Two different things share this page. Projects are systems I've built and actually run — mostly with AI, mostly on my own work. Play is the other half of my brain: photography, animation, design for its own sake. Neither is client work, which is why neither is on the front page.
A governed workspace where AI agents do real work against my files under written rules. Every decision that governs future work gets an ID, a date and a quote, so a rule made once is still in force a week later. The agents read the register before they act.
Open the modelRoles are discovered, read first-party where possible, and scored against a written rubric that the feedback then changes. Companies are tracked separately from the roles they post, because the two questions are not the same one.
Write-up comingNewsletters arrive, stay unread, and what is worth keeping is extracted into a catalog organized by the situation you are stuck on rather than by topic. Anything built from a headline alone is marked as such, so a thin source cannot masquerade as a read one.
Write-up comingOne scheduled run each morning, reading the sources I would otherwise not get to, and writing the result somewhere I will actually see it. The smallest of these systems and the one that proved the pattern.
Write-up comingAnything said in passing lands in an inbox. Anything that governs future work becomes a numbered decision with a date and a verbatim quote. Three scheduled agents read that register at the start of every run, so a rule made on a Tuesday is still in force the following week without anyone restating it.
Click any node to see what that part of the system actually does.