Work record / Stories

Cutting The Fleet Down To What Actually Earns Its Keep

By mid-2026 I had built a wide fleet of scheduled AI routines, and I ordered a hard review of every one of them.

By mid-2026 I had a wide fleet of scheduled AI routines covering briefings, outreach, content and audits. I ordered a full review of each one against what it still did. Then I ordered a second, adversarial pass to challenge some of the calls the first pass made.

I killed or paused several routines outright, even routines already built and running. They were a social media growth bot, a dead outreach directory, and a personal daily synthesis that had never produced anything real. I also refused to put a brand new weekly routine on a schedule. It stayed a manual command until it proved itself by hand.

My rule is to automate last and to add automation back only where I truly miss doing the work by hand. Use is what earns a routine its place. Wiring a new process straight into a cron would repeat the very automate-first mistake I had just named. Anything able to reach the team automatically also had to pass through me first. I am the approval gate, and nothing auto-broadcasts.

The number of live routines fell. In the survivors we removed dead integrations and channels, and we folded a redundant weekly audit into a simpler daily check. The new weekly review started as a manual ritual, run by hand for two weeks before it was allowed to run alone.

If a routine cannot point to something it caught or delivered, its cleverness does not matter, so cut it. A system does not get to keep running just because it exists.

Story details

All stories · Work record