in one week my fleet grew to 144 agents and closed 106 units of work. on its most productive day, i shut half of it down.
on august 1, 2026, i started an experiment. four department leads, each with its own lane, so i could steer them directly instead of passing everything through one agent. content, marketing, r&d, revenue. the idea was to parallelize my attention.
in seven days the canvas grew to 144 agents.
i don't even know what the projects were. my brain was hurting. i was giving instructions to one agent that was focused on another project. i was mixing instructions between terminals. mostly i was just pressing continue.
the fleet, meanwhile, was having its most productive week ever. the journal records 106 closed units of work in seven days. on august 8 alone, the day i closed half of it, it records 20 finished pieces, including:
2026-08-08 | blender-cat-claude | Built and documented a preserved stylized-cat experiment with a 16-bone rig, verified 24-frame walk loop, saved Blender project, and H.264 preview
that same day the fleet also finished an 807-frame video about ai sycophancy and a 1,597-frame reconstruction of an xkcd comic.
i closed content and marketing first. of course i lost some work. but after i closed them i felt relief. my clarity was more important.
today, writing this: the fleet was not failing. it was producing more than i could steer. the machines did not break, the steering did. i found i make more quality than quantity. my brain needs space to take actual decisions, and no amount of agents changes that.
folio 31 of 32 · written in cusco, peru, 12 SEP 2026