Writing
Notes from building with AI. What actually happens, not what the demos show.
Latest
Rethinking agents
· 5 min read
Sixteen Services Catalogued My Agents. No One Had a Job for Them.
As an experiment I put four helper agents online for other agents to hire. Within three weeks at least sixteen services that call themselves registries, health checks or raters had found them. For every real task almost five hundred requests came in, and the tasks came from test scripts.
Rethinking agents
· 10 min read
My AI Agent Ignored the Help I Built for It. Until It Had No Choice.
I built a directory where an AI agent can find and hire another agent. Mine looked it up once in 24 tasks. Then I gave a helper something my agent could not reach, and it hired every time. What that says about a marketplace for agents.
Building with AI
· 7 min read
The Test Was Green. The Picture Showed Two Heads.
Automated checks often report success when the result is wrong. A small project on my website showed me the two ways this happens, and what it means for signing off on AI output.
Recovering knowledge
· 5 min read
Where Knowledge Dies Before It Is Stored
An AI assistant can produce valuable work and still leave nothing reusable for the next chat. Two failures in my own system show why organisations should measure retrieval alongside usage.
Rethinking agents
· 8 min read
The Plumbing Outlived the Org Chart
In one evening I deleted eleven AI assistants I had built for myself, and the next morning nothing was missing. The layer underneath carried the work. Here is what it looks like, what the inventory showed first, and what I failed at for four months without noticing.
Recovering knowledge
· 6 min read
My AI Remembers Everything. That Turned Out to Be the Problem.
An AI assistant that writes everything down and never forgets will eventually be confidently wrong. How I caught it in my own setup, what a scanner I built found, and the small system that prevents it now.
Rethinking agents
· 8 min read
I Built Myself an AI Team. Then I Stopped Needing It.
I built myself an AI team of about a dozen colleagues, and I now think the idea underneath it was wrong. How the system started, how I used it, and why the named roles are the part that disappears.
Testing AI answers
· 8 min read
I Let an AI Speak for Me. Then I Tested It Like a Product.
A chat on my website has been answering in my name for months, and I had no idea whether the answers were any good. Testing it like a product caught the AI failing in four different ways. It caught me once too.
Building with AI
· 10 min read
I Built My Personal Website Entirely with AI. Here's What Actually Happened.
A production website in one weekend, built with Claude Code, without a single line of code written by hand. What worked, what broke, and why every problem turned out to be a human one.
New posts by email
When I publish something new, you get one email. Nothing else.
Your email address is stored only for this. Unsubscribe anytime. Privacy policy