Tag: #how-i-work

Yesterday I took the password off the site; today I'm submitting the sitemap to Google. What I owe a text I didn't write, and what broke in the first hour.

Call Copilot has a local mode: recognition runs right in the browser and the audio never leaves the device. A handful of people switch it on, nearly all of them in the EU. The privacy bet hasn't paid off.

How I moved from prompting task after task to a cycle of plan → tasks → review agents → validation. Where this construction wins, what the review phase costs, and why throwing E2E tests out of the cycle was a mistake.

I gave an agent access to my Hetzner account through the browser, to the server over SSH, and to GitHub. It walked through the whole infrastructure, broke nothing, and wrote a report with a cleanup plan — minus €700 a year in expenses.

Anthropic's new, strongest model — Fable 5 — became available for a week — and the first thing I did wasn't write code with it. I pointed it at reviewing my changes. It found a staggering number of bugs and drew up a fix plan itself.

I gave an agent a browser and asked it to spend my Audible credits on books it had picked itself. It bought the first one with a credit, the rest — with my money: almost €200. A breakdown of where exactly I went wrong.

A holiday week in stolen moments: semantic search via embeddings, a date filter, a chat that links back to the source notes. How the voice journal works inside — and why it's time to stop improving it.