One Founder, Eight Weeks, Four Hundred Agents
I delivered a working, deployed MVP in eight weeks, alone, by directing more than four hundred AI agents.
Everything I have published: essays and manuals from From Lab to Life, research papers, and the archive. Each work tells you where you can read it.

Start here
I delivered a working, deployed MVP in eight weeks, alone, by directing more than four hundred AI agents.
Large language model (LLM) agents now carry out work that outlasts a single session: building software, running literature reviews, writing books, producing campaigns. Every session, however, begins with an empty context window, and when the work moves to a different agent harness, a different model or a different computer, whatever the previous session knew is lost.
Written deliberately as a lesson rather than a product story, taught through The Curator v3.80.0.
Three kinds of context, three write rules, and a project that outlives the session, the harness, the model and the machine
Building a modern website in the age of AI agents: a manual, told through the new website of The Curator
The Curator ships as a native Mac app (v3.45.0). Covers the three layers (personal wiki, Shared Brain, agent working state), the menu-bar widget, bring-your-own-key across three providers at about €5 a month, and plain-markdown portability.
New essays from From Lab to Life arrive on Substack before they reach this library.
The confession piece about two years of hand-written handoff files. The flaw: the most degraded model writes the most important document.
I turned fifty-plus articles into a corpus you can question, download, and hand to your own AI.
Can a frontier orchestrator on a subscription delegate to local open-weight models?
Seven months, two generations, one production rebuild — and five things I learned that I did not expect to learn.
This paper reports a longitudinal design-science case study of a single production artefact — Lumina, a multi-tenant AI assistant platform — across two complete generations built fourteen months apart in capability terms and seven months apart in calendar terms.
The behind-the-scenes rebuild of Lumina from Gen-1 to an enterprise-grade Gen-2.
A summer manual for the curious non-developer who keeps seeing people build apps out of thin air and wondering: could I actually do that?
Running Lumina with real customers six months after launch. A Ljubljana massage studio hit a forgotten 2,000-character personality limit mid-conversation; the systemic fix raised the limit fourfold and added live counters.
Argues that competition has shifted from a fight for resources to a fight for intelligence, using the 19-day frontier-model suspension of June 2026 as evidence of single-vendor risk.
Putting the Organisation of Tomorrow to Work; Use Cases, Vendor Independence, and the Brain Behind the Framework
A follow-up seven weeks after the Shared Brain launch. The author's own wiki has 3,305 nodes and 15,095 edges at v3.0.6.
Two Anthropic announcements 24 hours apart. Sonnet 5 (30 June) becomes the new default, and its real story is follow-through on multi-step tasks.
We are learning to build and ship real digital products using AI coding agents. The gap between idea and shipped product is no longer technical, it is contextual.
How Claude Code plus Playwright via MCP became an autonomous testing loop in early 2025, before "loop engineering" had a name.
Every founder I know has the same problem with email support. In the early days, you handle it yourself. You are fast, you are personal, and you actually know the product. Then you get busier. Response times slip. Users start to notice. You think about hiring someone, but the volume does not justify it yet. So you buy a helpdesk tool, set up some canned responses, and tell yourself you will get to the backlog this weekend. You never get to the backlog.
Compares Kimi K2.7 and Claude Fable 5 using real production data (24 landing pages, 12 prompts).
Release notes for The Curator v3.0.1 beta.20 to beta.23: Shared Brain synchronisation improvements, a new test-automation infrastructure, and stability fixes.
Inside the machinery that powers modern AI development, and why understanding it matters even if you never write a line of code