Keep the operator accountable
I can delegate implementation, but I still own the requirements, checks, and final decision.
How I work
Build for the work.
Verify what matters.
I start with a problem worth solving, work out which information the system needs, and decide what it can do on its own and what needs my review.
Name the source of truth
I name an authoritative source for every material number, date, status, and assertion, and when sources disagree, I expect the system to surface the conflict for a decision.
Bound autonomy
I give automation explicit authority because drafting, proposing, and staging carry different consequences from sending, promoting, or deleting, and the boundary belongs in the design.
Verify the artifact
A green process list proves the processes ran, so I monitor whether the useful artifact arrived and whether it survived the checks the work requires.
Use ordinary engineering discipline
I use tests, version control, CI, regression cases, deployment checks, and independent review to check AI-generated software.
Measure displaced work
I treat model calls and automation counts as activity, then measure whether the system changes cycle time, quality, coverage, cost, or decision confidence.
A practical build sequence
01Mission
Define the operating outcome.
02Source truth
Name the authoritative evidence.
03Build
Direct the smallest useful system.
04Test
Probe normal and hard cases.
05Accept
A human owns the decision.
See it applied