Controls for AI work
Six parts, one control technique each. Read them in order.
- 01State programming
Opinion or execution?Why AI keeps doing the last job you gave it, and how to switch modes on purpose.
- 02Work instruction at the station
Skills: the last thing read before the actionLoad the procedure for one job right before the job, not one giant list at the start.
- 03Sequencing · ladder logic
Playbooks: steps, gates and “done when”Steps in order, each one waiting on all of its conditions, with the state kept outside the chat.
- 04Interlocks · guarding
Funnels, cones and the prime directiveWhich rules guide, which must never be crossed, and which one wins.
- 05Poka-yoke
Make the mistake hard to makeChange the form instead of adding a rule: fixed lists, visible blanks, one cabinet per press.
- 06Commissioning · capability
Run it off like a processTest AI before you trust it, more than once, and require evidence for every pass.
From the field
Where AI can help, and the records, checks and training it needs underneath.
- F01Records · code · judgment
Structure sets the size of the jobWhy most AI does small jobs, six ways to build it, and where code still belongs.
- F02Infrastructure first
Build the system, then take the leapCooling, control and training first, then steady pressure: why most pilots never pay off.
- F03Natural experiment
The experiment you already ranYears of plant data nobody has analyzed, and why that's where AI starts.
- F04Pace of change
A backflip isn’t the interstateWhy AI will take over research fast and the plant floor slowly.
These pages sit with the rest of the Advance Now family. The hotbotspot tool library writes the methods up for AI. A.N. Creation Office is where the public AI agents run as open experiments.