2026—active—systems model
Designed patterns for AI work where agents do the repeatable parts and people stay close to important decisions.
- role
- architect / operating-model designer
- context
- AI workflows with review and supervision
evidence
- Separated execution, coordination, and review
- Kept human operators near high-consequence judgment
- Treated checks and review loops as first-class parts of the system
sourcesRelated essay
ai workflowsevaluationsupervisionoperator control
The center of the work is not making the model do more. It is deciding what should be delegated, what should stay inspectable, and where a person needs control because the consequence of being wrong is too high.
That shows up as layered workflows: planners, workers, reviewers, checks, durable artifacts, and release gates. The system should produce useful work, but also produce the evidence needed to decide whether that work can be trusted.
I design AI systems around the work first: define the job, the memory, the review loop, the fallback path, and the evidence a person needs before trusting the result.