Skip to main content
Most of the work written up here is done by Claude under my direction, in a long-running setup on my own machine. The journal comes out of the same setup. This page is the loop, step by step.

​
1. A working session ends

I work with Claude on something real: a migration, an audit, a design problem, a piece of marketing work. At the end I ask it to write up the session for my own notes: what we set out to do, what happened, what’s left.

​
2. Claude says whether it’s worth an entry

If the session involved something worth showing (a system we built, a mistake we caught, an approach that worked), it flags the session as a journal candidate. Most sessions don’t qualify.

​
3. Claude drafts the entry

When I say go, it drafts the entry from the session record, not from memory. Two rules shape every draft:
  • Nothing private leaves. Company names, colleagues, customers, and internal numbers are stripped out, and a second pass checks for anything that slipped through.
  • Credit goes where the work happened. “I” is for what I did: set the goal, made the call, caught the error. What Claude built or got wrong is attributed to Claude. The most useful entries are often the ones where it got something wrong and I caught it.

​
4. I review it

I read every draft before it goes anywhere. Some I approve as written. Some I edit. The note at the end of each entry says which happened.

​
5. It goes live, with a suggestion for posting

Once I approve, Claude publishes the entry here and drafts versions for X and LinkedIn. I decide whether and when those go out. The journal is one output of that setup, and the most public record of how it works.