ARTHUR - BUILD IN LIVE #2
Systems Coming Online: Auth, Memory, Documents & Orchestration
Every week, Iâm opening up the build of Arthur â what weâve shipped, what weâve learnt, and whatâs coming next.
Edition #2 is where the system starts to feel real. The foundations have moved from theory into working parts you can actually touch.
Arthur is your mentor, co-founder and team. One conversation on the front. A full organisation working behind it â quietly, instantly, just for you..
If you wish to get involved, find out more, sign up for early access, simply join the community
Arthur is getting his organisation online
The focus this week was simple:
make the experience stable, predictable, and ready for real founders.
Arthur is now moving from âsmart chatâ into an organised, multi-agent workspace that understands you, remembers you, and produces work you can actually ship.
Hereâs what we built.
â
What we shipped
đ Authentication & access control
The app now has a proper identity layer powered by Amazon Cognito.
Clean login, predictable permissions, and the right walls in the right places. The boring but essential plumbing that keeps the whole system steady as we move into beta.
đȘ Early access sign-up (with built-in virality)
The new early access sign-up flow is complete and will roll out early next week.
This version has light virality baked in from day one:
share on social â earn Fuel Units
refer a founder â earn Fuel Units
join the community â earn Fuel Units
Simple actions. Clear rewards.
A clean way for founders to get earlier access â and for us to grow the waiting list without shouting.
đ Document generation â S3 (and Google Workspace)
Arthur can now generate documents on demand
(product specs, investor notes, briefs, updatesâŠ)
and store them straight into Amazon S3.
This means:
repeatability
consistency
versioning
no more âlost in chat historyâ
The system can finally produce work the same way a real team would.
For the MVP, Arthur will also plug directly into Google Docs, Sheets and Slides, using your existing Google Workspace as the backbone for sharing, versioning and live editability.
Your documents stay where your team already works; Arthur handles the heavy lifting in the background.
đ§ Message history (Memory Layer v0)
You can now reopen any past conversation â just like using ChatGPT or Claude.
Under the hood, this is the first layer of Arthurâs long-term memory system, which will later plug into The Source, RAG, and structured start-up profiles.
Right now this gives you:
continuity
fewer repeated questions
the sense of a system that actually remembers you
Itâs the start of a workspace that grows with you instead of resetting every time.
𧩠Tuning the brain: system prompts & orchestration
In parallel, weâre continually tuning all system prompts and the orchestration logic behind Arthur.
Weâre feeding in fresh context files all the time â founder history, start-up facts, decisions, recent outputs â so Arthur never answers in a vacuum.
Every reply runs against a living source of truth to keep it:
relevant
correct
up to date
consistent with what youâve already told us
This is the gap between âa smart chatbotâ and âa real co-founderâ.
Arthur doesnât just talk â he updates.
đ§Ș Jiva testing continues
Weâre still pushing on the agentic backbone.
Jiva is successfully spinning up agents from our blueprint, but the response flow through the model layer isnât yet as stable or predictable as we want â exactly the kind of friction we expect (and want) at this stage.
Better to expose and fix these failure modes in Week 2 than discover them in Week 9.
This is why Build in Live exists:
less theatre, more truth about what it takes to get a serious multi-agent system production-ready.
𧔠Behind the scenes: a full workforce coming online
Underneath all of this, weâve now defined and begun wiring in the full blueprint of Arthurâs agentic workforce â an organisation of specialist departments doing the deep work founders normally drown in.
Not one big model.
A team, with hierarchy and responsibilities.
At a high level:
The Orchestrator â the supervisor that routes jobs, checks quality, and enforces schemas, citations and freshness.
Facts & Changefeed â the source of truth that maintains the /facts spine, timestamps everything, and pushes updates downstream.
Legal, Governance, People, Finance, Security â the backbone agents producing the documents, policies and controls every start-up needs.
GTM, Content, Sales, Retention â the growth engine doing research, positioning, messaging, campaigns and pipelines.
Fundraising & IR â the capital-readiness team building decks, KPI tear sheets, investor maps and due-diligence packs.
Tech & Delivery â the internal CTO office: architecture, reliability, security, observability, and the operational scaffolding around it.
Each department has its own outputs, templates and quality gates.
Jivaâs job is to coordinate the entire organisation, so Arthur can quietly run a real company behind your conversation.
Weâre not building a chatbot.
Weâre bringing an entire workforce online â piece by piece.
đ§ Whatâs next
Connecting the new authentication layer to founder profiles and start-up data
Wiring document generation into the Operator so departments can produce structured work, not just text
Stress-testing message history with multi-step workflows and deeper context
Hardening the Jiva â model-layer routing so agent responses are stable, predictable, and consistent across providers
More orchestration tuning so the right âdepartmentsâ wake up at the right moment with the right brief
Rolling out early access sign-up + Fuel Unit mechanics next week
FOR THE â€ïž OF STARTUPS
Help steer what we build next
Iâm not building Arthur in a vacuum.
If thereâs a workflow, bottleneck or ugly bit of founder work you want Arthur to take off your plate, tell me.
What youâre trying to do
What slows you down
What âdoneâ would look like if Arthur nailed it
These go straight into the build plan for the next sprints.