Notes from building with AI. What actually happens, not what the demos show.
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.
I deleted eleven AI assistants I had built for myself in a single evening, and the next morning nothing was missing. What stayed was the layer underneath. Here is what it looks like, what the inventory turned up first, and the thing I failed at for four months without noticing.
For months I did not notice that the files my AI assistant draws its knowledge from had stopped being true. How I finally caught it, what a scanner I built found in my own setup, and the small system that prevents it now.
I built myself an AI team of about a dozen colleagues, each one designed by hand, and I am now convinced the idea underneath it was wrong. How the system started, how I used it, and why the part everyone notices, the named roles, is the part disappearing.
A chat on my website has been answering in my name for months, and I had no idea whether the answers were any good. So I started testing it like a product. It caught the AI failing in four different ways, and it caught me once too.
A production website in one weekend, built with Claude Code, zero lines of code by hand. The honest version of what worked, what broke, and why every single 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. Details: Datenschutz.