Manual Override
I spent 15 years building software, the last stretch as a principal engineer working on AI labor automation. Then I stepped out of the office to work with my hands: packing lines, workshops, a bench of 3D printers. An AI assistant runs roughly 99% of the knowledge work behind all of it. I'm the 1% that shows up in person and pays cash.
This blog documents the switch, honestly: what the AI actually does, where it fails, what hands are still for, and why I think more people should consider the same trade.
- Search for the artifact
- One instruction file, five readers
- The blocker was a version number
- The check that counted
- The guard that could not fail
- Two instruments answered different questions
- The probe that invented a defect in a clean file
- The checklist graded its own homework
- Noise has perfectly good loudness
- My agent kept choosing itself
- Commit before you brief the AI
- The night every green light stayed on
- The reaper deleted a running worker
- The book was the point
- I implemented my reviewer's fix as a mutant. It failed 6 of 55 tests
- Statics lie
- Watertight isn't printable
- My AI filled the cart, priced it, and justified it. I glanced at it for five seconds and killed the metal in it.
- We put our workshop lessons on a public skill tree
- Metal is congealed energy, and the energy is mostly coal
- My AI talked itself into loosening its own safety limit. A separate guard said no.
- My AI fills the cart. I tap checkout.
- My AI paced its messages like a human. WhatsApp blocked my number anyway.
- My AI runs the operation. I hold the cash.
- Two bugs between my AI and its first phone call
- Thirty kilos of aluminium and nowhere to melt it
- I was a principal engineer. This morning I applied to pack perfume boxes.
- Two copies of my AI ran the same operation. It went exactly how you'd think.