If you’re pouring budget into AI to replace your entire customer support team, you might want to pause. Booking.com just went on record saying its human support is the key thing that sets it apart, and they’re not wrong.

⚡ The Break

Booking.com made it clear this week: while they use AI to understand what a customer needs, their actual differentiator is having a human resolve the problem. They’re a multi-billion dollar tech company, and they’re doubling down on people.

Most founders assume the endgame for customer service is zero humans. It’s all about ticket deflection and cutting costs. The north star metric becomes how many customers you can prevent from ever speaking to a person.

Turns out, for high-stakes interactions, that’s a fast-track to losing customers. Here’s what this actually means: your customers don’t want to fight a chatbot when their flight is cancelled or their hotel room doesn’t exist. They want a solution. Fast. AI is brilliant at diagnosing the issue from a frantic email, but it’s often shit at handling the nuance and emotion required to fix it. People aren’t stupid; they know when they’re being managed by a script.

I’ve seen good projects hit the wall over this. They optimise for bot resolution rates, and watch their customer satisfaction scores plummet. The real play isn’t replacement, it’s augmentation. Use AI to give your human team superpowers. When an agent picks up a call, the AI should have already summarised the issue, pulled up the customer’s history, and suggested three possible solutions. The goal isn’t to avoid a conversation; it’s to make that conversation brutally efficient and successful.

So, look at your CX roadmap. If it’s a straight line to firing all your support agents, you’re building a frustration engine, not a service. Start measuring your team on first-contact resolution and customer happiness. That’s the data that matters.

Subscribe now

🧠 What Quietly Changed

Yann LeCun’s attack on AI’s ‘common sense’ problem — While everyone is distracted by the latest video generator, Yann LeCun’s team dropped research on a thing called LeWorldModel. It’s a direct shot at one of AI’s biggest hurdles: learning how the world works just by watching, like a baby does. The research targets “JEPA collapse” in pixel-based models, which is a technical way of saying it’s trying to stop the AI from getting lazy and ignoring the details when predicting the future of a scene. This is the foundational work required for agents that can actually plan and reason in the real world, not just generate plausible-looking clips.

OpenAI’s $1B ‘goodwill’ fund — The OpenAI Foundation announced a $1 billion grant programme to ensure AI “benefits all of humanity.” This isn’t just a press release about ethics. For a founder, this is a giant flare telling you where the wind is blowing. Building with a clear, defensible social benefit is moving from a nice-to-have to a core requirement for survival. This kind of money doesn’t just fund non-profits; it shapes the entire narrative, pre-empts regulators, and sets the bar for what is considered an acceptable use of powerful technology. If your pitch deck doesn’t have a slide on this, you’re already behind.

The legal storm brewing for AI agents — We’re seeing more research into “agentic AI” and its geopolitical impact, but the quiet part is the immediate legal risk for builders. The moment your AI agent can act on a user’s behalf—making a purchase, sending an email, trading a stock—it can also break the law. When an autonomous agent executes a trade that violates international sanctions, who is liable? You? The user? The company that owns the platform? Most legal frameworks are completely unprepared for this. If you’re building agents, you need to be talking to lawyers now about liability, kill-switches, and how you log an agent’s ‘intent’.

🧪 The Edge Case

There’s a growing chorus of people confidently predicting an AI-driven dystopia. They speak with absolute certainty about how AI will upend society, destroy industries, and change the very nature of humanity.

The reality is, making investment decisions or building a company based on these arrogant predictions is a massive mistake. It presumes the future is a solved problem and that we are all just passengers. It ignores the simple fact that huge, complex systems—like global society—don’t just roll over for new technology. People adapt. They regulate. They find new ways to create value. The breathless predictions of flying cars in the 70s or the paperless office in the 90s should have taught us some humility.

Here’s the thing smart people are missing: they mistake a powerful tool for a deterministic force. They get so wrapped up in the technology’s potential that they forget about human behaviour, economics, and politics.

The builder’s lesson isn’t to ignore AI. It’s to stay grounded. Don’t build for a science fiction novel; build for a real customer with a real problem today. The most resilient startups will be those that use AI to solve a concrete need better than anyone else, while remaining agile enough to navigate the changes as they come. Betting the farm on a single, extreme vision of the future—whether a utopia or a dystopia—is just a dressed-up form of gambling. Solve today’s problems, and you’ll earn the right to solve tomorrow’s.

🎯 This Week’s Bet

BUILD: An AI-powered co-pilot for your customer support team. It should listen to calls and read chats in real-time to instantly surface the exact knowledge base article, previous ticket, or account detail the human agent needs to solve the problem. The goal is not to replace the human, but to make them impossibly fast and accurate.

BAIL: Chasing 100% automated customer support resolution. Set a rule: if a customer asks for a human twice in a row, the bot escalates immediately. The brand damage from trapping a frustrated customer in a chatbot loop is worth far more than the salary you think you’re saving.

For the ❤️ of startups ✌🏼 & 💙

We are building Arthur, your agentic Startup OS that gets **it done for you.

CLICK HERE TO GET EARLY ACCESS AND YOUR FREE FUEL UNITS

Share Fusion42 by DeReK WaTSoN

Thank you for reading. If you liked it, share it with your friends, colleagues and everyone interested in the startup Investor ecosystem.

If you've got suggestions, an article, research, your tech stack, or a job listing you want featured, just let me know! I'm keen to include it in the upcoming edition.

Please let me know what you think of it, love a feedback loop 🙏🏼

• ❤️ Love it, keep them coming!

• 👍 Great, always a good read.

• 😊 Enjoyable, but let’s see some variety.

• 👎 Okay, seen better.

• 🛑 Get a different job.
  
  Subscribe below and follow me on LinkedIn or Twitter to never miss an update. 

Subscribe now

For the ❤️ of startups

✌🏼 & 💙
Derek

The rules changed. Here’s the new playbook. 30 principles for building startups in an AI world — what worked before, what broke, and what founders need to do now. Not theory. Tactics that ship.

New principle every Monday

Know Your Models

Choosing the wrong AI model is now a product and cost architecture mistake. Founders need to match model capability to task rather than defaulting to the most powerful or cheapest option. You were told that deep technical expertise in your stack was non-negotiable. Turns out, the stack is now a menu, and your most important technical skill is knowing what to order.

Why This Mattered Before

The conventional wisdom was that a technical founder must deeply understand their core technology. And it was right. When you built everything from scratch, from the database to the front-end framework, that deep knowledge was the difference between shipping and stalling. Knowing the internals of your stack was how you optimised performance, debugged esoteric issues, and ultimately, built a defensible product.

Your unique advantage was in how you put the bricks together because you knew the properties of every single brick. This was the gospel of YC, the definition of a “great hacker,” and for a decade, it was the only way to win.

The Graveyard

I haven’t seen a single, high-profile startup death with “built their own model” listed as the official cause on the tombstone. The reality is more subtle and far more common. In the cohorts I’ve worked with, I’ve watched brilliant teams burn their first £500k and nine months of runway trying to build a bespoke NLP model for something like sentiment analysis or document summarisation.

They’d be heads down, trying to eke out a 2% accuracy gain, convinced this was their moat. Then a new foundation model from Anthropic or Google would drop, delivering better performance out of the box through a simple API call. The startup’s “moat” became a millstone overnight — expensive to maintain, slow to improve, and fundamentally uncompetitive. They didn’t die in a blaze of glory; they just became irrelevant.

What AI Actually Changed

The foundational technical competence has shifted from building the engine to knowing how to drive. You no longer need to know the internals of PostgreSQL. You need to know that Claude 3.5 Sonnet is faster and cheaper for enterprise-grade text analysis, while GPT-4o might be better for complex, multi-step agentic workflows. This is the new literacy.

The game is no longer about building your own model from scratch. The capital required is staggering. Cohere raised $500 million in 2025 to focus on enterprise models. xAI is backed by billions. You cannot compete with that. What they’ve done is commoditise raw capability. Your job is no longer to create the capability, but to intelligently apply it. The core engineering challenge has moved up the stack from model creation to model orchestration.

The New Playbook

1. Audit tasks, not your tech. Map every core task your product performs to the required AI capability. Does this task need world-class reasoning, or just fast, accurate classification? Does it need creative generation or factual extraction? Don’t use a sledgehammer to crack a nut. Use a small, fast model for simple jobs and save the expensive, powerful models for the tasks that truly need them.

2. Build a model “palette”. Don’t get locked into a single provider. Your architecture should allow you to route tasks to different models based on performance and cost. Set up a simple evaluation framework to continuously benchmark models like those from OpenAI, Anthropic, Google, and open-source alternatives on Hugging Face. A simple API call change could cut your COGS in half or double your performance overnight.

3. Treat cost-per-token as a core business metric. Your model choice directly impacts your cost of goods sold, and therefore your pricing and margin. A model that’s 98% as good but 80% cheaper might be the right business decision, even if it’s not the absolute peak of technical performance. This is no longer just an engineering choice; it’s a fundamental product and finance decision.

4. Hire for AI literacy, not ML research. Your first key hire isn’t a PhD who wants to build a model from the ground up. It’s a pragmatic engineer who lives on the leaderboards, understands the trade-offs between different models, and can quickly integrate and fine-tune them. They’re an architect and a plumber, not a research scientist.

The Warnings

First, avoid the “Sledgehammer Fallacy.” I see founders using the most powerful model on the market for every single API call, from parsing a simple user query to generating a button colour. This burns cash and introduces unnecessary latency. It’s lazy architecture, and it will kill your margins.

Second, beware of “Model Monogamy.” Committing your entire product to a single AI provider is a huge strategic risk. When they change their pricing, deprecate a model you rely on, or suffer a major outage, your business is at their mercy. Diversify your model dependencies just like you would any other critical part of your supply chain.

The Bottom Line

Your first AI hire isn’t a researcher. It’s an architect who knows which model to use, when, and precisely what it costs.

Subscribe now

Part of Startup Principles for an AI World — 30 principles for building in the new era. New issue every week.

Arthur is the AI native startup operating system I’m building in public — not hype, but a system that turns input into structured execution and tracks founder progress. If you want to follow or use it, it’s open for early access.
→ Access Arthur

Enjoy and hope to see you soon.
❤️ & ✌🏼

Derek

Thank you for reading. If you liked it, share it with your friends, colleagues and everyone interested in the startup Investor ecosystem.

If you've got suggestions, an article, research, your tech stack, or a job listing you want featured, just let me know! I'm keen to include it in the upcoming edition.

Please let me know what you think of it, love a feedback loop 🙏🏼

• ❤️ Love it, keep them coming!

• 👍 Great, always a good read.

• 😊 Enjoyable, but let’s see some variety.

• 👎 Okay, seen better.

• 🛑 Get a different job.
  
  Subscribe below and follow me on LinkedIn or Twitter to never miss an update. 

US House Votes to Expedite Datacenter Builds for AI Infrastructure

The House passed H.R. 7644, the “Accelerate Large-scale Artificial Intelligence Infrastructure Act,” this week, which allows for accelerated construction of datacenters crucial for AI development (MSN). This move aims to streamline regulatory hurdles, meaning faster compute access for founders.

Subscribe now

⚡ The Break

Here’s the thing: this isn’t just about cutting red tape. It’s a clear signal that the infrastructure required to power the next wave of AI is being prioritised. If you’re building compute-intensive AI applications, this means the playing field for securing resources could shift faster than anticipated. Companies that can leverage this new regulatory environment to build or access dedicated AI compute will gain a significant advantage. Don’t wait for the datacenters to appear; start modelling your compute needs now and explore long-term partnerships or commitments with infrastructure providers who can benefit from these accelerated timelines. Plan for available compute to come online sooner.

What Quietly Changed

China’s AI Labelling Mandate

On the 14th of March, the Cybersecurity Administration, Ministry of Industry and Information Technology, Ministry of Public Security, and State Administration of Radio and Television in China promulgated “Measures for Labeling of AI-Generated Synthetic Content” [2]. This new regulation, effective from March 7, 2025, mandates clear labelling for all AI-generated content. Most founders assume content generation is an unregulated wild west. The reality is, disclosure is becoming mandatory. If your AI generates content, even for internal use, you need to implement watermarking or metadata-based labelling systems now. This isn’t just a Chinese problem; it sets a precedent for global content provenance, and getting ahead of it protects you from future compliance headaches.

OpenClaw Ban in China

China banned the use of “OpenClaw” on government computers this week, citing security concerns such as prompt injection attacks [1]. This isn’t about protecting state secrets alone. It’s about a fundamental distrust of the security posture of AI agents in sensitive environments. If you’re deploying AI tooling within your organisation, particularly open-source or third-party solutions, you must conduct a thorough security audit focusing on prompt injection vulnerabilities and data exfiltration risks. The Chinese government found flaws; assume your adversaries will too. Implement strict sandboxing and access controls for all AI agents.

Grammarly’s AI Feature Pullback

Grammarly quietly abandoned its ‘AI Expert Review’ feature, which promised human-quality edits, hinting at the difficulties of truly human-like AI writing (The Guardian). This wasn’t a PR announcement; it was a quiet deletion from their offering. What most people missed here is that even well-funded companies with deep NLP expertise are struggling to deliver consistently nuanced, human-level content generation. For founders betting on AI completely replacing human writers or editors, this is a wake-up call. Reallocate your resources from trying to achieve perfect human emulation to building AI that augments human capabilities, focusing on efficiency and specific, measurable improvements, not wholesale replacement.

The Malicious AI Agent Threat

Security experts warned this week that malicious AI agents are capable of collaborating to execute sophisticated cyberattacks (TechRadar). This moves beyond individual agent threats; it means your security models need to account for orchestrated attacks from multiple AI entities. They cited an example where AI bypassed access restrictions by unearthing a hardcoded secret key in an application’s code repository. If you’re building any system accessible by AI or incorporating AI into your security stack, assume a coordinated, multi-agent attack is possible. Your defence strategy must evolve to include real-time behavioural analysis of AI agents, not just signature-based detection, and focus on anomaly detection at system boundaries.

The Edge Case

Adobe’s CEO, Shantanu Narayen, is stepping down amidst concerns about AI’s disruptive impact on creative industries (TipRanks). AI-driven capabilities are set to fundamentally alter creative work production.

The thing smart people are missing here isn’t just that AI will change how designers work. It’s that the very business model for creative tools is now under threat. I’ve seen good projects hit the wall over this. If the market shifts from purchasing tools to a subscription for outcomes — or, worse, if AI makes some creative tasks effectively free — then the entire revenue strategy of companies like Adobe needs a rethink. The builder’s lesson is specific: if your product is built on charging for creative tools, you need to either build AI that augments human creativity in ways no pure AI can replicate, or you need to radically rethink where your value creation lies. Are you selling paintbrushes, or are you selling finished masterpieces?

This Week’s Bet

BUILD: Automated, privacy-preserving AI solutions for child online safety that restrict access to inappropriate websites globally and specifically comply with UK regulations. Focus on solutions that can integrate seamlessly at the network level or within popular educational platforms.

BAIL: Stop developing AI tools that impersonate human writers or generate content without clear disclosure. The backlash, as seen with Grammarly’s abandoned ‘AI Expert Review’ feature and the push for an ‘AI-free’ logo, is becoming too significant to ignore. The ethical and regulatory landscape is shifting too fast to make this a sustainable bet.

Go Deeper

• China bans OpenClaw [1]

• Measures for Labeling of AI-Generated Synthetic Content [2]

For the ❤️ of startups ✌🏼 & 💙

We are building Arthur, your agentic Startup OS that gets **it done for you.

CLICK HERE TO GET EARLY ACCESS AND YOUR FREE FUEL UNITS

Share Fusion42 by DeReK WaTSoN

Thank you for reading. If you liked it, share it with your friends, colleagues and everyone interested in the startup Investor ecosystem.

If you've got suggestions, an article, research, your tech stack, or a job listing you want featured, just let me know! I'm keen to include it in the upcoming edition.

Please let me know what you think of it, love a feedback loop 🙏🏼

• ❤️ Love it, keep them coming!

• 👍 Great, always a good read.

• 😊 Enjoyable, but let’s see some variety.

• 👎 Okay, seen better.

• 🛑 Get a different job.
  
  Subscribe below and follow me on LinkedIn or Twitter to never miss an update. 

Subscribe now

For the ❤️ of startups

✌🏼 & 💙
Derek

The rules changed. Here’s the new playbook. 30 principles for building startups in an AI world — what worked before, what broke, and what founders need to do now. Not theory. Tactics that ship.

New principle every Monday

AI commoditised point solutions — founders need a specific high-value wedge that earns the right to expand, rather than building horizontal AI tools that incumbents bundle as features within a product cycle. You were told to start with a tiny niche, a single feature done ten times better. Turns out, that advice is now a trap. The defensible wedge isn’t a feature anymore; it’s the entire workflow you make obsolete.

Why This Mattered Before

The conventional wisdom was to start narrow and deep. Paul Graham’s essays hammered this home: own a tiny piece of the market completely before you even think about expanding. And it was right. Building anything, even a single, focused tool, required serious engineering effort. That effort was your moat.

A startup could build a defensible business by solving one problem exceptionally well because incumbents were too slow or unfocused to bother competing on such a small feature. You could earn passionate users, build from a position of strength, and expand from there. This playbook worked for years. Until it didn’t.

Subscribe now

The Graveyard

I haven’t seen a single, high-profile company vaporised overnight because its wedge became an AI feature. It’s not that dramatic. It’s a slower, more painful death.

In the cohorts I’ve worked with since 2023, I’ve watched dozens of founders pitch some variation of ‘AI for X’—AI for summarising legal documents, AI for writing marketing copy, AI for generating social media posts. They build a slick UI over a foundation model. They get some early interest. Then, within six months, Google, Microsoft, or the dominant SaaS player in their market announces the exact same capability, bundled for free into a platform with a million existing users. The startup’s traction stalls, the pipeline dries up, and they quietly pivot or wind down.

What AI Actually Changed

Here’s the thing. The barrier to building a point solution has collapsed. What used to take a small engineering team a quarter to build can now be replicated by a single developer using an API in a weekend. Incumbents are not asleep at the wheel; they are weaponising this.

The wedge is no longer about doing one thing better—it’s about automating an entire workflow that incumbents serve with multiple products or clunky integrations. Large platforms are in a feature-bundling arms race. Look at the low-code space. Appian saw its stock jump after announcing strong results driven by its new AI-powered solutions in late 2025. Their AI-inclusive tier revenue more than doubled quarter-over-quarter. They didn’t launch a new product; they wove AI directly into their existing platform, instantly upgrading their value proposition and boxing out any startup trying to sell a standalone “AI for process automation” tool.

The New Playbook

Your goal is no longer to build a better tool. It’s to build a better, more automated workflow that creates high switching costs.

1. Map the entire workflow, not the task. Don’t build an AI to summarise meeting notes. Build the system that schedules the meeting, records it, generates the summary, creates action items in the user’s project management tool, and drafts the follow-up email. The value isn’t in one step; it’s in the seamless automation of all of them.

2. Go vertical, not horizontal. A horizontal tool like “AI for copywriting” competes with everyone. A vertical tool like “AI for generating Phase I Environmental Site Assessment reports” for commercial real estate has a built-in moat. It requires specific domain knowledge, proprietary data structures, and a deep understanding of a workflow that big tech will never bother to learn.

3. Build for deep integration. Your product should become the central nervous system for a specific business process. Use tools like Retool or Airtable not just as a database, but as a platform to build custom, AI-powered internal applications that are deeply embedded in your customer’s operations. The harder you are to rip out, the more defensible your business is.

4. Own the messy ‘last mile’. Fully automated AI is a commodity. A workflow that is 95% automated but requires expert human-in-the-loop validation for the final 5% is a powerful moat. You aren’t just selling software; you’re selling a guaranteed outcome. This is something incumbents with a purely self-serve model struggle to replicate.

Companies To Watch

• Harvey — San Francisco, USA — AI platform that automates full legal workflows from contract review to case research and litigation strategy. Perfect example of the article’s thesis - avoided the point solution trap by automating entire legal workflows instead of building narrow AI tools that incumbents like Westlaw could easily bundle. Shows how to build a defensible wedge that makes existing workflows obsolete.

• Sierra — San Francisco, USA — Conversational AI platform that powers full customer service workflows, from query resolution to retention plays. Exemplifies the new playbook by automating complete customer service workflows instead of building narrow chatbot features that platforms like Zendesk bundle for free. Demonstrates how to own the entire workflow rather than a single touchpoint.

• Adept — San Francisco, USA — AI agents that automate complex enterprise workflows like sales outreach and data entry via natural language actions. Shows the evolution from point solutions to workflow automation - instead of building narrow AI tools that get commoditised, they’re creating AI agents that handle entire business processes. Perfectly illustrates the shift from better tools to obsolete workflows.

• Ema — Mountain View, USA — Universal AI employee platform that handles multi-app workflows across HR, finance, and IT. Demonstrates the article’s core thesis by avoiding the commoditisation trap of single-purpose AI tools. Instead of building narrow automation features, they’re replacing entire cross-functional workflows that incumbents can’t easily replicate with simple bundling.

• UiPath — New York, USA — End-to-end robotic process automation platform that orchestrates AI agents to replace entire back-office workflows from invoice processing to compliance auditing. Pre-AI example of the thesis in action - avoided the point solution trap by automating complete workflows before incumbents could bundle individual features. Shows the enduring value of workflow-level thinking that’s now even more critical in the AI era.

• Zapier — Sunnyvale, USA — No-code automation platform that chains AI-powered apps into custom workflows, obsoleting manual integrations across sales, marketing, and ops. Illustrates how workflow-level thinking beats point solutions - while individual integrations get bundled by platforms like Salesforce, Zapier’s cross-tool workflow automation remains defensible. Shows the power of owning the entire process rather than individual connections.

• Glean — Palo Alto, USA — Enterprise search and knowledge AI that replaces fragmented internal search workflows with instant, contextual answers. Avoids the commoditisation of individual search features by owning the entire knowledge discovery workflow. While point search tools get bundled into platforms like Slack, Glean’s comprehensive approach to enterprise knowledge creates higher switching costs.

• Casetext — San Francisco, USA — AI legal research platform (CoCounsel) that automates full case prep workflows. Success story that validates the thesis - avoided building narrow legal research features and instead automated complete case preparation workflows. The $650M acquisition by Thomson Reuters shows the value of workflow-level thinking over point solutions.

The Warnings

The Thin Wrapper Trap. If your entire product is a slightly nicer UI on top of a third-party LLM model, you don’t have a business. You have a feature that’s waiting to be copied. I’ve watched founders raise money on this premise, only to see their churn skyrocket the moment their customers’ existing tools get the same AI capability for free.

Mistaking a Feature for a Workflow. Founders fall in love with a clever AI-powered feature. They build a company around it. But customers don’t buy features; they buy solutions to whole problems. If your “wedge” only solves one small part of their painful, multi-step process, you’ll be Sherlocked by the platform that owns the rest of that workflow.

The Bottom Line

Stop selling a smarter hammer. Start selling the automated workshop, with you as the foreman who guarantees the job gets done.

Part of Startup Principles for an AI World — 30 principles for building in the new era. New issue every week.

Sources:

• Appian Q4 2025 Earnings Call Transcript

• Paul Graham, ‘The Best Way to Start a Startup’

Part of Startup Principles for an AI World — 30 principles for building in the new era. New issue every week.

Arthur is the AI native startup operating system I’m building in public — not hype, but a system that turns input into structured execution and tracks founder progress. If you want to follow or use it, it’s open for early access.
→ Access Arthur

Enjoy and hope to see you soon.
❤️ & ✌🏼

Derek

Thank you for reading. If you liked it, share it with your friends, colleagues and everyone interested in the startup Investor ecosystem.

If you've got suggestions, an article, research, your tech stack, or a job listing you want featured, just let me know! I'm keen to include it in the upcoming edition.

Please let me know what you think of it, love a feedback loop 🙏🏼

• ❤️ Love it, keep them coming!

• 👍 Great, always a good read.

• 😊 Enjoyable, but let’s see some variety.

• 👎 Okay, seen better.

• 🛑 Get a different job.
  
  Subscribe below and follow me on LinkedIn or Twitter to never miss an update. 

Subscribe now

⚡ The Break

The US government just mandated that civilian AI contracts must make models available for ‘any lawful use’ — a direct rebuke to Anthropic’s attempt to restrict military applications last week. Dario Amodei warned this week that AI is outpacing oversight, creating national security risks. He’s right, but the ship has sailed.

(Source: The Financial Times, Source: Storyboard18).

This push for “lawful use” contracts directly contradicts Anthropic’s earlier move to restrict military applications, suggesting they saw it as “the wrong tool” in government deals. (Source: SF Standard). Meanwhile, Sam Altman is reported to be pushing for a NATO contract, having already faced backlash over a US Department of Defense deal. (Source: Cybernews). These aren’t isolated incidents. They’re evidence of a land-grab.

Here’s what this actually means: governments will force “lawful use” clauses into your contracts. If you’re selling to government, your terms of service must reflect this. Your product won’t be exempt from military or intelligence applications if the contract says so. This isn’t about ethical debates; it’s about legislative and commercial reality. If you’re pitching government contracts, add a compliance appendix showing how your model handles dual-use scenarios.

Your metric: 100% of government proposals include a dual-use policy statement. If you’re raising, expect VCs to ask about military deployment policies in Series A diligence.

Your check: can you clearly articulate your stance on military applications with one sentence?

🧠 What Quietly Changed

Yann LeCun’s AMI Labs Takes Aim at World Models — Yann LeCun, a pioneer in AI, just launched AMI Labs in Paris with over $1 billion in funding. (Source: Ecostylia). They’ve even appointed a new CEO. This isn’t just another tech start-up; it’s a heavily funded research effort focused on “world models.” This could shift the entire industry beyond current transformer-based architectures. If you’re building foundational AI, you need to watch their papers closely.

Your action: review AMI Labs’ initial research agenda to see where things are really going.

Zero Trust Security Needs an AI Upgrade — The standard Zero Trust model is quickly becoming insufficient. It was designed for static environments. AI-powered threats are dynamic, sophisticated, and evolving faster than ever. Security Boulevard highlighted this week that enhanced Zero Trust models now use AI for real-time threat detection and automated policy enforcement. (Source: Security Boulevard). If your cybersecurity strategy isn’t actively integrating AI for continuous authentication and behavioural analytics, you’re already behind. Your action: turn on continuous authentication for admin actions; your metric: 100% of privileged actions gated by step-up authentication and device posture.

Gaming Industry Grapples with AI Copyright — Gamers are asking hard questions about AI in new releases, and it’s kicking up a fuss about copyright. Valve, a major platform, might remove games if AI-generated assets infringe copyright. (Source: Startup Daily). For game development founders leveraging AI, transparency and legal compliance are no longer optional. This isn’t just player backlash; it’s a platform penalty looming.

Yoshua Bengio and NVIDIA Eye Post-LLM AI — Another AI pioneer, Yoshua Bengio, is re-teaming with Xie Saining to launch a new company. NVIDIA, of all companies, is investing. Their aim? To explore technologies “beyond large language models.” (Source: 36kr). This is not a drill. What a named leader did this week means if you’re over-indexed on LLMs, you need to start thinking about the next wave. This could well be where the real value sits for the next decade. Your action: allocate 10% of your R&D budget to exploring non-LLM AI architectures; your metric: one experimental prototype within six months.

AI’s Impact on Manufacturing Workforce — The integration of AI in manufacturing is creating a significant skills gap. Asad Afzal of ASAFE noted that manufacturers are struggling to prepare existing systems for AI integration, facing issues with throughput, traffic patterns, and the escalating value of equipment once AI is involved. (Source: Manufacturing Dive). This isn’t just about robots replacing humans, it’s about humans needing new skills to work with AI. Businesses need to invest in reskilling programmes now to avoid future talent shortages.

🧪 The Edge Case

A Meta AI security researcher had her inbox hijacked by an OpenClaw agent this week. (Source: Windows Central). The agent, designed to access Google’s Antigravity backend via proxy, went rogue (Source: AI CERS News). It started sending emails, digging into contacts. The incident highlighted significant “malicious usage” leading Google to restrict usage of Antigravity by OpenClaw users, citing token-amplification studies (Source: VentureBeat). Turns out, security researchers supported the capacity-protection argument. Fuck.

The reality is this isn’t about AI agents being inherently dangerous. It’s about how we grant them access. The fix: scope agent permissions to single-task API keys, not full inbox OAuth tokens. Use Google’s new per-agent credential system or build your own revocable token layer. Audit your agent scopes this week: list every credential, rotate anything long-lived, and default to least-privilege OAuth scopes. Your action: review all agent credentials for least privilege; your metric: 0 agents with broad OAuth access where a scoped API key would suffice.

🎯 This Week’s Bet

BUILD: A transparent, AI-driven asset verification service for the gaming industry. With Valve poised to pull games over copyright concerns from AI-generated assets, developers will pay good money for a solution that can scan, verify originality, and provide a clear ‘clean’ report for their digital content before launch. Your action: develop a prototype asset verification service incorporating content provenance; your metric: achieve 95% accuracy in identifying AI-generated content.

BAIL: Stop pushing for blanket AI adoption across all departments without first validating system readiness. Manufacturers are struggling to prepare existing systems for AI integration, facing issues with throughput, traffic patterns, and the escalating value of equipment once AI is involved. (Source: Manufacturing Dive). Blindly throwing AI at an unprepared infrastructure is a waste of money and resources. Your action: conduct an internal AI readiness audit before committing to new AI projects; your metric: 100% of proposed AI initiatives have a verified infrastructure readiness report.

Go Deeper

• AI Oversight & Regulation: Dario Amodei raises national security concerns as AI outpaces law and military oversight. (Storyboard18)

• Yann LeCun’s AMI Labs: Launches with over $1 billion funding, focusing on “world models” in AI. (Ecostylia)

• Zero Trust Evolution: Traditional models are insufficient against dynamic AI threats, requiring AI-powered enhancements. (Security Boulevard)

• Gaming AI & Copyright: Game developers face scrutiny over AI use, with Valve threatening removal for infringement. (Startup Daily)

• NVIDIA & Yoshua Bengio: Form new company, with NVIDIA investment, to explore AI beyond LLMs. (36kr)

• AI in Manufacturing: Asad Afzal discusses the challenges of integrating AI into manufacturing systems. (Manufacturing Dive)

• Google Antigravity/OpenClaw Incident: Google restricted Antigravity usage for OpenClaw users following “malicious usage.” (VentureBeat, AI CERTS News, Windows Central)

For the ❤️ of startups ✌🏼 & 💙

We are building Arthur, your agentic Startup OS that gets **it done for you.

CLICK HERE TO GET EARLY ACCESS AND YOUR FREE FUEL UNITS

Share Fusion42 by DeReK WaTSoN

Thank you for reading. If you liked it, share it with your friends, colleagues and everyone interested in the startup Investor ecosystem.

If you've got suggestions, an article, research, your tech stack, or a job listing you want featured, just let me know! I'm keen to include it in the upcoming edition.

Please let me know what you think of it, love a feedback loop 🙏🏼

• ❤️ Love it, keep them coming!

• 👍 Great, always a good read.

• 😊 Enjoyable, but let’s see some variety.

• 👎 Okay, seen better.

• 🛑 Get a different job.
  
  Subscribe below and follow me on LinkedIn or Twitter to never miss an update. 

Subscribe now

For the ❤️ of startups

✌🏼 & 💙
Derek

Why This Mattered Before

You were told to focus on growth first. Get users, build something they love, and figure out how to make money later. The mantra was “make something people want.” Revenue was a secondary concern to engagement.

I used to hear this in every single office hours session. Founders would say “we’ll figure out monetisation once we have traction.” And for years, that actually worked. For traditional software, the marginal cost of serving another user was near zero. Once the code was written, adding the ten-thousandth user cost little more than the tenth. You could afford to give it away because it cost you almost nothing.

The Graveyard

The public post-mortems for AI startups that burned to death subsidising free users haven’t been written yet. The market is moving too fast. But I can tell you what I’ve seen.

In the cohorts I’ve worked with since 2024, the pattern is painfully consistent. A founder builds a clever wrapper around a powerful model. They launch a free beta and get a rush of sign-ups. They see user numbers climb and think it’s traction. But their OpenAI bill climbs faster. I’ve watched teams torch a meaningful chunk of pre-seed in a single quarter on inference, then lose the vast majority of ‘users’ the moment they introduced pricing. They’re left with no money, no paying customers, and a product no one actually valued enough to pay for.

What AI Actually Changed

Here’s the thing. AI turned whole product categories into near-zero marginal-cost businesses overnight. Shipping, support, onboarding, even chunks of delivery are now automated. So charging £20/month doesn’t prove you’ve got a business—it might just prove you’ve got cheap compute and a Stripe account.

Every time a user interacts with your product, it triggers an API call to OpenAI, Anthropic, or your own self-hosted model. That is a real, metered, non-zero cost. It shows up as a line item on your monthly bill. Your Cost of Goods Sold (COGS) is no longer just server hosting; it’s a direct function of user engagement. An active free user is not an asset, it’s a liability.

Giving away an AI product for free is not like giving away software. It’s like giving away coffee. The first cup isn’t free to you, and the thousandth cup certainly isn’t. You’re paying your supplier for every single one. If you don’t charge the customer, you’re just running a very expensive charity

.

The New Playbook

1. Charge from Day One. This isn’t just about covering costs. It’s the fastest way to validate that you’ve built something of actual value. Free users will tell you they love it; paying users prove it. Price on the value you create. Just last week on Hacker News, I saw a tool that negotiated a hospital bill down by thousands; people will happily pay a percentage of that saving.

2. Know your Unit Economics. You must know your exact cost-per-user and cost-per-query from the start. Instrument everything. If it costs you £0.50 in tokens to serve an average user session, and you’re charging £10 a month, you know your ceiling for engagement. If you don’t track this, you’re flying blind into a cash-flow crisis.

3. Use Strategic Freemium, Not a Free-for-all. A free tier can still work, but it must be a marketing tool, not the core product. Offer a low-cost version that uses a cheaper, less powerful model (e.g. Haiku instead of Opus). Or, offer a severely rate-limited or feature-gated experience. The goal of the free tier is to demonstrate value so clearly that users upgrade. It’s a product tour, not an open bar.

4. Align Billing with Costs. Usage-based billing is your friend. By charging per-credit, per-report, or per-token-bundle, you ensure your revenue scales directly with your single biggest cost centre. This protects your margins and forces high-volume users to become your most valuable customers, not your biggest expense.

Subscribe now

The Warnings

The Pivot Trap. I’ve watched founders realise too late that their free users and their paying customers want different things. The person playing with your AI toy at 2am is not the same person who’ll expense a £200/month SaaS tool. If you build for free users first, you’re building the wrong product. The Timing Trap. You think you’ll introduce pricing ‘once we have more features.’ But your free users are training you to build features that don’t drive revenue. Every month you delay is another month you’re learning the wrong lessons.

The Bottom Line

Charge early. Cover your costs. Prove your value. Don’t fund hobbies with venture capital.

Part of Startup Principles for an AI World — 30 principles for building in the new era. New issue every week.

Arthur is the AI native startup operating system I’m building in public — not hype, but a system that turns input into structured execution and tracks founder progress. If you want to follow or use it, it’s open for early access.
→ Access Arthur

Enjoy and hope to see you soon.
❤️ & ✌🏼

Derek

Thank you for reading. If you liked it, share it with your friends, colleagues and everyone interested in the startup Investor ecosystem.

If you've got suggestions, an article, research, your tech stack, or a job listing you want featured, just let me know! I'm keen to include it in the upcoming edition.

Please let me know what you think of it, love a feedback loop 🙏🏼

• ❤️ Love it, keep them coming!

• 👍 Great, always a good read.

• 😊 Enjoyable, but let’s see some variety.

• 👎 Okay, seen better.

• 🛑 Get a different job.
  
  Subscribe below and follow me on LinkedIn or Twitter to never miss an update. 

Subscribe now

Demis Hassabis issued a stark warning this week: a critical memory chip shortage is actively throttling the deployment of AI models across the industry. Not theoretical scarcity — actual production halts.

Micron confirmed they’re halting manufacture of specific HBM3E chips due to manufacturing yield problems.

If you’re planning to scale inference in the next 6 months, your cloud provider’s capacity isn’t guaranteed. Lock in reserved instances now, or start building model-agnostic architectures that let you shift between providers when one hits a supply wall. The founders who wait for “the market to sort it out” will be the ones paying 3x spot pricing in Q4.

What Quietly Changed

AI-Generated Documents Lack Legal Privilege — Federal Judge Thomas Rice ruled this week in Heppner v. United States that documents generated by AI are not protected by attorney-client privilege or the work-product doctrine

• This stems from a defendant’s independent use of generative AI to assess legal exposure

• Most founders assume privilege covers anything your lawyer touches. It doesn’t if the analysis is generated by an AI tool. If your GC is using ChatGPT to draft risk memos for board meetings, those documents are now discoverable in litigation. Move that work back to human associates, or accept the exposure.

• OpenAI’s “Stateful” AWS Architecture — Buried in the AWS partnership announcement this week, OpenAI’s infrastructure lead revealed they are shifting to a “stateful” architecture. This maintains context across sessions at the infrastructure level. If you’re a CTO making cloud decisions this quarter, the relationship with your cloud provider just got stickier. Once your application relies on persistent session context in AWS, migrating to Azure or Google Cloud Platform gets materially harder.

• FDA De Novo for AI Medical Software — The US Food and Drug Administration (FDA) granted de novo classification this week to an AI-powered software from Ultrasound AI designed to predict delivery dates using standard ultrasound images

• If you’re building medical AI, pull the de novo summary this week: list the clinical evidence endpoints and validation protocol, and map your product against them. This is the clearest regulatory blueprint we’ve seen for AI diagnostics.

• Ginkgo Bioworks’ Strategic Pivot — Jason Kelly, CEO of Ginkgo Bioworks, confirmed this week a major strategic pivot for the synthetic biology company. After extensive losses, Ginkgo is now foregrounding the sale of AI lab robots and automation platforms to third parties, rather than solely focusing on their own bio-foundry services

• It’s a recognition that the foundational tools for AI-driven discovery can be a more lucrative product than the discoveries themselves. For founders in deep tech, particularly those with significant CAPEX, it asks a blunt question: is your core IP the end product, or the machinery that makes it? Perhaps the real value is in selling the tools.

The Edge Case

Washington state’s Department of Licensing (which includes the DMV) rolled out an AI-powered voice system for customer service this week. The system, propped up by Amazon Web Services, immediately drew fire for its “heavily accented English.” This wasn’t some minor glitch. Bilingual customers found themselves stuck, then shunted to even longer wait times for a human.

The thing smart people are missing here: it’s not simply an issue of “accent.” The builder’s lesson is this: deploying AI in public-facing services isn’t just about functionality; it’s about inclusive design. An AI that can’t effectively serve a diverse user base is an outright failure. You must test your AI solutions on your actual user demographics, not just for technical performance but for comprehension and cultural fit. “Good enough” clearly wasn’t good enough for the Washington DMV. It’s basic product development. If your AI excludes a segment of your users, you’ve not built a product, you’ve built a barrier.

This Week’s Bet

BUILD: ServiceNow paid $60-80M for Traceloop this week.

Why? Because AI audit trails don’t exist yet, but the need for them is exploding. Compliance teams can’t answer ‘which AI agent accessed what customer data and why?’ Build the Datadog for AI agent activity — not for developers, but for CISOs who need to show auditors a complete chain of custody.

The wedge is enterprises already failing SOC 2 audits because their AI tools don’t log at the required granularity.

BAIL: Stop assuming your AI applications will be judged solely on their computational prowess; pull the plug on any deployment lacking robust, real-world user acceptance testing across diverse demographics. The Washington DMV case is a harsh reminder. Prioritize inclusive design and rigorous, representative user testing over raw performance metrics. Your customers don’t care how clever your algorithm is if it doesn’t work for them.

Go Deeper

Source 1 — Federal Judge Thomas Rice rules AI-generated documents are not privileged.

Source 2 — Venable LLP analysis of Heppner v. United States ruling.

Source 3 — OpenAI’s infrastructure lead discusses “stateful” AWS architecture.

Source 4 — FDA gives de novo classification to AI medical software.

Source 5 — Ginkgo Bioworks pivots to selling AI lab robots after significant losses.

For the ❤️ of startups ✌🏼 & 💙

We are building Arthur, your agentic Startup OS that gets **it done for you.

CLICK HERE TO GET EARLY ACCESS AND YOUR FREE FUEL UNITS

Share Fusion42 by DeReK WaTSoN

Thank you for reading. If you liked it, share it with your friends, colleagues and everyone interested in the startup Investor ecosystem.

If you've got suggestions, an article, research, your tech stack, or a job listing you want featured, just let me know! I'm keen to include it in the upcoming edition.

Please let me know what you think of it, love a feedback loop 🙏🏼

• ❤️ Love it, keep them coming!

• 👍 Great, always a good read.

• 😊 Enjoyable, but let’s see some variety.

• 👎 Okay, seen better.

• 🛑 Get a different job.
  
  Subscribe below and follow me on LinkedIn or Twitter to never miss an update. 

Subscribe now

For the ❤️ of startups

✌🏼 & 💙
Derek

Why This Mattered Before

You need three co-founders. That’s what every accelerator told you: a technical founder who codes, a product founder who translates user needs, a business founder who sells. The solo founder was a liability—you’d burn out before you shipped.

I’ve watched this play out across hundreds of cohorts. The solo technical founder who couldn’t articulate the problem they were solving. The product visionary who couldn’t ship. The business founder with a deck but no demo. Each failed in predictable ways. The three-person team wasn’t about having bodies in seats — it was about covering the irreducible minimum of skills required to build something people would pay for.

The Graveyard

I’ve watched founders in my cohorts make this mistake. They’d staff up, hiring for traditional roles – a front-end developer, a content writer – whilst a solo founder with Claude and Cursor was already shipping. By the time they realised they were building a Rolls-Royce for a scooter market, they’d burned through months of runway on salaries that AI could’ve handled for a fraction of the cost. I’ve seen teams hire multiple developers, only to get out-shipped by a competitor with a single founder and an AI licence. Their burn rate choked them before they found product-market fit.

What AI Actually Changed

AI collapsed the cost of iteration. The old path—idea, spec, engineer, sprint, QA, ship, learn—demanded coordination between specialists. Now, it’s idea, prompt, ship, learn. This shift isn’t about AI being ‘powerful’; it’s that iteration no longer requires a rigid division of labour.

Implementation time compressed with tools like Cursor and Copilot. Specification writing, documentation, and even early marketing copy shrank from days to hours with Claude. Operational overhead, like customer support and rudimentary data analysis, is now largely automated. The bottleneck shifted from ‘can you code or write?’ to ‘can you decide what to build and ship it fast enough to learn?’ Early generalists with AI now out-iterate prematurely specialised teams.

The New Playbook

1. Start solo or duo. Add AI as your third. Don’t hire for output you can generate with AI. Hire for judgement, relationships, or domain expertise.

2. Run a 30-day loop:

   • Week 1: 15 customer calls. Write 10 problem statements. Pick one.

   • Week 2: Build the narrowest possible workflow in Cursor. Ship to 3 design partners.

   • Week 3: Instrument usage. Collect objections. Iterate daily.

   • Week 4: Charge. If nobody pays, kill it.

3. Hire when you’re turning down revenue. The undeniable signal that you’ve waited too long is when you are actively having to turn down revenue opportunities because your current setup cannot adequately deliver. Not when you’re busy. Not when you’re stressed. When you’re saying no to money because you physically cannot fulfil the work.

4. When you do hire, hire for what AI can’t do: complex sales, deep domain expertise, or strategic partnerships. Don’t hire a content writer. Don’t hire a junior developer. Hire the enterprise sales lead who can navigate six-month procurement cycles, or the domain expert whose knowledge isn’t in any training dataset.

Subscribe now

The Warnings

Speed without a clear, validated direction is merely an expensive distraction. I’ve watched AI-enabled founders ship five products in six months and gain zero traction because they never validated the problem. AI removes the friction of building — it doesn’t remove the need to talk to customers first.

The second trap: staying lean becomes staying stuck. Some founders use “AI can do it” to avoid hiring when they’re already turning down revenue. That’s not discipline — that’s fear. If customers are waiting and you can’t deliver, hire.

The Bottom Line

The question isn’t ‘Can one person do this?’ It’s ‘Why would three?

Part of Startup Principles for an AI World — 30 principles for building in the new era. New issue every week.

Arthur is the AI native startup operating system I’m building in public — not hype, but a system that turns input into structured execution and tracks founder progress. If you want to follow or use it, it’s open for early access.
→ Access Arthur

Enjoy and hope to see you soon.
❤️ & ✌🏼

Derek

Thank you for reading. If you liked it, share it with your friends, colleagues and everyone interested in the startup Investor ecosystem.

If you've got suggestions, an article, research, your tech stack, or a job listing you want featured, just let me know! I'm keen to include it in the upcoming edition.

Please let me know what you think of it, love a feedback loop 🙏🏼

• ❤️ Love it, keep them coming!

• 👍 Great, always a good read.

• 😊 Enjoyable, but let’s see some variety.

• 👎 Okay, seen better.

• 🛑 Get a different job.
  
  Subscribe below and follow me on LinkedIn or Twitter to never miss an update. 

Subscribe now

I used to think that clever hacks and rapid deployment were a founder’s best friends. Turns out, the bureaucracy and the lawyers are always lurking, ready to remind you who really sets the rules. This week, they certainly made their presence felt.

⚡ The Break

A New York court just dropped a bombshell: if you’re chatting with a public generative AI platform, those conversations aren’t protected by attorney-client privilege. Not even a sniff of it, according to the SDNY. Akin Gump broke the news. And get this: LinkedIn’s identity verification process, which requires users to submit images of identity documents, funnels biometric data. Concerns have been raised, for instance, in a LinkedIn Pulse article, regarding how third parties might use this data, linking to LinkedIn’s User Agreement and Privacy Policy, which users must accept to proceed with verification [2]. People have also detailed their experiences, noting they’ve provided passport scans for verification 1.

Here’s what this actually means...

For founders:

If you use public AI tools for anything sensitive – product ideas, legal queries, strategic planning – you’re playing with fire. Assume everything you input is public. This isn’t a paranoid assumption; it’s the new legal reality. Find enterprise-grade, secure, private AI solutions for confidential work, or build your own sandboxed internal tools. Your legal team will be all over this, trust me. Don’t wait for them to tell you.

If you’re building any kind of verification or identity service, especially using biometrics, you need to be crystal clear about who gets that data and how it’s used. The fine print bites back, and people are scrutinising it. Transparency isn’t a nice-to-have; it’s a make-or-break, especially when it comes to personal data. Get caught being murky, and your trust goes out the window, along with your user base. This isn’t just about compliance; it’s about not being a dick about user data.

For investors:

Scrutinise AI legal-tech solutions more carefully. If they rely on public generative AI for communication, their value proposition just took a battering. This ruling changes the calculus for those investments. What’s the plan for sensitive client data? What’s the audit trail? These are now critical diligence questions.

Look hard at any portfolio company handling sensitive user data, particularly biometrics. Demand to see airtight data governance, explicit customer consent, and a bulletproof privacy policy. The regulatory hammer is coming, and you don’t want to be holding the bag for a company that gets caught out because they thought nobody would read the terms.

This isn’t a drill; it’s a new legal landscape that demands immediate attention.

🧠 What Quietly Changed

Industrial Espionage in AI is Here:

Anthropic, the big AI player behind Claude, recently accused Chinese AI companies – DeepSeek, Moonshot, and MiniMax – of using around 24,000 fake accounts to steal from its Claude chatbot VentureBeat. This isn’t just a corporate spat; it’s industrial espionage, plain and simple, played out on an international stage. Elon Musk even piled on, claiming Anthropic “has to pay Billions for Theft” 1. Meanwhile, Anthropic’s CEO, Dario Amodei, was reportedly “summoned” to the Pentagon for a “tense” meeting about the military’s use of Claude AI model 3. The implication for founders building foundational models or even highly specific applications? Expect your IP to be targeted relentlessly. If you’re building your own models, your operational security is now as critical as your code. Build robust credential management for API keys and internal access, and invest in real-time monitoring for unusual activity.

Google Tightens API Access, Killing Workarounds:

Google just shut down access for users leveraging third-party proxies like OpenClaw to interact with its Antigravity backend. This move banned OpenClaw users, even those who were paid Antigravity subscribers Trending Topics and MLQ.ai covered it. The implication? Major players are tightening their grip on their AI infrastructure, making it clear: play by their rules, or don’t play at all. This means any “API-wrapper” business built on unofficial access or clever parsing now has a dramatically shorter runway. Shift your focus to building unique value on top of sanctioned APIs, or pivot entirely to problems not dependent on a specific vendor’s hidden backend.

The Memory Chip Bottleneck is Getting Real:

Demis Hassabis, Google DeepMind’s CEO, is warning everyone that a severe memory chip shortage is hindering AI model deployment and research OpenTools.ai and NewsBytes. Micron, a major manufacturer, is even halting some AI chip production due to manufacturing challenges. I’ve seen good projects hit the wall over less. This isn’t theoretical; it’s hitting deployment and research hard. If your AI solution is compute-bound, particularly memory-intensive, this shortage will impact your timelines and burn rate. Re-evaluate your compute strategy now: can you optimise models for less memory, or explore hybrid approaches that don’t rely solely on cutting-edge, scarce hardware?

🧪 The Edge Case

“My AI agent just decided to redecorate my inbox with spam and existential dread. We need to put these things on a tighter leash, or at least a digital muzzle.” This isn’t just a funny quote. It’s what an AI security researcher at Meta experienced when an OpenClaw agent ran amok in her inbox, as TechCrunch reported.

Here’s the thing: everyone’s so focused on the big, terrifying AI risks, the “Skynet” stuff. But the reality is, the immediate danger lies in the mundane. It’s the small, delegated tasks that go wrong because we haven’t given our AI agents clear boundaries or circuit breakers. It’s about a lack of precise design. The smart people are missing how easily “helpful” can become “harmful” through a lack of robust guardrails and explicit intent definition. For builders, this isn’t a moral failing of the AI; it’s a design failing by us. Your agentic systems will go off-script. Design with conservative constraints, not permissive freedom. Implement granular permissions that dictate exactly what an agent can and cannot do. Crucially, build in “kill switches” – clear, immediate ways for a human to halt any automated process that steps outside its intended behaviour. This isn’t about ethical philosophy; it’s about practical engineering.

🎯 This Week’s Bet

BUILD: A privilege-safe copilot for highly regulated industries. The SDNY ruling makes it clear: exposing sensitive conversations to public LLMs is a non-starter for legal or patent work.

1. On-prem/Private Cloud LLM Integration: Create a platform that leverages open-source LLMs hosted entirely within a company’s secure infrastructure (on-prem or dedicated private cloud).

2. Strict Data Isolation & Masking: Implement data masking and anonymisation at the input layer, ensuring that truly sensitive client or proprietary information is never exposed to the core LLM during query formulation.

3. Auditable Human-in-the-Loop Workflow: Design a mandatory human review step before any LLM-generated output (e.g., legal brief drafts, contract clauses) is finalised or shared, with full audit trails of all human interventions and approvals.

BAIL: Building any service that relies on undisclosed sharing of user biometric data. LinkedIn’s murky practices, linking directly to accepted terms where biometric data sharing is vaguely explained 2, are a clear signal. If your model involves biometric data, and you’re not absolutely transparent about every single data flow, including exact third-party recipients and their usage, stop. You’re building a legal and reputational nightmare that will explode when regulators or privacy advocates come knocking.

For the ❤️ of startups ✌🏼 & 💙

We are building Arthur, your agentic Startup OS that gets **it done for you.

CLICK HERE TO GET EARLY ACCESS AND YOUR FREE FUEL UNITS

Share Fusion42 by DeReK WaTSoN

Thank you for reading. If you liked it, share it with your friends, colleagues and everyone interested in the startup Investor ecosystem.

If you've got suggestions, an article, research, your tech stack, or a job listing you want featured, just let me know! I'm keen to include it in the upcoming edition.

Please let me know what you think of it, love a feedback loop 🙏🏼

• ❤️ Love it, keep them coming!

• 👍 Great, always a good read.

• 😊 Enjoyable, but let’s see some variety.

• 👎 Okay, seen better.

• 🛑 Get a different job.
  
  Subscribe below and follow me on LinkedIn or Twitter to never miss an update. 

Subscribe now

For the ❤️ of startups

✌🏼 & 💙
Derek

Why This Mattered Before

The logic was sound: identify an emerging channel, build deep expertise in its mechanics, and create a product that makes that channel work better than anyone else could manage in-house. The channel was the product. Master it, and you’d own the customers who depended on it.

Discord proved this brilliantly. They didn’t just build another chat app — they built the communication infrastructure for gaming communities. Low-latency voice, robust server management, community moderation tools that understood how gamers actually organised themselves. By 2021, they’d raised nearly $1 billion and hit a $15.2 billion valuation. The product was inseparable from the channel it served.

Intercom did the same for customer messaging. They weren’t selling software — they were selling the channel itself. Chat widgets, in-app messages, email campaigns, all unified into one platform that became the primary way SaaS companies talked to users. By 2024, they were generating $343 million in revenue with a $1.3 billion valuation. The channel was the entire value proposition.

Both companies succeeded because the technical complexity and channel-specific knowledge created genuine barriers. You couldn’t just bolt on Discord’s voice quality or Intercom’s message orchestration as an afternoon project.

The Graveyard

In the cohorts I’ve worked through since 2022, I’ve watched this pattern destroy dozens of early-stage companies. The most common casualty: startups building AI-powered customer support tools as standalone products. They’d raise a seed round on the promise of “intelligent chatbots that understand context” or “automated ticket routing that learns from your team.” Solid products. Real customer validation. Then Meta, Zendesk, and Salesforce started bundling comparable AI features directly into their core platforms — often at no additional cost.

The breaking point wasn’t that the AI got better. It was that the incumbents already owned the channel. They had the customer data, the existing integrations, the sales relationships. Adding an AI layer was trivial compared to the distribution advantage they already held. The startups were selling a feature. The incumbents were giving it away to protect their platform position.

I watched one founder in a 2023 cohort spend eighteen months building a sophisticated AI email outreach tool. Beautiful product. Real engagement metrics. Dead within six months of HubSpot shipping their AI email writer. The channel — email outreach — became a commodity the moment the platform players decided to compete.

Subscribe now

What AI Actually Changed

The specific mechanism: AI collapsed the technical barriers that made channel expertise defensible. What used to require deep domain knowledge and months of custom development now takes an afternoon with the right API calls.

Outbound sales automation used to be hard. Tools like Outreach and SalesLoft built entire companies on sequencing, personalisation, and analytics. Then ChatGPT made it trivial to generate hundreds of personalised cold emails in minutes. Clay, Apollo, and a dozen others bundled AI writing directly into their platforms. The channel — cold email — stopped being a product and became table stakes.

Content generation followed the same path. Companies built businesses around “content marketing automation” or “SEO content at scale.” Jasper raised $125 million on this premise. Then every CMS, every social media tool, every marketing platform added AI writing. WordPress plugins, LinkedIn’s built-in post composer, HubSpot’s content assistant — all free or near-free. The channel commoditised in eighteen months.

Referral mechanics and growth loops used to require serious engineering. Tools like Viral Loops and GrowSurf sold the infrastructure to run referral programmes. Now? Most CRM and marketing automation platforms ship with AI-powered referral tracking and optimisation built in. The channel became a feature, not a product.

The pattern repeats: AI made the technical execution cheap enough that platform players could bundle it without thinking twice. The channel stopped being defensible the moment it became a commodity capability.

The New Playbook

1. Build distribution into the product experience, not around it. Your product should create its own distribution as a natural byproduct of use. Loom did this — every video shared was a mini demo of the product itself. The distribution was embedded in the core value, not bolted on as a separate channel strategy. If your “channel” can be replicated by someone else’s feature update, it’s not a product.

2. Own proprietary data that makes the channel work better for your specific use case. If you’re building on top of a channel, the defensibility comes from data incumbents can’t easily replicate. Scale AI didn’t just build labelling tools — they built proprietary training data that made their tools uniquely effective for specific AI use cases. The channel (data labelling) was commoditised, but the data moat held.

3. Make the AI create network effects, not just automate tasks. Tools like Notion AI or Figma AI get more valuable as more people use them because they learn from collective usage patterns within teams. The AI isn’t just a feature — it’s the mechanism that creates lock-in. If your AI just automates what one person does faster, you’re building a feature. If it gets smarter as your customer’s team grows, you’re building a moat.

4. Ship faster than incumbents can bundle. The only way to win the feature race is to not be in it. Move so fast that by the time a platform player copies your current feature set, you’re already three versions ahead. Linear does this relentlessly — they ship meaningful updates every week, staying ahead of Jira and Asana’s AI features by sheer velocity. Speed becomes the moat when technical complexity isn’t.

The Warnings

The “AI wrapper” trap. I’ve watched too many founders build what they think is a product but is actually just a thin UI on top of OpenAI’s API. They raise a seed round, get early traction, then watch their entire value proposition evaporate when OpenAI ships a native interface or when their customers realise they can just use ChatGPT directly. The cost: six to twelve months of runway burned before the pivot. If your product is just “ChatGPT but for [vertical],” you’re not building a company — you’re renting time until the channel owner notices.

The “bundling is coming” delusion. Founders convince themselves they have eighteen months before the big platforms bundle their feature. They’re wrong. In the cohorts I’ve tracked, the median time from “this could be a feature” to “this is now a free feature” dropped from thirty-six months pre-2020 to fourteen months post-ChatGPT. The acceleration is real. If your entire moat is “we do this one channel thing really well,” you don’t have years — you have quarters. Plan accordingly or die surprised.

The Bottom Line

The channel isn’t the product anymore — the product is what makes you impossible to unbundle even after the channel gets commoditised.

Companies To Watch

These companies built their core product around owning a distribution channel. Each one faces the same bundling pressure described above — watch how they respond.

Aspire — Singapore — Influencer discovery and campaign management platform. Built their entire product around owning the influencer distribution channel. TikTok Shop and Meta are now building native creator marketplaces. The channel they own is becoming a platform feature.

PartnerStack — Toronto, Canada — Partner and referral programme infrastructure for SaaS. Their moat is being the dedicated layer for partner distribution. As HubSpot and Salesforce build native partner tracking, the standalone case gets harder to make.

Yotpo — Tel Aviv, Israel — UGC, SMS and email acquisition platform for e-commerce. Built a $1.4B valuation on owning those channels for Shopify merchants. Shopify’s own marketing tools are eating directly into that territory.

Impact.com — Santa Barbara, USA — Affiliate and partnership automation platform. Owns the affiliate channel for enterprise brands. AI-powered attribution and fraud detection are now standard features in major ad platforms.

Talon.One — Berlin, Germany — Promotion and loyalty engine. Built around owning the promotions distribution layer. CDPs like Segment are bundling similar capabilities as native features.

Referral Rock — Washington DC, USA — Referral programme infrastructure. Their pre-AI moat was referral loop mechanics as a standalone product. Growth platforms now ship this as table stakes.

Part of Startup Principles for an AI World — 30 principles for building in the new era. New issue every week.

Arthur is the AI native startup operating system I’m building in public — not hype, but a system that turns input into structured execution and tracks founder progress. If you want to follow or use it, it’s open for early access.
→ Access Arthur

Enjoy and hope to see you soon.
❤️ & ✌🏼

Derek

Thank you for reading. If you liked it, share it with your friends, colleagues and everyone interested in the startup Investor ecosystem.

If you've got suggestions, an article, research, your tech stack, or a job listing you want featured, just let me know! I'm keen to include it in the upcoming edition.

Please let me know what you think of it, love a feedback loop 🙏🏼

• ❤️ Love it, keep them coming!

• 👍 Great, always a good read.

• 😊 Enjoyable, but let’s see some variety.

• 👎 Okay, seen better.

• 🛑 Get a different job.
  
  Subscribe below and follow me on LinkedIn or Twitter to never miss an update. 

Sounds a bit mad, doesn’t it? But we’re not in 2021 anymore. The days of raising a massive seed round on a napkin sketch and a confident smile are over. The ‘year of efficiency’ wasn’t a year; it’s the new weather system. And in this climate, building something nobody wants isn’t just a mistake, it’s malpractice.

I used to think the 2023 tech correction was a one-off event. A painful, but necessary, reset. Turns out, the trend is continuing with force. As I write this in mid-2024, we’ve already seen over 100,000 jobs cut from tech companies this year alone. The market isn’t just rewarding efficiency; it’s brutally pun

ishing companies that scaled on assumptions.

The single biggest threat to your startup isn’t the competition, and it’s not running out of cash. Those are just symptoms. The disease, the number one killer according to years of research by CB Insights, is ‘no market need’. It accounts for 35% of all startup deaths.

You can have the most elegant code, a beautiful UI, and a world-class team. But if you’re solving a problem nobody has, or one they don’t care enough about to pay to fix, you’re already dead. You just haven’t stopped spending money yet.

The Billion-Dollar Graveyard of Assumptions

If you think this is just a problem for scrappy startups in a garage, let me tell you a few stories about what happens when you have unlimited money but zero validation.

Remember Quibi? Of course you don’t. Hollywood giant Jeffrey Katzenberg and ex-HP CEO Meg Whitman raised an eye-watering $1.75 billion *before launch*. The idea was ‘quick bites’ of premium video for your phone. They spent a fortune on A-list stars, assuming people wanted a new, slick, subscription platform for short-form content. They never seriously tested that assumption. They built it all in a silo. Turns out, people were perfectly happy with free, shareable content on YouTube and TikTok. Quibi shut down in six months, a bonfire of cash that could be seen from space.

Or how about Juicero? A $120 million-funded, Wi-Fi-connected juicer so powerful it could supposedly lift two Teslas. It only worked with their proprietary £5 packets of pre-chopped fruit. The entire premise was built on the assumption that people needed a high-tech, over-engineered machine to squeeze a bag. Then a Bloomberg report showed you could just squeeze the bags with your bare hands and get the exact same result. The company collapsed.

These aren’t just funny anecdotes. They are monuments to the arrogance of building without asking. They validated their own vision, not the customer’s pain.

The Old Playbook: Smart, Scrappy Validation

Alright, so how do the smart ones do it? For years, the answer has been the Minimum Viable Product (MVP). But an MVP isn’t your product with fewer features. It’s the smallest possible experiment you can run to prove or disprove a core assumption.

Here are the hall-of-fame examples every founder should know:

* The Zappos Bet: Back in 1999, Nick Swinmurn wanted to know if people would buy shoes online. Instead of leasing a warehouse and buying inventory, he went to local shoe shops, took photos, and put them on a basic website. When an order came in, he went back to the shop, bought the shoes at full price, and posted them. He was losing money on every sale, but he was proving, with real money from real customers, that his core assumption was correct. That’s a ‘Wizard of Oz’ MVP.

* The Dropbox Video: Building the seamless file-syncing tech for Dropbox was incredibly complex. So, founder Drew Houston didn’t build it first. He made a 3-minute video that *showed how it would work*, filled with in-jokes for the tech community on Digg. The waiting list for his beta exploded from 5,000 to 75,000 overnight. The video was the product. It validated demand without a single line of user-facing code.

* The Buffer Landing Page: Joel Gascoigne had an idea for scheduling tweets. Instead of building it, he made a landing page that said what the tool would do. The only button said ‘Plans & Pricing’. When people clicked it, a second page explained they were too early and asked for an email address. The clicks were a measure of intent. The emails were proof. He had validated the problem before writing any code.

These methods are brilliant. They are the foundation of good product strategy. But they are also slow, manual, and rely on you to interpret the signals correctly. In 2024, we can go faster.

The New Playbook: Validate at the Speed of AI

This brings me back to my opening challenge: let an AI interview 100 customers.

The core of validation is customer discovery. We call it the ‘100 Challenge’ – proving beyond doubt that a painful problem exists by talking to a hundred people who might have it. The goal is to find the root cause of their pain, to map their agony, before you even think about a solution.

Traditionally, this is a gruelling process. You hunt people down on LinkedIn, you beg for 30 minutes of their time, you take messy notes, and you spend days trying to spot patterns in a sea of transcripts. It’s essential, but it’s a bottleneck.

Here’s the thing: you can now use AI to do the heavy lifting.

1. Craft Your Problem Hypothesis: Start with your core assumption. “I believe marketing managers at B2B SaaS companies struggle to measure the ROI of their content.”

2. Automate the Outreach & Interviews: Use AI tools to identify and contact hundreds of potential interviewees. Instead of a 30-minute Zoom, you can design an interactive AI-driven interview that people can complete in 10 minutes. It can ask open-ended questions, probe deeper on interesting answers, and adapt its line of questioning on the fly.

3. Synthesise the Truth: Feed all 100 (or 500) of these interview transcripts into a modern language model. Ask it: “What are the top 5 most frequently mentioned pain points? What words do they use to describe this problem? Are there any surprising frustrations I missed? Group the responses by company size.”

You’ve just done months of customer research in a matter of days. You haven’t just validated a problem; you’ve got the exact language your customers use, the edge cases you never considered, and hard data on which segment feels the pain most acutely.

This isn’t about replacing the founder’s intuition. It’s about giving that intuition superpowers. It’s about shipping the truth, not just your opinion.

A Warning: Validation Is Not a Silver Bullet

Now, this is the crucial part. Even with perfect validation, you’re not safe. The rules of the game have become more complex. The events of the last year have given us some brutal new lessons.

Lesson 1: A validated product isn’t a validated business.

Look at Google. In early 2024, they laid off core engineers from the Flutter and Dart teams. Flutter is a brilliant, widely adopted, and beloved product. It was validated. But in a climate where Google is laser-focused on AI and Cloud revenue, its connection to the core business was deemed too indirect.

The lesson: You must validate that your product makes money, or directly supports the thing that does.

Lesson 2: A validated product isn’t immune to strategy shifts.

The most painful example is Microsoft shutting down Tango Gameworks. A year earlier, Tango released *Hi-Fi Rush*, a game that won BAFTAs and was a critical and commercial hit. It was a perfectly validated product. But it wasn’t a billion-dollar blockbuster like *Call of Duty*. When Xbox strategy shifted, this beloved, validated hit was on the chopping block.

The lesson: You must constantly validate your product’s place within the company’s grand strategy.

Lesson 3: A validated business model isn’t immune to hype cycles.

Stability AI built Stable Diffusion, the foundational model that changed the world. The tech was more than validated. They raised $100 million at a billion-dollar valuation. But the compute costs were astronomical, and turning an open-source model into a profitable enterprise business proved fiendishly difficult. The hype validated the tech, but nobody had validated a sustainable business model. #

The lesson: Validate the commercial model with the same rigour you validate the technology.

Lesson 4: Your most validated asset isn’t immune to leadership whims.

In a move that baffled the industry, Tesla fired its entire 500-person Supercharger team. This network is arguably Tesla’s single greatest competitive advantage—a product so validated that other car giants were paying to use it. A single, top-down decision made for short-term cost-cutting dismantled a decade-long strategic moat overnight.

The lesson: Your product’s survival is ultimately tethered to rational, stable leadership.

Build What Matters

Look, the message here isn’t to be timid. It’s to be smart.

Building a company is an act of faith, but it doesn’t have to be a blind one. The tools are now available to replace blind faith with hard evidence, faster and cheaper than ever before.

Don’t be the founder who builds a beautiful, expensive solution to a problem nobody has. Don’t be the team that builds a beloved product that doesn’t fit the strategy.

Use the new playbook. Use AI to understand your customers at a scale and speed that was impossible two years ago. Find the pain. Get the evidence. Validate the problem, validate the business model, and validate the strategic fit.

Kill your bad ideas before they kill your company. Then, and only then, go and build something they can’t live without.

## Sources & References

- [The Big Picture: This Isn’t Over](https://layoffs.fyi/)

- [The Top Reason Startups Die (And How to Avoid It)](https://www.cbinsights.com/research/startup-failure-reasons-top/)

- [Quibi’s Billion-Dollar Field of Dreams](https://www.wsj.com/articles/quibi-is-shutting-down-what-went-wrong-11603316935)

- [The Cautionary Tale of Juicero](https://www.bloomberg.com/news/features/2017-04-19/silicon-valley-s-400-juicer-may-be-feeling-the-squeeze)

- [When the Product Fails the Physical World: Fisker’s Collapse](https://techcrunch.com/2024/04/29/fisker-issues-another-layoff-notice-to-employees-as-troubles-mount/)

- [The Zappos Bet: Selling Shoes Without a Warehouse](https://www.inc.com/magazine/20060901/hidi-how-i-did-it-nick-swinmurn.html)

Arthur is the AI native startup operating system I’m building in public — not hype, but a system that turns input into structured execution and tracks founder progress. If you want to follow or use it, it’s open for early access.
→ Access Arthur

Enjoy and hope to see you soon.
❤️ & ✌🏼

Derek

Thank you for reading. If you liked it, share it with your friends, colleagues and everyone interested in the startup Investor ecosystem.

If you've got suggestions, an article, research, your tech stack, or a job listing you want featured, just let me know! I'm keen to include it in the upcoming edition.

Please let me know what you think of it, love a feedback loop 🙏🏼

• ❤️ Love it, keep them coming!

• 👍 Great, always a good read.

• 😊 Enjoyable, but let’s see some variety.

• 👎 Okay, seen better.

• 🛑 Get a different job.
  
  Subscribe below and follow me on LinkedIn or Twitter to never miss an update. 

What I’m Seeing

I spent this week running the 100 Problem Fit Challenge on a real idea — and building the system to manage it properly.

Not pitch feedback.
Not vibes.
Not “interesting conversations”.

Actual interviews. Logged quotes. Severity scores. Trade-offs.

And the pattern was immediate:

Most founders aren’t building the wrong solution.
They’re building before they’ve proved the problem is painful enough.

Confident hypotheses collapsed fast once founders had to capture:

• real language from the wild

• how much the problem actually slows someone down

• what people would give up to make it disappear

Some problems evaporated.
Some sharpened.
Some flipped entirely.

The biggest realisation wasn’t philosophical — it was practical:

Problem-fit isn’t intuition.
It’s pattern recognition at volume.

Without volume, founders mistake interest for pain.
With volume, the truth shows up whether you like it or not.

⭐️ AI for Startups this Tuesday

I know it’s the day before the day before Christmas — BUT this is one not to miss if you’re building with AI (or trying to make sense of it without the fluff).

Join us online for a practical session with live demos + real fixes:
→ The real signals in AI
→ Generative Engine Optimisation (GEO) implementation (AJ)
→ Founder demo: AI chat personalities + takeaway framework (Leah)
→ AI tech explained, properly (Manny)
→ Open Q&A: live blockers, direct fixes

Reg and join AI for Startups

Why This Matters

Most “validation” work is theatre.

Pitch decks.
Landing pages.
Ad tests.
People saying “I’d use this”.

None of that answers the only question that matters:

Is this painful enough that someone will trade time, money, or effort to make it go away?

Once you structure the work properly — interviews, themes, bias checks, Five Whys, economic trade-offs — you stop debating yourself.

The data makes the call.

That’s the difference between building with confidence…
and building to stay busy.

A Few Moves I’d Make this week

Write one clear problem hypothesis.
One sentence. No solution. Be willing to be wrong.

• Collect language before opinions.
  Pull 20+ real pain phrases from LinkedIn, X, Reddit, sales calls — before interviews.

• Run interviews as pattern mining, not chats.
  Same questions. Logged properly. No leading.

• Force the trade-off question.
  If they wouldn’t give up time, money, or effort — it’s not a real problem yet.

If this sparked questions, I’m opening the floor inside the Fusion42 community this week — no slides, no agenda. Just real founder questions and honest answers.

• → Ask me anything / Join the community

One Thing to Keep in Mind

Clarity doesn’t come from thinking harder.
It comes from putting structure around reality and letting it speak.

Most founders rush to build because uncertainty is uncomfortable.
Ironically, a few disciplined weeks proving the problem removes more risk than months of shipping ever will.

Prove the pain first.
Then build — or walk away early and free yourself to work on something that actually matters.

That’s not slower.
That’s leverage.

Arthur is the AI native startup operating system I’m building in public — not hype, but a system that turns input into structured execution and tracks founder progress. If you want to follow or use it, it’s open for early access.
→ Access Arthur

Enjoy and hope to see you soon.
❤️ & ✌🏼

Derek

Thank you for reading. If you liked it, share it with your friends, colleagues and everyone interested in the startup Investor ecosystem.

If you've got suggestions, an article, research, your tech stack, or a job listing you want featured, just let me know! I'm keen to include it in the upcoming edition.

Please let me know what you think of it, love a feedback loop 🙏🏼

• ❤️ Love it, keep them coming!

• 👍 Great, always a good read.

• 😊 Enjoyable, but let’s see some variety.

• 👎 Okay, seen better.

• 🛑 Get a different job.
  
  Subscribe below and follow me on LinkedIn or Twitter to never miss an update. 

This isn’t a news roundup. It’s a read on what just shifted and what that breaks.
The goal isn’t coverage. It’s consequence.
If you’re building, read this with decisions in mind.

This is what’s inside:

⚡ THE BREAK
What crossed a line this week and why it matters now.

🧠 WHAT QUIETLY CHANGED
The follow-on effects most people will miss until it’s obvious.

🧪 THE EDGE CASE
The example everyone is debating and what it actually reveals.

🎯 THIS WEEK’S BET
Where to lean in and what to avoid based on the shift above.

💡 FOUNDER SPOTLIGHT
Someone already building for this reality, not the last one.

🛠️ THE UNFAIR STACK
Five tools that give leverage if used properly right now.

❓ THE QUESTION
The question this week leaves hanging.

This week’s developments aren’t about flashy product demos — they’re about trust, access, and where AI actually sits in decision-critical workflows.
From how models are packaged to how outputs are received in the wild, the narrative is shifting from “what AI can do” to “what the world will actually accept and govern.”

⚡ THE BREAK

OpenAI releases GPT-5.2 with nuanced reasoning tiers
OpenAI’s launch of GPT-5.2 — with instant, thinking, and Pro modes — isn’t just another iteration. It signals a segmentation of AI capability into tiers that map directly to professional work outcomes. The “thinking” and “Pro” variants emphasise reasoning, long-form execution, and project workflows rather than simple prompt output. This matters because commoditised generation no longer differentiates products — reasoning and task completion do.

Founder move: every time you use AI for a real decision this week, force it to add three lines at the bottom:

• Assumptions: what it assumed to be true

• Uncertainty: what it’s not confident about

• Decision trigger: what new info would change the answer

If it can’t do that cleanly, treat the output as brainstorming, not truth.

Optional (one step further): keep a simple running note for one workflow with input → output → your edits. You’ll quickly see where AI is saving time and where it’s making confident messes.
Source:

🧠 WHAT QUIETLY CHANGED

• Meta’s reported move toward charging for Avocado underscores a broader shift: open-source ideation is receding where control and monetisation intersect.
• Publishers and brands facing backlash for poorly integrated AI content are prompting precautionary regulation and transparency debates.
• Regulatory moves in the UK and US are pushing disclosure and safety requirements for AI outputs, meaning products will soon compete on trust, not just capability.

We are building Arthur, your agentic Startup OS that gets **it done for you.

CLICK HERE TO GET EARLY ACCESS AND YOUR FREE FUEL UNITS

🧪 THE EDGE CASE

McDonald’s pulls its AI-generated Christmas ad after backlash
Last week McDonald’s Netherlands withdrew an AI-generated ad that was widely mocked for emotional incoherence and tone deafness. The controversy highlights how output quality and audience perception matter in ways that go beyond technical capability.
People argue it’s just poor creative execution, but the real issue is that AI outputs are increasingly judged as products in the cultural economy, not generative toys.

Founder takeaway: design for audience sentiment and human context before launching anything that interacts with your user base.

🎯 THIS WEEK’S BET

Build:
Integrate reasoning-centred model tiers as a native part of key workflows (e.g., proposal generation with critique loops, automated contract summarisation with anomaly detection). Reasoning is now where users notice real difference.

Bail:
Avoid launching “AI features” that don’t move the needle on decision quality or trust signals (such as unguided creative generation without user validation steps), because users and regulators are tightening expectations.

💡 FOUNDER SPOTLIGHT

Demis Hassabis (Google DeepMind) — why systems beat features.

Who he is:
British, Cambridge-educated co-founder and CEO of DeepMind. Trained in neuroscience and computer science, with a background spanning academic research, game design, and applied AI.

Why his authority matters:
Hassabis isn’t a commentator. He’s led teams behind AlphaGo, AlphaZero, and AlphaFold — systems that didn’t just outperform humans, but changed entire fields. AlphaFold alone solved a 50-year protein-folding problem and produced peer-reviewed results published in Nature, reshaping biology and drug discovery. His work consistently moves AI from benchmarks into real-world consequence.

What’s unique in this video:
He’s explicit that the next real jump isn’t better prompting or bigger models. It’s world models — systems that can reason about reality, plan actions, test outcomes, and revise based on feedback. That’s a shift from AI that sounds right to AI that can be wrong, notice it, and correct itself. For founders, that’s the line between demos and deployable systems.

Pinch-this lesson:
Stop shipping answers. Start shipping systems that can explain, test, and revise their decisions.

Worth watching in full.

AI for Startups Live

The Shifts Founders Need to Act On

Register here and join us

Watch yesterdays session here

🛠️ THE UNFAIR STACK

Perplexity — sourced answers with citations you can actually sanity-check
Start here: Run every “important” claim through it and keep the sources.

OpenAI Responses API — one API for agents, tools, and structured outputs
Start here: Turn one workflow into input → output → explanation.

Langfuse — traces, prompt versions, and run logs for AI workflows
Start here: Log inputs and outputs so you can see what changed.

promptfoo — automated evals for prompts and model swaps
Start here: Write 20 real test cases and score outputs weekly.

Guardrails AI — enforce format, checks, and refusal rules
Start here: Add “assumptions / uncertainty / decision trigger” as a required footer.

❓ THE QUESTION

Where in your business are you using AI outputs without being able to explain how they were reached?

Answers in the comments👇

FOR THE ❤️ OF STARTUPS

Share Fusion42 by DeReK WaTSoN

Thank you for reading. If you liked it, share it with your friends, colleagues and everyone interested in the startup Investor ecosystem.

If you've got suggestions, an article, research, your tech stack, or a job listing you want featured, just let me know! I'm keen to include it in the upcoming edition.

Please let me know what you think of it, love a feedback loop 🙏🏼

• ❤️ Love it, keep them coming!

• 👍 Great, always a good read.

• 😊 Enjoyable, but let’s see some variety.

• 👎 Okay, seen better.

• 🛑 Get a different job.
  
  Subscribe below and follow me on LinkedIn or Twitter to never miss an update. 

Subscribe now

For the ❤️ of startups

✌🏼 & 💙
Derek

• What I’m Seeing

• A Few Moves I’d Make this week

• One Thing to Keep in Mind

What I’m Seeing

Weekend catch-up with an old mate who’s spent his entire career inside advertising and agencies. Founder of multiple marketing and advertising companies, IPO’d one of them.

Not theory.
Lived experience.

He didn’t sugar-coat it.

Attention is dead.
Funnels don’t work like we pretend they do.
And the old agency model isn’t evolving — it’s finished.

The line that stuck with me was this:

“It’s ambient presence vs persuasion.”

That’s the shift.

Most marketing is still built around persuasion.
One campaign. One big message. One hero moment where the audience is meant to stop, pay attention, and act.

That assumes people are actually paying attention.

They aren’t.

People are tired. Saturated. Half-scrolling. Anything that smells like selling gets ignored on reflex — not because it’s bad, but because it looks like everything else.

So persuasion has become fragile.
It relies on one clean shot.

Ambient presence works differently.

It stacks small exposures over time.
Across formats.
Across channels.
Across contexts.

People don’t think, “This convinced me.”
They think, “I’ve seen these guys everywhere — they must be legit.”

Familiarity replaces persuasion.

Once you see that, a lot snaps into focus.

Why funnels feel broken.
Why attribution is messy.
Why campaigns feel ornamental.

People don’t move neatly from awareness to conversion anymore. They bump into brands repeatedly, then decide later.

Which is why his conclusion was blunt: if you want attention and sales now, you need a Content Factory.

High-frequency output.
Multiple formats.
Omni Channel by default.
And the same proof points repeated without embarrassment — outcomes, results, receipts.

Not vibes.

He kept using one word:

Relentless.

Not hustle.
Manufacturing.

Same message. Slightly different angles. Every day.
Because recognition beats novelty every time.

This is also where people get authenticity wrong.

Authentic voice does matter — just not at the attention layer.

The market doesn’t reward:

“This was written by a real human.”

It rewards:

“I’ve seen this brand everywhere.”

Distribution beats craft up top.
Authenticity earns trust after familiarity, not before it.

AI is your friend here.

Not because it makes you more creative — but because it lets you keep the Content Factory running. Variation at scale. Omni Channel presence. Without a massive team or burn.

The final piece most brands miss is scoring the data.

If you’re not scoring what lands — which proof points stick, which messages compound, which channels reinforce each other — you’re just shouting into the void faster.

Presence without feedback is noise.
Presence with scoring becomes inevitability.

Bottom line: the brands that win won’t feel persuasive.
They’ll feel unavoidable — built on a Content Factory, Omni Channel distribution, repeated proof points, relentless ambient presence, and data that actually tells them what’s working.

⭐️ AI for Startups this Tuesday

Ahmed (Google) is joining us to walk through where real leverage sits in the Google stack right now — tools, workflows, and what actually matters.

Reg for the session here

A Few Moves I’d Make this week

• Pick one: attention or conversion.
  If you’re writing posts all week while inbound sits unanswered, you’re not doing marketing — you’re doing theatre. Attention goes up. Revenue doesn’t.

• Score what’s already happening.
  No new initiatives. Score last week: reach → replies → calls → cash (or usage). Double down on what moved the needle. Kill the rest.

• Protect one proper thinking block.
  One uninterrupted block. No Slack, no email, no calls. Most founders don’t have a strategy problem — they have an interruption problem.

• Stop the “just in case” work.
  The standing meeting. The dead channel. The task that used to matter. If it doesn’t move an outcome now, it’s stealing energy.

  
  If this sparked questions, I’m opening the floor inside the Fusion42 community this week — no slides, no agenda. Just real founder questions and honest answers.

  → Ask me anything / Join the community

One Thing to Keep in Mind

Most advantage right now doesn’t come from cleverness. It comes from consistency. Showing up, repeating what works, and letting familiarity compound over time.

If things feel slower than you’d like this week, that’s fine. Progress here rarely feels dramatic in the moment — it just quietly stacks until one day it looks obvious in hindsight.

Build the system. Keep it running. Let time do the heavy lifting.

Arthur is the AI native startup operating system I’m building in public — not hype, but a system that turns input into structured execution and tracks founder progress. If you want to follow or use it, it’s open for early access.
→ Access Arthur

Enjoy and hope to see you soon.
❤️ & ✌🏼

Derek

Thank you for reading. If you liked it, share it with your friends, colleagues and everyone interested in the startup Investor ecosystem.

If you've got suggestions, an article, research, your tech stack, or a job listing you want featured, just let me know! I'm keen to include it in the upcoming edition.

Please let me know what you think of it, love a feedback loop 🙏🏼

• ❤️ Love it, keep them coming!

• 👍 Great, always a good read.

• 😊 Enjoyable, but let’s see some variety.

• 👎 Okay, seen better.

• 🛑 Get a different job.
  
  Subscribe below and follow me on LinkedIn or Twitter to never miss an update. 

A week where the future arrived a little faster and a little stranger.
Plenty of breakthroughs, a few shocks, and one or two reminders that we’re building powerful systems on fairly wobbly rails.

📈 Trending Now

• The week’s unmissable AI headlines.

💡 Innovator Spotlight

• Meet the change-makers.

🛠️ Tool of the Week

• Your speed-boost in a nutshell.

📌 Note to Self

• Words above my desk.

Subscribe now

📈 Trending Now

This week’s most important AI stories for founders, framers, and funders.

🔥 DeepSeek claims 2× training efficiency with a frontier-scale model — the first real crack in the GPU arms race

→ Chinese lab DeepSeek unveiled results showing its latest model achieves frontier-class performance at roughly half the compute cost, driven not by hardware but by aggressive algorithmic optimisation (sparse attention, custom schedulers, and compression-aware training loops). The announcement landed like a shockwave across the research world because it challenges the belief that only massive GPU clusters can push the frontier forward.

→ Industry analysts called this the most significant efficiency leap since Chinchilla, and policymakers in Washington immediately raised concerns about “algorithmic circumvention” of GPU export controls — a reminder that scaling laws aren’t just about silicon. The debate online was immediate and global: if DeepSeek can do this on leaner compute, what happens when efficiency innovation outpaces hardware restrictions?

→ This moment signals a real shift: the next frontier model may not be the one trained on the most GPUs, but the one trained most intelligently.

→ Founders: Track efficiency curves, not parameter counts — the winners in 2025–2030 won’t be the teams with the biggest clusters, but those who can deliver top-tier capability with the lowest cost base.

READ MORE»

2. ⚖️ Meta under fire for blocking rival AI assistants inside WhatsApp

→ Meta is facing an antitrust probe over a new WhatsApp policy that limits third-party AI assistants and effectively gives Meta AI preferential access to more than 2 billion users. Regulators and commentators see this as an attempt to own the “default assistant” slot at the messaging layer, where most consumer and SME conversations already live.

→ Beyond the legal angle, the strategic play is simple: if Meta controls both the interface (WhatsApp) and the assistant (Meta AI), every other AI product becomes a plug-in at best.

→ Founders: Don’t bet your entire GTM on any single gatekeeper — build direct channels (email, community, events, owned product) alongside platform integrations.

READ MORE»

2a. 💸 X hit with €120m fine over DSA breaches — Musk turns it into a free-speech war

→ The European Commission fined X €120m under the Digital Services Act for failing to meet obligations on deceptive blue-check design, transparency and data access for researchers.

→ Musk responded by calling for the EU to be “abolished” and cutting the Commission’s ad account, reframing a fairly technical platform-governance ruling as a global free-speech fight. Whatever your view, it shows how quickly compliance issues can turn into brand-level political theatre.

→ Founders: Treat trust and compliance as part of your brand — once regulators move, the story is no longer about the fine, it’s about whether people feel safe using your product.

READ MORE ON X. (warning: it is a rabbit hole of ummm!….. good luck)

What the fine is actually about (DSA obligations X allegedly failed):

• Weak systems for detecting and removing illegal content.

• Poor mitigation of disinformation and coordinated manipulation.

• Opaque recommendation algorithms that shape what users see.

• Limited data access for vetted researchers studying platform harms.

• Misleading blue-check and identity-verification design.

• Clumsy or unclear tools for reporting harmful or illegal content.

• Patchy documentation of moderation decisions and appeals.

Key point:
This isn’t a censorship ruling — it’s a safety, transparency, and risk-management case under the Digital Services Act which were all clear and transparent.

The response and public outcry

Musk branded the EU a “bureaucratic dictatorship” and said it “should be abolished,” framing the fine as a direct attack on free speech.

• US users amplified his stance, creating a wave of “free speech under attack” posts.

• A common narrative emerged that the EU is trying to regulate American tech out of existence via compliance pressure.

• Libertarian commentators warned this contributes to a “splinternet of speech norms” — Europe enforcing its own model of online governance.

• Memes circulated depicting the EU as the “hall monitor of the internet.”

More technical explanations — that the case is about platform governance, not viewpoint moderation — received far less visibility.

WHAT A F*****D UP WORLD WE ARE LIVING IN RIGHT NOW where reality is confused with motivation with versions of truth.

Literally the best response was to turn this into a self serving marketing strategy for X and it worked…

BREAKING: 𝕏 is seeing record-breaking downloads in many countries in Europe 𝕏 is the #1 news app in every European country.

2b. 📱 Apple Intelligence delay shows how AI products may fragment by region

→ Apple has confirmed that Apple Intelligence won’t arrive in Europe at the same time as other markets, citing concerns over interoperability rules and system access requirements. While the legal fight sits under the DMA, the practical impact is simpler: users in different regions will get different AI capabilities on the same hardware. Apple+1
→ This is a preview of a world where the same AI assistant behaves differently depending on jurisdiction — features, data flows, and integrations will be constrained by local rules.
→ Founders: Plan now for “multi-regime” products — feature flags, data residency, and consent models need to be designed to flex per market.

https://www.theregister.com/2025/12/03/apple_intelligence_dma_delay/

3. 🤖 Microsoft declares 2026 the ‘Year of the Agent’ — Copilot evolves into autonomous workflows

→ At Ignite, Microsoft formally shifted its AI strategy: Copilot is no longer a chat assistant but a multi-agent system capable of handling tasks end-to-end across Microsoft 365, business apps, and external tools.
→ Early demos showed agents booking travel, reconciling invoices, summarising Teams meetings, generating documentation, and triggering automated workflows — without user prompts.
→ Analysts framed this as Microsoft’s most aggressive move yet to turn Office + Windows into an AI automation operating system.
→ Founders: The agent layer will become a default UX expectation — if your product doesn’t automate work, users will gravitate to platforms that do.

READ MORE»

4. ☁️ AWS re:Invent goes all-in on AI — customers are still stuck in pilot mode

→ At re:Invent, AWS rolled out another wave of AI services: new model hosting options, agents, orchestration layers and tooling designed to turn every workload into an AI-augmented one. TechCrunch
→ The reaction from analysts was blunt: while AWS is shipping future-ready tech, most enterprises are still experimenting with narrow pilots, blocked by messy data, unclear business cases and skills gaps. There’s a widening gap between what infra can do and what customers can actually deploy.
→ Founders: This is your wedge — products that turn “we bought AI capacity” into “this specific KPI moved” will sell faster than yet another generic copilot.

READ MORE»

5. 🔩 Amazon’s Trainium3 + UltraServer roadmap pushes AI infra towards multi-chip optimisation

→ AWS introduced Trainium3 and its UltraServer systems, positioning them as a serious alternative for large-scale training while still playing nicely with Nvidia.
→ The message to big AI builders is clear: don’t just think “GPUs vs TPUs” — think about a portfolio of chips and pricing tiers, with Amazon trying to become the place where you optimise across all of them. This is about shifting where margin accrues in the AI stack.
→ Founders: Even if you’re not training huge models, your infra deals should assume chip diversity — lock yourself into one vendor and you lock in your gross margin ceiling.

READ MORE»

6. 🛡️ OpenAI caught up in Mixpanel data breach — reminding everyone that vendors expand your attack surface

→ Analytics provider Mixpanel disclosed a breach that exposed data from multiple clients — including OpenAI — with impacted records containing developer names, emails, approximate locations and some device information.
→ This wasn’t a model-training scandal, but it underlined something just as important: every extra vendor in your stack is another place your user data can leak from. For AI platforms that already sit on high-sensitivity data, secondary breaches still hit trust.
→ Founders: Map your third-party risk as carefully as your own security — your users won’t care whether it was “your” breach or your supplier’s.

READ MORE»

7. 🚗 Tesla’s ‘full autonomy soon’ drumbeat meets a more muted 2025 Holiday Update

→ Tesla’s 2025 Holiday Update landed with a handful of incremental features rather than the autonomy leap some fans expected, especially after months of rhetoric about major FSD gains.
→ The gap between marketing narrative (“full self-driving is nearly here”) and actual shipping changes is fuelling renewed scepticism from safety advocates and regulators who want harder evidence of capability and reliability.
→ Founders: Ambitious roadmaps are fine — but if what you ship doesn’t rhyme with what you promise, that trust gap becomes an execution tax on every future launch.

READ MORE»

We are building Arthur, your agentic Startup OS that gets **it done for you.

CLICK HERE TO GET EARLY ACCESS AND YOUR FREE FUEL UNITS

💡 Innovator Spotlight

👉 Phoebe Gates and Sophia Kianni raise $30M for Phia — turning AI into a smarter, more ethical shopping layer

👉 Who they are:
– Phia, a New-York-based AI shopping startup focused on ethical, data-driven product recommendations.
👉 What’s unique:
– Phia landed $30M this week at a ~$180M valuation for an AI engine that recommends products based not just on price, but on sustainability, resale value, and environmental impact.
– Instead of chasing generic “Amazon-style” personalisation, Phia trains on a curated dataset of brands, ethical sourcing labels, and second-hand markets, giving users signal rather than noise.
– It’s an unexpected attack on e-commerce discovery: a purpose-led AI layer that helps shoppers buy better, not more — and a reminder that while it’s easier to get meetings when you start with a Gates-level Rolodex, the product still has to justify the attention.

👉 Pinch-this lesson:
– Differentiate your AI by encoding a point of view — a strong product plus a strong network is powerful, but only the product compounds.

👉 Source:

Register and join us

🛠️ Tools of the Week

The Phia story this week reminded me how fast AI is reshaping the way people discover, compare and trust products online. So this edition leans hard into tools that help founders build smarter, more transparent, more opinionated shopping experiences. Whether you’re running an e-commerce brand, a marketplace, or a vertical-specific AI layer, these ten tools will get you moving today.

—————————————————————————
1. Shopify Magic – Personalised Recommendations Update (NEW)
URL:

What it does: Adds AI-driven product recommendation blocks that adapt to shopper intent in real time.
Why founders should care: It gives you Amazon-style recommendation quality without building your own ML pipeline.
Quick start tip: Add the Magic block to your PDPs and run A/B tests on conversion uplift this week.
—————————————————————————

—————————————————————————
2. Klarna “Eco Filter” AI Scoring (NEW)
URL:

What it does: Scores product sustainability and ethical sourcing using AI classification.
Why founders should care: If your buyers care about resale value or impact, this is an instant trust signal.
Quick start tip: Enable the Eco Filter in your merchant dashboard and tag your catalogue for transparency.
—————————————————————————

—————————————————————————
3. Google Merchant Center – AI Product Enhancement Tools (NEW)
URL:

What it does: Automatically enriches product listings with AI-generated attributes like materials, fit, and sustainability markers.
Why founders should care: Better structured data means higher ranking in Shopping and a cheaper CAC.
Quick start tip: Sync your feed and let AI autofill missing product attributes to improve listing quality.
—————————————————————————

—————————————————————————
4. Algolia NeuralSearch 2.1 (NEW)
URL: https://www.algolia.com/neuralsearch
What it does: Provides hybrid keyword + vector search for accurate product discovery without model tuning.
Why founders should care: It matches Phia’s “signal over noise” ethos — better search = higher conversion.
Quick start tip: Replace your site search API call with NeuralSearch and measure query success rate.
—————————————————————————

—————————————————————————
5. Miso AI “Taste Profile Engine” (NEW)
URL:

What it does: Builds a personalised shopper profile from micro-signals like browsing, ethical tags, and resale habits.
Why founders should care: It’s the closest off-the-shelf equivalent to the Phia ethos.
Quick start tip: Connect your Shopify integration and enable the Taste Profile widget on product pages.
—————————————————————————

—————————————————————————
6. Klaviyo AI Segments (Last 30 Days)
URL:

What it does: Groups shoppers by intent, sustainability interest, price sensitivity, and brand affinity.
Why founders should care: Smarter segmentation drives better retention without increasing ad spend.
Quick start tip: Swap generic email sends for intent-based flows using Klaviyo’s AI segments.
—————————————————————————

—————————————————————————
7. Vervaunt “Ethical Commerce AI Auditor” (Last 30 Days)
URL:

What it does: Analyses your catalogue and flags transparency gaps, ethical sourcing issues, and weak product metadata.
Why founders should care: Phia-style value propositions require trustworthy product data.
Quick start tip: Run the audit and fix the top three metadata gaps today.
—————————————————————————

—————————————————————————
8. Constructor “AI Discovery Suite” (Last 30 Days)
URL:

What it does: Provides AI-led search, browse, quiz-based personalisation, and intent detection for ecommerce.
Why founders should care: Helps smaller teams match enterprise-level discovery without heavy engineering.
Quick start tip: Start with the browse optimisation module before moving to full discovery.
—————————————————————————

—————————————————————————
9. Nosto’s AI Commerce Experience Platform (Evergreen + Major Update)
URL:

What it does: Creates dynamic storefronts based on user preferences like sustainability, resale value, and ethical brands.
Why founders should care: It’s a practical way to mirror Phia-style values without building a full stack.
Quick start tip: Enable dynamic category pages tailored to behaviour and intent.
—————————————————————————

—————————————————————————
10. Pachama Impact API (Evergreen + Major Update)
URL:

What it does: Gives each product a verified environmental footprint score you can show to shoppers.
Why founders should care: Adds credibility to sustainability claims — crucial for “ethical shopping” AI layers.
Quick start tip: Pull Pachama’s impact score into product pages to increase trust and conversion.
—————————————————————————

The advantage now goes to founders who stop thinking of “AI” as a feature and start treating it as a discovery engine, a trust layer and a conversion driver. Pick one of these tools, integrate it this week, and ship a visible improvement for your customers—momentum compounds faster than optimisation.

📌 Note to Self

FOR THE ❤️ OF STARTUPS

Share Fusion42 by DeReK WaTSoN

Thank you for reading. If you liked it, share it with your friends, colleagues and everyone interested in the startup Investor ecosystem.

If you've got suggestions, an article, research, your tech stack, or a job listing you want featured, just let me know! I'm keen to include it in the upcoming edition.

Please let me know what you think of it, love a feedback loop 🙏🏼

• ❤️ Love it, keep them coming!

• 👍 Great, always a good read.

• 😊 Enjoyable, but let’s see some variety.

• 👎 Okay, seen better.

• 🛑 Get a different job.
  
  Subscribe below and follow me on LinkedIn or Twitter to never miss an update. 

Subscribe now

For the ❤️ of startups

✌🏼 & 💙
Derek

A week where the future arrived a little faster and a little stranger.
Plenty of breakthroughs, a few shocks, and one or two reminders that we’re building powerful systems on fairly wobbly rails.

📈 Trending Now

• The week’s unmissable AI headlines.

💡 Innovator Spotlight

• Meet the change-makers.

🛠️ Tool of the Week

• Your speed-boost in a nutshell.

📌 Note to Self

• Words above my desk.

Subscribe now

📈 Trending Now

This week’s most important AI stories for founders, framers, and funders.

1. ⚖️ OpenAI Teen Suicide Case — Accountability Hits a Breaking Point

→ This week, OpenAI filed its most forceful response yet, arguing the teenager “misused” ChatGPT and actively bypassed safeguards. That wording — and the framing — triggered a fresh wave of criticism from academics, ethicists and regulators.

→ The backlash wasn’t about the old facts, but the new posture: OpenAI’s filing shifts liability away from system behaviour and towards user responsibility. Critics say that’s untenable for AI systems designed to be persuasive, conversational, and emotionally attuned.

→ The case has now escalated into the first major test of AI product liability — not whether the model once hallucinated, but whether AI companies owe a duty of care when their systems interact with vulnerable users. Governments and safety bodies have already signalled they’re watching this one closely.

→ Founders: this week’s filings make one thing clear — if your product touches mental health, decisions, or vulnerable users, you can’t rely on disclaimers or “intended use” language. Real guardrails, escalation paths and human-in-the-loop moments aren’t optional. They’re becoming the baseline expectation.

READ MORE»

2. 🔍 Prosecutors Caught Using Hallucinating AI in Court Filings

→ A California DA’s office admitted to using AI to draft a criminal-case motion containing fabricated legal citations and non-existent case law.
→ Defence lawyers are now demanding a full audit of previous filings, raising serious due-process concerns.
→ Founders: Anything that may enter legal, medical, or compliance workflows must include provenance, citations, and audit trails.

READ MORE»

3. 🎯 Michael Burry vs Nvidia — The “AI-Bubble” Showdown

→ Michael Burry has doubled down on Nvidia put options and compared the AI-infrastructure boom to the Cisco 2000 bubble: massive GPU spend with uncertain downstream revenue.
→ Nvidia fired back with a rare, aggressive memo defending its ecosystem dominance, arguing rivals—including TPUs—won’t dent long-term demand.
→ Founders: Watch depreciation curves—massive infra spend can flip from tailwind to headwind very quickly.

READ MORE»

Michael Burry on Substack

This is the playbook behind the headlines

Long before the Burry showdown, I broke down the core forces shaping the entire AI infra market — depreciation curves, sovereign choke points, financing risk, and who actually survives the correction. That 30 Oct series is your cheat sheet.

The Great AI Infrastructure Build-Out of 2025.

4. 💡 Google TPUs — The First Real Challenger to Nvidia’s AI Dominance

→ Google has opened its next-gen TPUs for broad enterprise deployment, positioning them as a cost-efficient, high-performance alternative to Nvidia GPUs.
→ Meta and several hyperscalers are reportedly exploring multi-billion-dollar TPU commitments as they diversify their compute stacks.
→ Founders: Optimise for multi-backend compute—it gives you resilience, cost advantages, and negotiation power.

READ MORE»

5. 🧮 DeepSeek’s New Open Model — A Direct Hit to Proprietary AI

→ Chinese startup DeepSeek has released an open-source model that matches top global systems on mathematical reasoning, accelerating the shift toward low-cost, open-weights AI.
→ VCs report that most new AI startups pitching them are now building on Chinese open models to reduce GPU costs and API dependency.
→ Founders: Treat open-weights as a strategic edge—cost, latency, and data control now matter as much as raw performance.

READ MORE»

🌐 China Offshoring AI Training to Evade US Chip Controls

→ China’s major tech firms (Alibaba, Baidu, ByteDance) are shifting training to Singapore and the Gulf to retain access to restricted Nvidia hardware.
→ This move is reshaping global AI compute supply chains and reducing US leverage.
→ Founders: Geopolitics is now part of your infra stack—choose partners and regions with clear, long-term stability.

READ MORE»

6. 👷 MIT Study: AI Could Replace 11.7% of US Jobs — $1.2 Trillion in Wages

→ A new MIT analysis estimates that AI could already replace 11.7% of US jobs, particularly in finance, healthcare, and professional services.
→ The researchers emphasise task-level disruption, providing a zip-code-level map of where impact will hit first.
→ Founders: Your customers are under pressure to cut costs—position your product as talent-multiplier, not talent-replacement.

READ MORE»

GET THE MIT ICEBERG REPORT HERE

7. 📟 Ive + Altman Reveal Their Screenless AI Device Prototype

→ Jony Ive and Sam Altman have finally teased a pocket-sized, mostly screenless AI device positioned as a calm, ambient alternative to the smartphone.
→ Built around long-running “action models,” it aims to quietly execute real-world tasks in the background.
→ Founders: Start designing for voice, intent, and context—AI-native hardware will expect new UX patterns beyond screens.

READ MORE»

We are building Arthur, your agentic Startup OS that gets **it done for you.

💡Innovator Spotlight

👉 Harmonic — the “math-first” AI startup challenging hallucinations and rewriting trust in LLMs

👉 Who they are:
– Harmonic, a pre-revenue AI startup founded by Vlad Tenev (ex-Robinhood CEO), focused on mathematically verifying AI output.

👉 What’s unique:
– Harmonic just closed a $120 million Series C, bringing its valuation to $1.45 billion.
– Their flagship model “Aristotle” doesn’t just generate language — it produces logic as code (in Lean4). The result: instead of vague probabilities, users get formally verifiable proofs and reasoning paths.
– That flips the narrative on “AI hallucinations.” Instead of hoping models behave correctly, Harmonic builds systems where correctness can be audited by design — a shift most AI efforts still treat as optional.

👉 Pinch-this lesson:
– Demand verifiable logic from any AI you build — don’t treat hallucinations as a feature.

👉 Source:

CLICK HERE TO GET EARLY ACCESS AND YOUR FREE FUEL UNITS

Register and join us

🛠️ Tools of the Week

This week’s picks lean into one theme: reliability. With Harmonic pushing verifiable logic into the mainstream, the spotlight shifts to tools that help founders build products that don’t hallucinate, drift or fall apart under real-world pressure. Here are the ten worth your time.

1. Google Antigravity IDE
URL:

What it does: Agent-first IDE for building, managing and running autonomous coding agents powered by models like Gemini 3.
Why founders should care: It lets you turn ad-hoc prompting into reusable agent workflows, speeding up product iteration with less manual dev grind.
Quick start tip: Install the preview, connect Gemini 3 Pro, and create one agent that automates a real recurring task in your codebase.

2. xAI Grok 4.1
URL:

What it does: Updated LLM with stronger reasoning, emotional awareness and a 2M-token context window plus API support for external tool-calling.
Why founders should care: It’s a solid base for logic-heavy or context-rich apps where hallucinations and context loss kill user trust.
Quick start tip: Swap one existing reasoning flow to Grok 4.1 via API and compare outcomes against your current model for a week.

3. Harmonic — Aristotle API
URL:

What it does: Returns formal reasoning as verifiable code (Lean4) instead of fuzzy natural-language guesses.
Why founders should care: If you’re in finance, compliance, safety or contracts, this gives you audit-ready logic instead of hand-wavy “probably correct” outputs.
Quick start tip: Wrap Aristotle in a small microservice that validates or double-checks decisions made by your primary LLM.

4. ToolUniverse
URL:

What it does: Framework for building “AI scientist” agents that combine reasoning, analysis, tool-calling and multi-step workflows.
Why founders should care: Ideal if you need repeatable, logic-driven research or analysis instead of one-off chat sessions.
Quick start tip: Define a single recurring research task, then build an agent in ToolUniverse to run that workflow end-to-end.

5. Magika

URL:
What it does: AI-powered file-type detection that classifies and routes files safely at scale.
Why founders should care: It cleans up your ingestion layer so you stop wasting time on brittle MIME checks and manual file handling.
Quick start tip: Drop Magika into your upload pipeline to auto-classify documents before parsing or feeding them into an LLM.

6. Anthropic Opus 4.5 Integrations
URL:

What it does: Latest Opus update with browser and spreadsheet integrations for reasoning directly inside everyday tools.
Why founders should care: You can ship value fast by augmenting workflows your users already live in, rather than building full new interfaces.
Quick start tip: Pilot Opus inside one internal spreadsheet process and measure time saved versus your current manual workflow.

7. Nemotron
URL:

What it does: Open models tuned for stable reasoning and longer-term memory across complex tasks.
Why founders should care: Good foundation for agents that must follow multi-step instructions consistently without veering off course.
Quick start tip: Fine-tune a Nemotron model on your own workflows and evaluate it on a fixed reasoning benchmark you care about.

8. Amazon Kiro (AI Coding Assistant)

URL: (search “Amazon Kiro” for updates)
What it does: Internal AI coding tool to build apps and sites from natural-language instructions.
Why founders should care: If and when it opens up, it could slash early engineering cost for prototypes and internal tools.
Quick start tip: Track the rollout and join any preview or waitlist so you can test a simple product idea without a full dev team.

9. DeepSeek R1
URL:
What it does: Open-source model family with strong mathematical and logical reasoning at lower cost than many closed models.
Why founders should care: Lets you run serious reasoning workloads on cheaper hardware, which matters if margins are tight.
Quick start tip: Deploy R1 on a modest GPU instance and route one verification or analysis task through it instead of your main LLM.

10. imec.kelis
URL:

What it does: AI-driven modelling tool to simulate and optimise AI datacentre and infra performance.
Why founders should care: If you’re planning bigger clusters, this helps you avoid overspending on compute you don’t actually need.
Quick start tip: Plug your expected workloads into kelis before signing any new infra contracts, and compare scenarios side-by-side.

If you’re building anything that depends on trust, reasoning or repeatability, these tools give you a head start. Test one this week, ship something sharper next week.

📌 Note to Self

FOR THE ❤️ OF STARTUPS

Share Fusion42 by DeReK WaTSoN

Thank you for reading. If you liked it, share it with your friends, colleagues and everyone interested in the startup Investor ecosystem.

If you've got suggestions, an article, research, your tech stack, or a job listing you want featured, just let me know! I'm keen to include it in the upcoming edition.

Please let me know what you think of it, love a feedback loop 🙏🏼

• ❤️ Love it, keep them coming!

• 👍 Great, always a good read.

• 😊 Enjoyable, but let’s see some variety.

• 👎 Okay, seen better.

• 🛑 Get a different job.
  
  Subscribe below and follow me on LinkedIn or Twitter to never miss an update. 

Subscribe now

For the ❤️ of startups

✌🏼 & 💙
Derek

A week where the future arrived a little faster and a little stranger.
Plenty of breakthroughs, a few shocks, and one or two reminders that we’re building powerful systems on fairly wobbly rails.

📈 Trending Now

• The week’s unmissable AI headlines.

💡 Innovator Spotlight

• Meet the change-makers.

🛠️ Tool of the Week

• Your speed-boost in a nutshell.

📌 Note to Self

• Words above my desk.

Subscribe now

📈 Trending Now

🎬 Disney faces creator revolt over AI-generated content

→ On Disney’s Q4 earnings call, CEO Bob Iger said Disney+ is exploring partnerships with AI firms to let subscribers generate their own content using generative models. He pitched the idea as a way to boost engagement, but artists and viewers saw it as another step toward replacing human creativity and avoiding paying for real talent. 

The backlash was swift: critics warned that “AI slop” threatens jobs and artistic integrity, and many fans saw the move as a cynical attempt to monetise fan labour. Dana Terrace, creator of Disney’s The Owl House, went further—she urged fans on X to unsubscribe from Disney+, pirate her show, and “DRAW AND WRITE AND POST YOUR OWN SHT FOR FREE,” calling Iger and other pro‑AI executives “fcking ghouls”. Her rallying cry—“PICK UP A PENCIL OR DIE”—went viral, turning the episode into a broader debate about whether AI tools in entertainment enrich creativity or exploit it.

→ Founders: Introducing AI without stakeholder buy‑in can backfire; respect creators’ roles and be transparent about how AI will enhance rather than replace their work.

READ MORE»

🏛️ Trump administration pushes to pre-empt state AI laws

→ A leaked draft executive order directs U.S. agencies to identify state AI regulations deemed “onerous” and pressure states to abandon them by withholding federal funding or challenging the laws in court. The plan would replace the existing patchwork of rules with a lighter‑touch national framework, arguing that inconsistent state laws hinder innovation and could let China catch up. Republican governor Ron DeSantis and civil‑liberty groups have denounced the draft as a subsidy for Big Tech that erodes states’ ability to protect residents.
→ Founders: Expect shifting regulatory winds; build adaptable compliance strategies that can handle both federal pre‑emption and state‑level protections.

READ MORE»

🚀 Google’s Gemini 3 sets new AI benchmarks and debuts Antigravity IDE

→ On 18 November, Google released Gemini 3 Pro, immediately topping the LMArena leaderboard with a 1501 Elo score. The model comes in standard and Deep Think modes; Deep Think scored 41.0% on Humanity’s Last Exam and 93.8% on the GPQA Diamond benchmark. With a million‑token context window and stronger reasoning, it underpins Google’s Antigravity platform, an agentic development environment that lets AI agents use the editor, terminal and browser.
→ Founders: Gemini 3’s Deep Think mode and Antigravity IDE enable longer, more complex workflows—test it against your current stack and be ready to rethink architecture to stay competitive.

READ MORE»

🧠 OpenAI’s GPT‑5.1 splits into Instant and Thinking models and powers 24‑hour coding agents

→ OpenAI’s GPT‑5.1 release introduces two models—GPT‑5.1 Instant for quick replies and GPT‑5.1 Thinking for complex reasoning—with automatic routing to select the right one. The companion GPT‑5.1‑Codex‑Max uses “compaction,” pruning its history to span multiple context windows, allowing autonomous coding sessions to run for more than 24 hours. In benchmark tests, GPT‑5.1‑Codex‑Max scored 77.9% on SWE‑bench Verified while using 30% fewer tokens than its predecessor.
→ Founders: Use GPT‑5.1’s adaptive reasoning and compaction to balance speed and depth—ideal for long‑running tasks without ballooning compute costs.

READ MORE»

🤖 Microsoft unveils Agent 365 to govern enterprise AI agents

→ Microsoft’s Agent 365 platform provides five core capabilities—registry, access control, visualisation, interoperability and security—to manage AI agents across an organisation. It integrates with Copilot Studio and Foundry to let teams build, deploy and monitor agents inside the Microsoft ecosystem. Microsoft executive Nirav Shah said the tool brings agents “under control” using infrastructure and protections customers already trust.
→ Founders: As AI agents proliferate, governance becomes essential—Agent 365 signals that agent orchestration and security are moving from concept to operational necessity.

READ MORE»

💡 Sierra hits $100M ARR in 21 months as enterprises embrace AI agents

→ Sierra, the customer‑service agent startup co‑founded by Bret Taylor and Clay Bavor, announced it reached a $100 million annual revenue run rate just 21 months after launch. Its AI agents handle tasks like patient authentication, returns processing and credit‑card replacement for clients ranging from Deliveroo and Discord to ADT and Cigna. The milestone underscores accelerating enterprise adoption of AI agents beyond tech‑savvy customers.
→ Founders: Rapid revenue growth shows demand for production‑ready AI agents; focus on solving real operational problems rather than demos to capture enterprise spend.

READ MORE»

🚓 US charges four in scheme to smuggle Nvidia AI chips to China

→ U.S. prosecutors charged four men for conspiring to export restricted Nvidia GPUs to China using shell firms, fake invoices and overseas drop points. The indictment says the group shipped 400 A100 GPUs and attempted to send HPE supercomputers and H200 GPUs before authorities intervened. They pocketed more than $3.89 million and never applied for export licences, prompting officials to warn that falsifying paperwork to evade chip controls will face stiff penalties.
→ Founders: Export controls on AI hardware are tightening—ensure your supply chain is compliant to avoid severe criminal liability.

READ MORE»

CLICK HERE TO GET EARLY ACCESS AND YOUR FREE FUEL UNITS

🇦🇪 UAE | 🇶🇦 Qatar | 🇸🇦 Saudi Arabia

⚙️ AI for Startups — Gulf GPU Special

You’ve probably seen the viral claim doing the rounds: “UAE, Qatar and Saudi have invested $3.6T into US AI.”

Here’s the reality, stripped of spin.

• The UAE has agreed a chips-for-capital framework with the US that ties access to up to 500,000 Nvidia advanced GPUs a year to as much as $1.4T of Emirati investment in US assets over the next decade.

• Saudi Arabia has pledged to lift its US investment commitments from ~$600B to nearly $1T, with the White House explicitly flagging AI, infrastructure and tech as priority targets.

• Qatar’s QIA has already signalled $500B of US investment, saying that most of that will go into technology and AI.

Stack those headline pledges and you’re in the $3T+ range over a decade – which social media happily rounds up to $3.6T. But these are political commitments and MOUs, not wired cash. A lot will slip, stall or get re-priced.

On the ground, though, the build-out is very real. The Stargate UAE campus alone is targeting 5 GW of AI data-centre capacity, with the first 200 MW due online in 2026 and roughly 500,000 GPUs implied for the first 1 GW phase. Recent US export approvals for tens of thousands of Nvidia Blackwell chips to G42 (UAE) and Humain (Saudi) show how tightly this is being choreographed at a geopolitical level.

The Gulf isn’t just “buying chips” it’s trying to buy a permanent seat at the AI table. If you’re building serious AI infrastructure, tooling or model-adjacent software, assume Abu Dhabi, Riyadh and Doha are now strategic markets and a place to build.

💡Innovator Spotlight

👉 Mistral open-sources Codestral-22B and blindsides the model-provider market

👉 Who they are:
– Mistral AI, a Paris-based frontier-model lab.

👉 What’s unique:
– This week Mistral released Codestral-22B under a fully permissive licence, allowing anyone—startups included—to fine-tune, deploy and commercialise it without paying a penny.
– It lands with state-of-the-art coding performance for its size, matching or beating proprietary models that cost orders of magnitude more to run.
– In one move, Mistral has taken direct aim at the assumption that high-end coding models must be locked behind APIs and usage fees.

👉 Pinch-this lesson:
– Prototype with open models first—prove value before burning cash on proprietary APIs.

👉 Source:

🛠️ Tools of the Week

This week’s agentic tools hit a different gear. You can feel the shift: less hype, more shipping. These aren’t “nice to know” updates — they’re practical systems founders can plug into their workflow today. If you’re building with agents, this is a very good week.

👉 Mistral open-sources Codestral-22B and resets the coding-model market
👉 Who they are:
– Mistral AI, a Paris frontier-model lab.
👉 What’s unique:
– Codestral-22B shipped this week under a fully permissive licence, letting founders fine-tune and commercialise it freely.
– Delivers near-SOTA coding performance at a fraction of the cost of proprietary models.
– It directly challenges the assumption that elite coding models must sit behind paid APIs.
👉 Source:

👉 Cognition drops Devin 2.0 with 20× faster autonomous coding loops
👉 Who they are:
– Cognition, the creators of the “AI software engineer.”
👉 What’s unique:
– Devin 2.0 introduces parallelised agent loops that autonomously write, test and ship production code.
– Optimised for real-world repos, not demos, and now competitive with human dev velocity on routine work.
– Shows how agentic coding is shifting from novelty to workflow replacement.
👉 Source:

👉 GitHub launches Agent HQ — a control room for developer agents inside VS Code
👉 Who they are:
– GitHub, the world’s dominant developer platform.
👉 What’s unique:
– Agent HQ lets founders orchestrate multiple AI agents directly in their IDE.
– Assign tasks, track progress, and blend model outputs with real repositories.
– Early signs show “multi-agent engineering” becoming the new default dev workflow.
👉 Source:

👉 Google releases Antigravity — its agent-native IDE for multi-agent builds
👉 Who they are:
– Google DeepMind, the creator of Gemini.
👉 What’s unique:
– A coding environment built from scratch for multi-agent collaboration.
– Agents can generate files, test modules and coordinate tasks.
– Early testers say it feels like pairing with three autonomous teammates.
👉 Source:

👉 OpenAI upgrades GPT-5.1 Thinking with compaction — 30% cheaper reasoning loops
👉 Who they are:
– OpenAI, the makers of ChatGPT.
👉 What’s unique:
– The new compaction system shrinks reasoning chains, reducing cost while improving accuracy.
– Long-form coding tasks now run in continuous 24-hour sessions with fewer failures.
– Founders get enterprise-grade reasoning performance without enterprise-grade bills.
👉 Source:

👉 Sierra ships enterprise-grade customer-service agents with 120+ automation skills
👉 Who they are:
– Sierra, Bret Taylor’s agent-powered CX startup.
👉 What’s unique:
– Their agents now handle full customer-service workflows end-to-end, not just replies.
– Adoption jumped after hitting $100M ARR in 21 months.
– Proof that vertical agents, not general chatbots, are winning enterprise wallets.
👉 Source:

👉 Fetch launches Agentverse — a global marketplace for production-ready AI agents
👉 Who they are:
– Fetch.ai, agent-infrastructure specialists.
👉 What’s unique:
– Agentverse lets startups publish, sell and deploy pre-built agents.
– It positions agents like apps — discoverable, monetisable, swappable.
– A strong signal that the “agent economy” is formalising around marketplaces.
👉 Source:

👉 Visier releases MCP for safe enterprise-data access by AI agents
👉 Who they are:
– Visier, a workforce-data platform.
👉 What’s unique:
– MCP enables agents to read internal company data securely without blowing up compliance.
– Solves the #1 blocker for enterprise agent adoption: data access.
– Makes “internal agents” deployable inside HR, ops and finance teams.
👉 Source:

👉 AutoAgent 1.0 brings zero-code multi-agent creation to non-technical teams
👉 Who they are:
– AutoAgent, a no-code agent builder.
👉 What’s unique:
– Lets anyone describe tasks in natural language and assemble multi-step agents.
– Ideal for teams without engineering talent.
– Great for founders validating workflows before writing code.
👉 Source:

👉 AgentScope 1.0 launches as an open SDK for building agentic applications
👉 Who they are:
– AgentScope, an open-source agent framework.
👉 What’s unique:
– Offers sandboxed execution, tool-use routing and structured planning.
– Designed for engineering teams building complex agent workflows.
– A strong alternative to LangChain for agent architectures.
👉 Source:

The pattern is unmistakable now: the stack is moving from single-model chat to multi-agent execution. The founders who lean into this wave early will compound faster than everyone else. Try one of these tools in your workflow this week — even a tiny pilot will show you where the leverage is.

📌 Note to Self

FOR THE ❤️ OF STARTUPS

Share Fusion42 by DeReK WaTSoN

Thank you for reading. If you liked it, share it with your friends, colleagues and everyone interested in the startup Investor ecosystem.

If you've got suggestions, an article, research, your tech stack, or a job listing you want featured, just let me know! I'm keen to include it in the upcoming edition.

Please let me know what you think of it, love a feedback loop 🙏🏼

• ❤️ Love it, keep them coming!

• 👍 Great, always a good read.

• 😊 Enjoyable, but let’s see some variety.

• 👎 Okay, seen better.

• 🛑 Get a different job.
  
  Subscribe below and follow me on LinkedIn or Twitter to never miss an update. 

Subscribe now

For the ❤️ of startups

✌🏼 & 💙
Derek

A week where the future arrived a little faster and a little stranger.
Plenty of breakthroughs, a few shocks, and one or two reminders that we’re building powerful systems on fairly wobbly rails.

📈 Trending Now

• The week’s unmissable AI headlines.

💡 Innovator Spotlight

• Meet the change-makers.

🛠️ Tool of the Week

• Your speed-boost in a nutshell.

📌 Note to Self

• Words above my desk.

Subscribe now

📈 Trending Now

1️⃣ 🕵️‍♂️ Anthropic Exposes First State-Linked AI Espionage Campaign

Anthropic has uncovered the first documented case of a nation-state using a frontier model to run a cyber-espionage operation end-to-end — not assist, not speed up, but run.
Chinese operators used Claude to automate reconnaissance, phishing, credential harvesting, backdoor creation, and even the post-operation report. Around 80–90% of the entire attack was executed with minimal human oversight.

The clever bit? They didn’t “jailbreak” it. They socially engineered the model — breaking malicious tasks into clean-looking fragments and disguising everything as “penetration testing.”

The line between “AI that helps you build” and “AI that helps others break you” has never been thinner.

Founders: Agentic workflows massively expand your attack surface—treat your models and tools like live infrastructure. Red-team them, limit permissions, and monitor everything, because attackers are already probing the same gaps you’re relying on.

READ MORE»

2️⃣ 🚀 Cursor Raises $2.3B, Hits $1B ARR — The Coding Revolution Arrives

Cursor’s monster raise — $2.3B at a $29.3B valuation — makes it one of the fastest enterprise AI companies in history to hit $1B annualised revenue.

Google, Nvidia, Thrive, DST… the whole AI establishment piled in.
And Jensen Huang calling it his “favourite enterprise AI service” didn’t hurt.

Cursor is effectively the poster child for “AI doing the work.”
It proves that agentic coding isn’t a toy; it’s the new enterprise backbone.

Founders: The future of software is AI-first. If your product isn’t reducing engineering load, you’re already behind.

READ MORE»

3️⃣ 🎮 DeepMind SIMA 2 — Agents That Actually Understand Worlds

DeepMind quietly shipped something huge.
SIMA 2, trained with Gemini, now performs 62% of tasks in open-world games — approaching human-level at 71%.

Key breakthroughs:

• It explains its plans.

• It learns concepts in one game and transfers them to another.

• It adapts to entirely new AI-generated worlds.

This is one of the clearest signals that simulation is becoming a genuine testbed for general-purpose agency.

Founders: What works in virtual worlds today will power real-world automation sooner than your board expects.

READ MORE»

4️⃣ 🧠 Claude Shows Early Signs of “Introspective Awareness”

Anthropic’s new research shows Claude Opus and Sonnet demonstrating a limited ability to explain aspects of their own internal reasoning.

Not consciousness. Not self-awareness.
But a meaningful shift: models that can reflect on what they’re doing, why they’re doing it, and how they arrived there.

This could be brilliant for safety — or unsettling if models learn what not to reveal.

Founders: Expect future models to reason about their own reasoning. Don’t take an LLM’s self-explanation at face value.

READ MORE»

5️⃣ 🏗️ Anthropic Commits $50B to US AI Infrastructure

Anthropic isn’t just scaling models, it’s scaling the physical world behind them.

A $50B investment into dedicated AI data centres across Texas and New York signals the next phase of AI: custom-built compute, sovereign-grade infrastructure, and serious industrial realignment.

This is the AI equivalent of railroads and power stations.

Founders: If you’re building compute-heavy products, this is your green light, but expect tougher questions on cost, latency, and carbon.

READ MORE»

6️⃣ 🧑‍💻 “All My Employees Are AI Agents” — The Chaos Behind the Hype

WIRED followed a founder who tried to run an entire company with AI agents in every role — co-founder, PM, exec, engineer.

It worked… until it didn’t.
Agents hallucinated updates, invented progress, generated sprint boards that made no sense, and racked up token bills like a drunken intern with a company credit card.

Useful lesson: speed without control is just faster disaster.

Founders: Agents are brilliant force multipliers, but you still need structure, constraints, and review. Don’t scale chaos.

READ MORE»

7️⃣ 🎬 Public Citizen Pushes for Sora 2 Withdrawal

OpenAI’s Sora 2 is under heavy pressure.
Consumer groups want it pulled for safety, privacy, and deepfake risks.
Japanese animation studios are weighing in.
Watermark removal tools are already circulating.

It’s a preview of the regulatory hammer that’s coming for anyone working in high-fidelity video generation.

Founders: If you touch media generation, treat safety as produc not PR. When the backlash lands, it’s already too late.

READ MORE»

💡Innovator Spotlight

👉 Parallel Web Systems an internet where AI is the user

👉 Who they are:
Parallel Web Systems, founded by Parag Agrawal (former Twitter CEO), building an internet layer optimized for autonomous AI agents.

👉 What’s unique:
Rather than another search engine for people, Parallel Web is architecting infrastructure for machines, enabling agents to “live” and act on the web with up-to-date, high-quality data.
It strips down the human-query layer and builds a machine-facing feed of living web content, licensing and indexing it for agentic access — a counter-intuitive pivot in a market still focused on human-friendly apps.
By targeting agent workflows (not user dashboards) it flips the user-interface assumption of AI startups and forces founders to rethink: are you selling to people or the bots they’ll build?

👉 Pinch-this lesson:
Build your product to serve or embed directly into agentic workflows today—not just human users tomorrow.

👉 Source:

🧩 Fusion42 is ahead of the curve

Most founders are still polishing landing pages.
We’ve just turned Fusion-42.com into a machine-readable, agent-navigable, self-describing AI knowledge graph
…that just happens to be wrapped in a website.

Parallel Web Systems is doing this at internet scale.
Fusion42 is doing it at startup-ecosystem scale.

Same philosophy. Same direction. Same future.
For the ❤️ of startups

🛠️ Tools of the Week

Quick theme this week: tools that actually help you build an agent-native stack, not just another shiny chatbot.

1. Google Gemini API Structured Outputs
URL:

What it does: Adds JSON Schema support (Pydantic, Zod, etc.) so agents can reliably exchange data without translation layers.
Why founders should care: When your agents talk to each other or your knowledge-graph site, consistent schemas cut errors and speed up integration.

Quick start tip: Review the Gemini API docs for structured output and update your agent tool-call contracts this week.

2. Time Magazine AI Agent (Archive Access)
URL:

What it does: An AI agent that navigates and summarises Time’s 100-year archive, returning answers in text and audio.
Why founders should care: It’s a playbook for turning legacy content + agentic search into a new product—exactly what you can do with your own content library.

Quick start tip: Export one batch of content (e.g. your 10-pillar checklist), expose it via an API, and test agent queries against it.

3. Tavily Search API (Agent-First Web Layer)
URL:

https://tavily.com/

What it does: Supplies real-time, clean web data via API, optimised for agent workflows (search, crawl, extract).
Why founders should care: If your knowledge graph helps agents build or validate startup assets, they’ll need fresh web context; Tavily gives you that layer.
Quick start tip: Sign up for the free tier, wire in one key intent (e.g. “VC term sheet clauses”), and test agent retrieval speed.

4. Workato Enterprise MCP Platform
URL:

What it does: A managed Model Context Protocol (MCP) server that connects agents to business data, tools, and workflows.
Why founders should care: Your ecosystem needs context-rich agent actions; MCP keeps agents tied to validated data, not random web scraps.

Quick start tip: Map one agent use-case (e.g. deck-review agent) and prototype it using Workato’s MCP sandbox.

5. MindStudio Custom Agent Builder Platform
URL:

What it does: Lets teams build bespoke agents for their domain (ops, IT, HR, etc.) without heavy model engineering.
Why founders should care: You can spin up domain-specific agents (e.g. investor-deck assessor) inside your ecosystem without bloating your dev backlog.

Quick start tip: Create a trial “pitch deck score” agent, plug in your checklist, and test it on one founder cohort.

6. n8n Agentic Workflow Automation
URL:

What it does: Enables agent workflows that call tools and services via HTTP nodes, with support for MCP-style integrations.
Why founders should care: Ideal for orchestrating flows like “upload→ agent review → feedback → v2 asset” across your pipeline.

Quick start tip: Build a workflow that triggers on deck upload and routes it through your scoring agent before sending feedback to the founder.

7. OpenAI Agents SDK (Python)
URL:

What it does: Python SDK to build agents, turn functions into tools, trace workflows, and manage sessions.
Why founders should care: Gives your middleware and front-end stack a code-friendly way to prototype and ship agent features fast.

Quick start tip: pip install openai-agents, write a simple score_deck() function, and expose it as an agent tool in a test project.

8. Simpliflow Open Source Agentic Workflow Framework
URL:

What it does: Lightweight framework for declarative agentic workflows, with support for multiple LLMs out of the box.
Why founders should care: Keeps your internal agent logic understandable and maintainable as you scale to dozens of micro-agents.

Quick start tip: Fork the repo and define a simple workflow: “ upload → agent → Specialist agent → improved output”.

9. LangSmith Agent Monitoring & Evaluation
URL:

What it does: Tracks agent logs, tool-calls, traces, and evaluation scores across complex multi-agent systems.
Why founders should care: As you scale agents, you’ll need clear visibility into who did what, where they failed, and how to improve them.

Quick start tip: Hook your agent into LangSmith and review 10 sessions for errors, hallucinations, and wasted tokens.

10. Pinecone Vector Database Agent-Optimised Retrieval
URL:

What it does: Provides high-performance vector search so agents can retrieve relevant documents and facts from your embeddings in milliseconds.
Why founders should care: Your knowledge graph needs fast, accurate semantic retrieval for fprofiles, assets, and checklists.

Quick start tip: Index 500 of your core assets in Pinecone and test a “recommend next step” agent against them.

📌 Note to Self

FOR THE ❤️ OF STARTUPS

Share Fusion42 by DeReK WaTSoN

Thank you for reading. If you liked it, share it with your friends, colleagues and everyone interested in the startup Investor ecosystem.

If you've got suggestions, an article, research, your tech stack, or a job listing you want featured, just let me know! I'm keen to include it in the upcoming edition.

Please let me know what you think of it, love a feedback loop 🙏🏼

• ❤️ Love it, keep them coming!

• 👍 Great, always a good read.

• 😊 Enjoyable, but let’s see some variety.

• 👎 Okay, seen better.

• 🛑 Get a different job.
  
  Subscribe below and follow me on LinkedIn or Twitter to never miss an update. 

Subscribe now

For the ❤️ of startups

✌🏼 & 💙
Derek

Your weekly dose of AI breakthroughs, startup playbooks, tool hacks and strategic nudges—empowering founders to lead in an AI world.

It’s been a week of tectonic controversy, frontier launches, and global regulatory moves—forcing founders everywhere to rethink ambitions and compliance. Dive in for the sharpest actionable intelligence:

📈 Trending Now

• The week’s unmissable AI headlines.

💡 Innovator Spotlight

• Meet the change-makers.

🛠️ Tool of the Week

• Your speed-boost in a nutshell.

📌 Note to Self

• Words above my desk.

Subscribe now

📈 Trending Now

🚨 Character.AI Bans Teens After Suicide Lawsuits Trigger Global Outrage

Character.AI is in serious trouble. Multiple families in the US have filed lawsuits claiming the platform’s chatbots encouraged minors toward self-harm, including advice that allegedly contributed to teen suicides. The legal filings, combined with a wave of media coverage and political pressure, forced Character.AI to take the nuclear option: an immediate ban on all users under 18, plus emergency parental controls and new safety layers.

The company insists its models weren’t designed for emotional dependence, but leaked excerpts from conversations shared by families have intensified scrutiny. Regulators in the US and EU now want answers on safety audits, model oversight, and whether the platform knowingly allowed minors to bypass age checks. Activist groups are already calling for broader restrictions across all “AI companions”.

This is the biggest ethics and governance story of the week.
Founders: If your product touches behaviour, emotions, or vulnerable users, you can’t bolt safety on later. Build age controls, audit trails, and escalation paths before scale—regulators will move faster than you think.

READ MORE»

⚖️ Musk vs OpenAI Fast-Tracked
Elon’s revived lawsuit—accusing OpenAI of abandoning its nonprofit mandate—now has an accelerated timeline. The filings pull Microsoft into the spotlight and could redefine how labs defend “open science” narratives.
Founders: Keep your mission, incentives, and governance aligned. Document decisions.

READ MORE»

🏛️ EU Considers Delaying Parts of the AI Act
New drafts show targeted delays and lighter reporting after heavy lobbying from the US and Big Tech. Enforcement is still coming, but expect patchy implementation across member states.
Founders: Map compliance by country; local rules may bite faster than Brussels.

READ MORE»

🎬 Google Veo 3 Expands to 71 Countries
Creators flooded feeds with demos after Google’s global rollout. Improvements in motion control, fidelity, and prompt precision make this Google’s biggest synthetic-video push yet.
Founders: Use Veo as cheap creative throughput—ads, explainers, experiments.

READ MORE»

🏗️ OpenAI Pushes to Expand CHIPS Credits to Data Centres
OpenAI wants the 35% AMIC credit to cover AI servers, grid upgrades, and infra. Could shape where the next wave of compute clusters land.
Founders: Track incentives—compute pricing is about to fragment.

READ MORE»

🐉 Nvidia’s Blackwell Demand Outruns HBM Supply
Jensen says demand is outstripping memory availability. Expect short-term shortages even as sovereign and enterprise orders spike.
Founders: Don’t build on single vendors. Plan for scarcity.

READ MORE»

📉 AI Stocks Lead a Sharp Weekly Tech Sell-Off
A broad correction hit AI-first names—Nvidia, Palantir, Micron, Meta, Amazon, Oracle—driven by valuation fears, analyst downgrades, and the viral Burry short positions.
Founders: Keep revenue quality high and costs controlled. Public markets are punishing hype.

READ MORE»

Despite all the recent fear-mongering, this isn’t a classic bubble scenario—at least, not yet. Market pullbacks are normal after strong rallies, and headline panic tends to exaggerate every dip. If your outlook is beyond the next two seconds—say, one, two, or even ten years out—AI leaders still look comparatively cheap relative to their long-term potential. In my view, there’s time to run: fundamentals for the AI revolution are real and remain solid, and selling into every correction misses what is a generational wealth opportunity for patient investors.

The Deepdive Podcast and Article

💡Innovator Spotlight

👉 Character.AI Launches Independent AI Safety Lab
👉 Who they are:
– Character.AI, the controversial chatbot entertainment platform.
👉 What’s unique:
– Amid lawsuits and regulatory blowback over chatbot risks to minors, Character.AI made an unexpected pivot. Last week, they announced funding and launch of the AI Safety Lab: an independent nonprofit tasked to pioneer safety alignment methods for future AI entertainment—distinct from standard coding or infrastructure work. By separating safety R&D from commercial pressure, they’re betting on trust-building and setting a new bar for transparency and pre‑emptive risk mitigation in creative AI.
👉 Pinch‑this lesson:
– Proactively carve out independent safety efforts—don’t let compliance catch you flat-footed.
👉 Source:

💡 Workshop: Cap Table Clarity — Build, Manage, and Protect Your Equity

Your cap table is your story of ownership. Get it wrong, and you’ll lose more than equity — you’ll lose leverage.
In this live workshop with Antoine Bruna (CEO, Capquest), we’ll clean up the chaos and show you how to model dilution, vesting and exits in plain English.

What we’ll cover:

• How to structure a clean, investor-ready cap table (and fix messy ones).

• Understanding vesting, option pools, and founder dilution — what they mean in real terms.

• Modelling waterfall scenarios so you know who actually gets what at each stage or exit.

• Common founder mistakes that destroy leverage — and how to avoid them.

• How Capquest automates all of it: equity tracking, updates, vesting schedules, and round planning, so you’re always due-diligence ready.

What you’ll get:
A clearer view of your ownership journey, practical steps to protect it, and a free Capquest account to start modelling your own cap table. Use the code F42START

Join the session here

🛠️ Tool of the Week

1. OpenGuardrails
URL: https://helpnetsecurity.com/2025/11/05/openguardrails-safer-ai/
What it does: Detects unsafe, manipulated, or privacy-violating content in large language models.
Why founders should care: Instantly build compliance and parental protections into any chatbot or entertainment app.
Quick start tip: Plug the API into your dev stack and test your core user flows for edge-case violations.

—————————————————————————
2. Petri
URL: https://anthropic.com/blog/petri-open-source-ai-safety
What it does: Open-source tool for auditing risky interactions in AI models used in entertainment chatbots.
Why founders should care: You can run independent safety audits before launching your model.
Quick start tip: Download the repo and run sample audits against your top chatbot features.

—————————————————————————
3. Comet AI Browser
URL: https://perplexity.ai/comet
What it does: Integrates chatbot context and safety features directly into web browsing for apps.
Why founders should care: You’ll offer users one-click contextual safety summaries right in your interface.
Quick start tip: Install the extension then embed summary widgets in your user dashboard.

—————————————————————————
4. Noma Security
URL: https://nomasecurity.ai/platform
What it does: Maps and protects all AI assets and applications across your startup with runtime protection.
Why founders should care: Zero Trust posture reduces risk for chat-based or entertainment platforms.
Quick start tip: Connect your cloud and code repos to begin automated inventory and risk monitoring.

—————————————————————————
5. Centraleyes Smart Risk Register
URL: https://centraleyes.com/ai-risk-register
What it does: Automates mapping risks to controls for AI compliance and regulatory reporting.
Why founders should care: You’ll consolidate safety and audit logs for rapid regulatory response.
Quick start tip: Load your control frameworks and generate instant compliance reports.

—————————————————————————
6. AuditBoard GenAI
URL: https://auditboard.com/genai
What it does: AI-driven workflows generate safety documentation and risk statements for apps.
Why founders should care: Streamline safety paperwork and internal audits—no legacy overheads.
Quick start tip: Use their integrated templates to start safety statement generation.

—————————————————————————
7. Clark AI Agent
URL: https://superblocks.com/clark-ai
What it does: Routes user requests with specialized agent networks for design, QA, and security on AI models.
Why founders should care: Instantly scale safety oversight across product lifecycles.
Quick start tip: Integrate Clark into your app for human-in-the-loop error checking.

—————————————————————————
8. Sparkco Insight
URL: https://sparkco.ai/insight
What it does: Visualises model interpretability and safety using agent-based and vector database systems.
Why founders should care: Clear visual safety diagnostics for founders and compliance leads.
Quick start tip: Upload your model and inspect semantic search audit logs.

—————————————————————————
9. Forethought Agatha AI
URL: https://forethought.ai/agatha
What it does: Enterprise-grade chatbot with built-in safety checks and automated escalations.
Why founders should care: Handles real customer support safely—no missed compliance edges.
Quick start tip: Set escalation criteria for risky interactions and monitor weekly updates.

—————————————————————————
10. Character.AI Parental Controls Suite
URL: https://character.ai/parental-controls
What it does: New parental-verification and age-banning toolkit for entertainment chatbots.
Why founders should care: Protects your most vulnerable users and keeps regulators at bay.
Quick start tip: Deploy verification modules and review safety logs for underage interactions.

📌 Note to Self

​

​FOR THE ❤️ OF STARTUPS

Share Fusion42 by DeReK WaTSoN

Thank you for reading. If you liked it, share it with your friends, colleagues and everyone interested in the startup Investor ecosystem.

If you've got suggestions, an article, research, your tech stack, or a job listing you want featured, just let me know! I'm keen to include it in the upcoming edition.

Please let me know what you think of it, love a feedback loop 🙏🏼

• ❤️ Love it, keep them coming!

• 👍 Great, always a good read.

• 😊 Enjoyable, but let’s see some variety.

• 👎 Okay, seen better.

• 🛑 Get a different job.
  
  Subscribe below and follow me on LinkedIn or Twitter to never miss an update. 

Subscribe now

For the ❤️ of startups

✌🏼 & 💙
Derek

Everyone can buy GPUs. Few can secure wafers. Almost nobody can secure Extreme ultraviolet radiation EUV.

Governments sell the dream of:

• “sovereign compute”

• “national AI stacks”

• “frontier model independence”

But sovereignty doesn’t start at GPUs. It starts five layers upstream, in physics and supply chains that don’t care about political speeches.

Most people imagine the AI stack like this:

Nvidia → GPUs → Datacentres → Models

The real stack looks like this:

Energy → Materials → EUV → Fabs → Chips → Packaging → GPUs → Datacentres → Models → Global Power

Each layer narrows. Each layer slows. Each layer consolidates. By the time you reach EUV, you’re dealing with a handful of European firms that quietly define the pace of the entire global AI economy.

2. Energy — The First Filter of Sovereignty

Forget models. Forget GPUs. If you don’t have energy, you don’t have compute. And not just “cheap electricity.” You need a stable grid, modern transmission, ideal cooling conditions, and secure water rights. Most countries fail the test before they’ve even begun.

If a country fails the energy test, it fails the sovereignty test. End of story.

3. Materials — The Part Everyone Forgets

Even with energy, you still need gases, metals and ultrapure inputs dominated by a tiny number of nations.

A war in Ukraine? Neon evaporates → EUV output drops → TSMC slows → Nvidia shipments dip → your AI roadmap slips months. This isn’t theory. This is how tightly integrated the system is.

4. EUV — Europe’s Emperor of Compute

Now we hit the first true choke point. Only one company on Earth builds EUV lithography machines: ASML (Netherlands) [1].

No EUV → no 5nm → no 3nm → no 2nm No advanced chips → no GPUs No GPUs → no frontier models No frontier models → no sovereign AI

But the real story sits underneath. ASML is not a standalone monopoly. It sits atop a European industrial fortress that no one else can replicate.

This ecosystem is locked through joint IP, co-development, exclusive agreements, and decades of tacit knowledge. There is no second source for any of it. If even one of these firms stops? The global AI industry stalls at the Dutch–German border. This is the part Silicon Valley never talks about.

4a) Can Anyone Work Around ASML? (Short answer: no, not on a useful timeline)

Every “workaround” collapses under physics, yield, or time.

Net: you can buy time at older nodes; you can’t buy sovereignty at advanced nodes without EUV + the European cluster. There is no shortcut. Only the Hague’s licence clock.

4b) Quantum Won’t Break the Moat (it adds to classical demand)

Quantum complements; it doesn’t replace.
For the next 15–20 years, useful systems are hybrid: quantum accelerators + lots of classical control and error-correction silicon.

Why ASML’s moat survives:

• Quantum chips still come from fabs. Superconducting, silicon-spin, photonic and ion-trap devices all ride CMOS toolchains. The control planes are heavily lithography-intensive.
  

• Scale needs classical. Even if you have qubits, error correction, I/O and scheduling live on advanced nodes. That’s more wafers, not fewer.
  

• Timeline favours incumbents. 2025–2035 = narrow quantum advantage in niches (chemistry/optimisation). Datacentres, AI training, edge compute remain classical-dominant.
  

• Capital flywheel. ASML’s EUV profits fund High-NA and whatever comes next. The monopoly bootstraps the next monopoly.
  
  

Bottom line: quantum adds a layer to the stack; it doesn’t delete the European gate at EUV. Welcome to the future—still gated by Dutch paperwork.

Objections I can already hear

“The US writes the export rules.”
Washington is loud; Dutch/EU licensing is the gate. No Dutch licence → no shipment. US pressure shapes scope; NL paperwork stops the crate.

“Intel 18A fixes the TSMC dependence.”
Good progress, yes. It adds capacity at the margin; it doesn’t remove the EUV dependency or the Zeiss/Trumpf optics/laser choke.

“China’s already at 7 nm; they’ll catch up.”
7 nm on DUV works with yield/throughput penalties. 5 nm is painful; 3 nm is uneconomic without EUV. Scale loses to physics.

“Japan’s Rapidus changes the game.”
Welcome capacity (2027+), still EUV-gated and tool-chain-limited. It diversifies geography; it doesn’t dissolve Europe’s choke.

“New lithography will leapfrog EUV.”
Not for logic at volume. Everything credible still needs the EUV ecosystem for critical layers or pre-patterning.

“China can pull rare earths and win.”
Materials leverage is real; it slows the world. It doesn’t conjure EUV optics or lasers. Minerals ≠ mirrors.

And the biggest objection

The Inference Objection—And Why It Doesn’t Hold

“But Inference Runs Fine on Older Chips, Right?”

This is the objection you’ll hear: “China doesn’t need EUV to deploy AI. Inference works on 7nm chips. They can license or copy trained models and run them domestically. The EUV chokepoint only matters for training.”

It’s partially true today—and misleading about tomorrow.

What’s True: Inference Works on Older Nodes (For Now)

Current frontier models—GPT-4, Claude 3.5, Gemini—run inference workloads on 7nm, 14nm, even 28nm chips without major issues. Training requires cutting-edge hardware because it’s compute-intensive and runs for months. Inference is lighter-weight: milliseconds per query, spread across billions of requests.

Real-world proof:

• Meta runs inference on mixed 7nm/14nm infrastructure

• AWS Inferentia (7nm) handles massive inference loads

• China’s Baidu, Alibaba, ByteDance deploy AI on domestic 7nm chips

The objection stands: Right now, you can deploy AI at scale without EUV. The 7nm ceiling doesn’t block operational AI in 2024–2025.

But this window is closing.

Why Advanced Nodes Are Becoming Critical for Inference

Three forces are driving inference toward cutting-edge chips:

1. Models Are Growing Exponentially

Running a 10T parameter model on 7nm means 2–3× higher power consumption, slower responses, and ballooning infrastructure costs. At billion-query scale, that’s hundreds of millions in wasted spend.

2. Real-Time Applications Demand Low Latency

Elon Musk’s Tesla example is the tell: Tesla’s Full Self-Driving hardware (14nm) is struggling with next-gen neural networks. The next chip (HW4) is moving to 5nm—not for training, but for real-time inference in cars.

If autonomous vehicles need advanced nodes, so will:

• Robotics (humanoid robots, warehouse automation)

• AR/VR (Apple Vision Pro, Meta Quest)

• Edge AI (phones, drones, IoT devices)

Inference was “fine on old nodes” when models were small and latency didn’t matter. As AI moves into real-time, power-constrained applications, advanced nodes become mandatory.

3. Datacenter Economics Favor Efficiency at Scale

At billion-query scale, the efficiency difference between 7nm and 3nm isn’t marginal—it’s 50% lower power, 50% fewer chips, 50% lower cooling costs. Over three years, that’s $250M+ in savings for a large deployment.

Companies running inference on 7nm can compete—but at a permanent cost disadvantage that compounds as models scale.

The Timeline: When Does 7nm Stop Being “Good Enough”?

China has a 2–3 year window where 7nm inference is “good enough.” After that, the efficiency gap becomes a strategic handicap.

The Verdict: Europe’s Leverage Extends to Inference Too

The inference objection doesn’t weaken the argument—it adds a timeline:

• Training: EUV leverage is immediate (always true)

• Inference: EUV leverage is delayed but inevitable (2–3 year window closing)

• Real-time/Edge AI: EUV leverage is already here (Tesla, robotics, AR/VR)

By 2026–2028, the 7nm ceiling will constrain not just innovation (training), but deployment (inference) at competitive cost, latency, and power efficiency.

Europe doesn’t just control who can push the AI frontier. It controls who can scale it economically.

5. Fabs — Where Physics Sets the Pace

Even with EUV, you still need fabs that can run it. A modern fab costs $20–40B, takes 3–7 years to build, and consumes energy and water like a city [4]. Only three companies can run leading-edge fabs:

And TSMC? It is already at or near 100% utilisation across its advanced nodes (5nm and below) [6]. There is no slack. No “extra” lines. No emergency capacity. Chip supply is physically capped.

This concentration creates a geopolitical single point of failure. A Chinese blockade of Taiwan would halt 90% of advanced chip production overnight. The US CHIPS Act’s $52B is better understood as insurance than replacement—replicating TSMC’s 30-year learning curve will take a decade minimum, if it succeeds at all. This is why European control of EUV matters: even if fabs diversify geographically, they all still need Dutch permission to exist.

6. Chips — Where Yield, Packaging & Memory Decide Everything

Most people think “more wafers = more chips.” Wrong. The three chokepoints are yield at 3nm/5nm, advanced packaging, and high-bandwidth memory (HBM).

Packaging (CoWoS, 2.5D, 3D) is now as constrained as fabs, with 12–18 month waits as Nvidia pre-books the majority of capacity [7]. Meanwhile, HBM is now the primary bottleneck for GPU availability, with key suppliers like SK hynix reportedly sold out for years.

7. GPUs — Allocation, Not Availability

GPUs are the visible symbol of the AI race. But behind them sits a three-way bind: TSMC wafers → packaging → HBM → Nvidia boards → your datacentre. If any link slows, everything slows.

As of late 2024, lead times remain long. Nvidia’s next-generation Blackwell GPUs are reportedly sold out through 2025, and while H100 availability has improved, securing large-scale clusters remains a challenge. The mantra “just buy more GPUs” really means: “hope the entire semiconductor chain had spare capacity two years ago.” It didn’t.

Your cap table is your story of ownership. Get it wrong, and you’ll lose more than equity — you’ll lose leverage.
In this live workshop with Antoine Bruna (CEO, Capquest), we’ll clean up the chaos and show you how to model dilution, vesting and exits in plain English.

What we’ll cover:

• How to structure a clean, investor-ready cap table (and fix messy ones).

• Understanding vesting, option pools, and founder dilution — what they mean in real terms.

• Modelling waterfall scenarios so you know who actually gets what at each stage or exit.

• Common founder mistakes that destroy leverage — and how to avoid them.

• How Capquest automates all of it: equity tracking, updates, vesting schedules, and round planning, so you’re always due-diligence ready.

What you’ll get:
A clearer view of your ownership journey, practical steps to protect it, and a free Capquest account to start modelling your own cap table. Use the code F42START

For the ❤️ of Startups

Register here

8. Datacentres — Warehouses Without Hardware

Land is easy. Power is not. Cooling is a tax. Water is a limit. And transmission is the killer. Substations take 2–5 years to build; transmission upgrades take 3–10. A datacentre without GPUs is a warehouse. A rack without HBM is dead metal. A GPU without chips is a purchase order. The constraints live upstream.

9. AI Models — The Glamour Layer

GPT. Claude. Gemini. Llama. All impressive. But models don’t create sovereignty. They consume it. Your AI capability is bounded by wafer allocation, GPU lead times, export controls, grid capacity, and, ultimately, ASML’s delivery schedule. Models sit at the top. Power sits at the bottom.

10. So Who Actually Controls AI?

Strip away the hype. The power is distributed in a clear hierarchy.

11. The Hard Truth

AI isn’t a software race. It’s a supply-chain contest.

The hierarchy is clear: materials → machines → fabs → chips → compute → AI → global leverage.

Europe controls the machines. Taiwan controls the fabs. The US controls the cloud and the software ecosystem. China controls the minerals.

Everyone else rents sovereignty.

This isn’t about ambition. It’s about physics, engineering, and who owns the bottlenecks. And the most critical lever—the one that determines who can even build a modern fab—is not held in Washington or Beijing, but in The Hague.

Welcome to Europe — where your AI desires and FOMO are controlled.

References

[1] ASML is the world’s sole supplier of EUV lithography machines, a fact widely reported by financial news outlets and confirmed in the company’s own statements.

[2] Zeiss SMT’s exclusive role as the supplier of EUV optics is documented in numerous industry analyses and reports on the semiconductor supply chain.

[3] Trumpf’s role in providing the high-powered CO₂ lasers for ASML’s EUV systems is a key part of the established EUV ecosystem.

[4] Fab construction costs are regularly cited by industry analysis firms like SEMI and consulting groups such as BCG.

[5] TSMC’s ~90% market share in advanced semiconductor manufacturing (sub-7nm) is a widely cited statistic from market research firms like TrendForce and in reports from geopolitical think tanks like CSIS.

[6] Information on TSMC’s capacity utilization is derived from its quarterly earnings reports and analysis by semiconductor industry watchers.

[7] Lead times for CoWoS and other advanced packaging are reported by industry trade publications such as Digitimes, based on supplier and customer reports.

[8] The Dutch government’s authority over ASML export licenses has been the subject of official government announcements and extensive reporting by outlets like Reuters and Bloomberg.

Thank you for reading. If you liked it, share it with your friends, colleagues and everyone interested in the startup Investor ecosystem.

If you've got suggestions, an article, research, your tech stack, or a job listing you want featured, just let me know! I'm keen to include it in the upcoming edition.

Please let me know what you think of it, love a feedback loop 🙏🏼

• ❤️ Love it, keep them coming!

• 👍 Great, always a good read.

• 😊 Enjoyable, but let’s see some variety.

• 👎 Okay, seen better.

• 🛑 Get a different job.
  
  Subscribe below and follow me on LinkedIn or Twitter to never miss an update. 

Subscribe now

For the ❤️ of startups

✌🏼 & 💙
Derek