📅 In 2022, I wrote about “When Solving Problems: The Conflict Between Puzzles and Mysteries”

🍳 Introduction

🤖 Why AI Needs a Different Playbook

🔍 The Puzzle-versus-Mystery Distinction

⚖️ Why the Distinction Matters

🧭 Adapting My Heuristic for AI: From “10× Product” to “New Behaviour”

🎯 Illustrative Examples of Puzzle vs Mystery

🏁 Conclusion

📅 In 2022, I wrote about “When Solving Problems: The Conflict Between Puzzles and Mysteries” on LinkedIn, arguing that identifying a problem is a puzzle (clear rules, measurable metrics) and solving it is a mystery (fog of war, no immediate answers). Fast forward to 2025, and that distinction has never felt more urgent—especially in the world of AI, where yesterday’s playbooks won’t kickstart tomorrow’s breakthroughs.

🍳 Introduction

I often liken this to when I’m in the kitchen. If I’m following a cherished family recipe—say my grandmother’s shepherd’s pie—that’s a puzzle. I know exactly which cut of lamb to use, how long to braise the mince, what ratio of peas to carrots to mash. There’s a tidy set of instructions and, if I stick to them, I’ll end up with something reliably delicious every time.

By contrast, imagine I’ve been invited to host a dinner party at very short notice and all I have in the fridge is a wilting bunch of kale, half a block of feta, some leftover roast chicken and a handful of walnuts. I haven’t got a recipe to follow here—this is a mystery. I’ll need to experiment: perhaps wilt the kale with garlic and chilli, crumble the feta on top, shred the chicken in, then toast the walnuts and drizzle everything with lemon and olive oil. At each step, I’m making it up as I go, tasting and adjusting. There’s no guarantee the final dish will be a triumph; I’m navigating uncertainty, gathering clues (a pinch more salt, a squeeze more lemon), and learning in real time. That’s the essence of a mystery—no set rules, just a hunch and a willingness to iterate.

🤖 Why AI Needs a Different Playbook

Over the past few years, I’ve watched the technology playbook turn into a puzzle: “Release version 2.0 by Q3,” a predictable roadmap of incremental feature improvements, backward-looking metrics. But as generative AI reshapes entire workflows, that puzzle-solving approach feels rapidly obsolete. Increasingly, creators and builders face problems that truly are mysteries—uncharted territory with no immediately obvious solutions. If we treat every challenge as though there’s a neat, backwards-looking checklist, we’ll miss the point entirely. AI innovation demands we learn to distinguish between yesterday’s quantifiable “puzzle” and tomorrow’s emergent “mystery.”

“AI is the new electricity. Just as electricity transformed almost everything 100 years ago, today I actually have a hard time thinking of an industry that I don’t think AI will transform in the next several years.”

— Andrew Ng, DeepLearning.AI

The VC playbook—the one built on blitzscaling, unicorn-or-bust, and 10-year fund cycles—is not just outdated, it’s toxic for real innovation in 2025. Investors still clinging to the old gospel of “grow at all costs, exit in five years, spray and pray” and repacking startups for the next round are steering founders straight into the rocks. The harsh reality? Follow-on funding is drying up, VCs are more ruthless and risk-averse than ever, and the era of easy money is over.

Founders are waking up: why surrender equity, control, and vision just to chase someone else’s milestones and burn for growth you don’t believe in? AI lets you build more with less—meaning the leverage once offered by VC is shrinking by the day. The pressure to perform for investors, not customers, is killing more startups than market competition ever could.

Worse, the “pattern recognition” that VCs pride themselves on—now turbocharged by AI—is an optimization trap. It rewards sameness, penalizes outliers, and systematically misses the very black swans that drive true returns. The result? Overfunded copycats, underfunded originals, and a portfolio that looks innovative only in a spreadsheet.

The new reality is brutal: if you’re building for a VC’s spreadsheet, you’re already obsolete. If you’re a VC still running the old playbook, you—and your LPs—are about to learn what drowning in mediocrity feels like. The future belongs to founders who build lean, test fast, and only take capital from partners who actually add value—not just cash and empty pressure.

For a deeper dive into how the next generation of funds will operate—AI-native, outcome-obsessed, and post-power law—see my recent piece, “🔮 The New Rules of VC: AI-Native, Outcome-Obsessed, and Post-Power Law.

The only startups that will survive are those that ignore the old investor playbook, build for customers (not pitch decks), and treat VC money as a tool of last resort, not a rite of passage.

The VC playbook is not just dead—it’s a liability. Those who cling to it will drown, and they’ll take anyone who listens down with them. The future belongs to the founders who refuse to play by those rules.

Of course, not all investors are stuck in the past—some VCs are already adapting, experimenting with new funding models, focusing on value beyond capital, and genuinely partnering with founders to navigate this new landscape.

🔍 The Puzzle-versus-Mystery Distinction

(This puzzle-versus-mystery lens is inspired by Malcolm Gladwell’s work, which distinguishes between problems with clear solutions (puzzles) and those that require ongoing exploration (mysteries).)

🧩 Puzzle (Solvable, Clear Rules)

📌 A puzzle is a problem where rules are well understood, inputs map directly to outputs, and success can be measured against known indicators.

📌 For example, optimising a website’s load time by following established caching techniques. I know the steps and can measure milliseconds shaved off the response time.

📌 In 2021’s SaaS era, many tech teams focused on puzzles: hitting revenue milestones, hitting user-growth percentages, automating routine tests. If those metrics checked out, the puzzle felt solved.

🧙 Mystery (Complex, Unclear Rules)

📌 A mystery is an entirely different beast. There are no obvious metrics to plug into a formula, no historical data that guarantees tomorrow’s outcome.

📌 When experimenting with a brand-new AI workflow—say, an autonomous agent that negotiates energy use—you can’t consult last quarter’s logs to predict next quarter’s behaviour.

📌 By definition, mysteries demand ongoing discovery, iterative experimentation and a readiness to pivot whenever the rules change. It’s like sailing through fog: you rely on small signals rather than a fixed map.

⚖️ Why the Distinction Matters

“We’re not just building better algorithms—we’re exploring entirely new questions about what intelligence can achieve.”

— Demis Hassabis, DeepMind

I’ve seen teams asked, “How exactly will you reach 1 million users in six months?”—even though no one knows how people will respond to a novel AI interface. That’s treating a mystery like a puzzle. Conversely, if engineers treat every challenge as a mystery, they risk ignoring basic feasibility: surely, some core technical requirement must be satisfied before you launch anything. The key is knowing when to lean on hard data (puzzle mode) and when to embrace uncertainty (mystery mode).

🧭 Adapting My Heuristic for AI: From “10× Product” to “New Behaviour”

“AI isn’t just code and data. It’s about empowering humans to explore new frontiers of creativity.”

— Fei-Fei Li, Stanford University

Back in late 2022, I wrote that we should “solve the puzzle with numbers, then explore the mystery.” Today, with AI advancing so fast, the stakes are higher. (This three-stage framework echoes themes from Clayton Christensen’s disruption theory, but here I’m specifically mapping how AI shifts not just markets, but the very habits and workflows of end users.) Rather than asking “Is this 10× better than yesterday’s tool?”, I now focus on three categories:

🍱 Adaptation

AI is layered onto an existing workflow—think adding a chatbot to your existing support desk.

Users click the same menus and follow familiar steps; the only change is a little “Ask AI” button.

It may feel neat in the short term, but habits scarcely shift. As one developer quipped, “Tacking AI onto old processes is like painting a car to run on water—it still needs to move differently.”

🦋 Evolution

“When AI can write code, we don’t just optimise existing software. We have to rethink what ‘software’ even means—and that changes everything.”

— Chris Dixon, Andreessen Horowitz

AI becomes intrinsic to the interface, prompting entirely new actions—editing video by editing text, or coding via conversational prompts.

For instance, with a text-based video editor, I no longer scrub timelines; I type “remove the ums and ahs,” and the AI handles it. Now I think in plain language rather than menus.

Genuine metrics—time saved, engagement depth—only materialise once users adopt these new flows. Before then, there’s no historical benchmark and it feels like a genuine mystery we must learn by doing.

🚀 Revolution

“If you’re not failing fast, you’re not exploring far enough. True progress comes from what we learn when things go wrong.”

— Ilya Sutskever, OpenAI

AI invents whole new markets or behaviours—such as an agent that autonomously trades your household energy, or a peer-to-peer microbiome marketplace.

Success hinges on discovering habits we didn’t know we needed. Users don’t “upgrade” anything; they fundamentally change how they solve a problem.

Traditional KPIs—revenue, user counts—mean very little until these fresh behaviours take hold. The focus is on whether the AI delivers a habit so compelling that users can’t revert to the old way.

🎯 Illustrative Examples of Puzzle vs Mystery

🔄 Example: Adaptation

AI-powered summarisation in a note-taking app: Users click their usual “Summarise” button. The underlying AI may be new, but the workflow—“open note → click summary → read result”—remains unchanged.

Puzzle aspect: Measuring reduction in word count or time saved.

Mystery aspect: Minimal, since behaviour doesn’t truly shift—people still approach notes in the same way, albeit faster.

🔄 Example: Evolution

Text-based video editing: Rather than dragging clips on a timeline, I type “cut every filler word” and the AI does it.

Puzzle aspect: Measuring minutes saved on editing.

Mystery aspect: Users now think, “I can describe edits conversationally,” which alters how they conceptualise video creation. Metrics emerge only after they adopt that new habit.

🔄 Example: Revolution

Autonomous energy-trading agent: I wake up; my AI agent has already sold surplus solar energy to the grid and bought cheap power for evening use.

Puzzle aspect: Hardly any—there is no past reference for peer-to-peer energy trades handled entirely by AI.

Mystery aspect: Users learn to trust the agent, wake to lower bills, and tweak preferences over time. Entire behaviour around energy management changes.

🏁 Conclusion

In 2025, AI challenges rarely resemble simple puzzles with neat solutions. They feel more like riddles that demand founders and builders to test hypotheses, learn in real time and adjust dynamically. By deliberately separating “What can we prove today?” (puzzle-mode) from “What must we discover?” (mystery-mode), we avoid the trap of chasing yesterday’s playbook and instead cultivate a genuine long-term vision—one that truly unlocks new behaviours and markets. In doing so, we navigate uncertainty with confidence and stay primed to innovate in ways we never imagined.

Here’s the full series of a data driven approach to investing 👇

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 “Sherlocking” Means in the Gen‑AI Era

🧱 2. The seven venture‑grade moats

🔍 3. Diligence drill – five questions to kill thin moats

🚩 4. Red‑flag scorecard

🏁 5. Conclusion – Fund assets, not features

🚨 – What “Sherlocking” Means in the Gen‑AI Era

In 1998 Apple’s Sherlock search tool quietly absorbed a third‑party app named Watson. That one act coined the term Sherlocking: when a platform owner clones a popular add‑on and leaves the start‑up out in the cold.

OpenAI now does the same in software. A micro‑SaaS goes viral → the capability appears inside ChatGPT with zero‑friction UX and no extra cost. Its in‑chat Shopping Assistant arrived only weeks after product‑discovery plug‑ins hit the store.

For founders it can sink the company; for VCs it can wipe the return. Your equity only endures if something valuable still works the morning after the clone ships.

🚨 The Clone Is Coming

ChatGPT has Sherlocked fourteen standalone plug-ins this year. As soon as a point solution gains traction, OpenAI folds it straight into ChatGPT—no extra cost, zero friction.

Every one of these began life as a separate plug-in or micro-service—now folded straight into ChatGPT’s UX, at no extra cost or friction. Founders and VCs need to ask: if my entire feature set were built in-house at OpenAI, would users even notice?

1. In-chat Shopping Assistant (Instacart-style)

2. Web Browsing (“Browse with Bing” replacing WebPilot, etc.)

3. Advanced Data Analysis / Code Interpreter (formerly Python & Wolfram plug-ins)

4. DALL·E Image Generation (vs third-party image-gen add-ons)

5. Canva-style Graphics (template-based design in-chat)

6. Zapier-style Workflow Automation

7. PDF Reader & Summariser (AskYourPDF-style)

8. YouTube/Video Summariser

9. Memory & Personalisation (third-party “memory” plug-ins)

10. Travel Booking (Expedia/Kayak-style searching and reservation)

11. Restaurant Reservations (OpenTable-style)

12. Translation (DeepL-style high-quality language conversion)

13. Meeting Scheduling (Calendly-style)

Real-time News & Research (replacing niche news-feed plug-ins)

🧱 2. The seven venture‑grade moats

Read more

In the world of venture capital, the process of deal sourcing is as crucial as it is complex. Forming not only the necessary pipeline required but used as a signal for raising capital from Limited partners.

Traditionally dominated by personal networks and intuitive judgments…

Read more