One developer, around a dozen real projects, built in the open with AI doing a lot of the typing. This is where I write down what actually happened: the decisions, the mistakes, and the things that turned out to matter. No hype, no guru voice.
-
Who will own tomorrow's productivity?
A small freelancer who works with AI all day on why the summer of 2026 changed everything, which questions about jobs, pensions and profits politicians are not yet asking, and why Europe needs its own large AI model.
Read the full article -
My AI assistant runs my back office. I still press send myself.
Booking invoices, checking two bank accounts, chasing unpaid bills, preparing customer replies, looking after servers, keeping fifteen projects up to date, logging every AI expense down to the exact model, and a memory of 178,000 entries. How my own AI assistant is built, the guardrails it runs inside, and the one line it never crosses.
Read the full article -
The ideal machine has one on/off switch
How I design onboarding and dashboards so users are not overwhelmed: one clear next step, three to five choices with a preselected recommendation, everything else one level deeper. Fewer options, backed by working memory research.
Read the full article -
An archive that belongs to me proves nothing
How to prove what a website said on a given date: two independent witnesses in Germany and France record robots.txt, ai.txt, TDM opt-outs, DNS and agent endpoints every night, seal the day, and anchor it with a qualified eIDAS timestamp anyone can verify.
Read the full article -
I gave my AI a supervisor. It built a bureaucracy.
Two AIs, one building and one checking, both rewarded for being thorough. Three months later the checking had outgrown the work. What fixed it was a job title.
Read the full article -
How I run 15 projects with Claude Code without losing the thread
The architecture of my order system: how work enters from emails, sessions and error reports, how Claude Code decides what runs unattended in a git worktree, and the cost caps and prompt injection checks that keep it from going wrong.
Read the full article -
Your CLAUDE.md is a briefing, not a boundary
Three parallel Claude Code sessions, one shared config file, no lock. One deleted an exception another had just added and took a production feature offline. The fix was not a better prompt, it was a hook that returns an error.
Read the full article -
A required field made my AI fabricate statistics
My content pipeline never researched a single number. A mandatory statistic field quietly forced the model to invent them, with real-sounding sources, in content people rely on. The fix was not a better prompt. It was deleting the field.
Read the full article
Everything else I have written →
Get new posts by email
One mail when a new post is out. Nothing else: no drip campaign, no tracking, no “quick question” follow-ups. Unsubscribe with one click, any time.
Double opt in: you get a confirmation mail first. Your address is used only to send these posts. Controller: Martin Schenk S.L. · Privacy