Execution
Do the work. Execution operates through hat-based workflows — ordered sequences of behavioral roles that structure how work progresses. Each unit moves through its workflow in iterative cycles called bolts.
Key Artifacts
- Completed deliverables for each unit
- Quality gate results demonstrating standards compliance
- Operational plan for the delivered outcome
- Progress notes and decision records
Hats and Workflows
A hat is a behavioral role assumed during execution. Hats are not people — they are modes of operation that define focus, constraints, and expected outputs.
| Hat | Focus |
|---|---|
| Planner | Analyze the unit, identify approach, create execution plan |
| Executor | Produce the deliverable according to the plan |
| Reviewer | Evaluate deliverables against success criteria |
A workflow is an ordered sequence of hats. Workflows are configurable per domain:
Bolts: The Iteration Cycle
Work progresses through bolts — iteration cycles within units. Each bolt advances the work and produces a reviewable increment. Quality gates fire after every executor action, creating a tight feedback loop.
Operating Modes
Supervised
Human approves each significant action. AI proposes, human validates, AI implements, human reviews. Best for novel, high-risk, or foundational work.
Observed
AI works while human watches in real-time. Human can intervene but does not block progress. Best for creative, subjective, or training scenarios.
Autonomous
AI iterates independently until success criteria are met, using quality gate results as feedback. Human reviews the final output. Best for well-defined tasks with verifiable criteria.
Domain Examples
Software: A bolt might implement a feature, run tests, and fix failures.
Marketing: A bolt might draft content, run brand review, and refine based on feedback.
Research: A bolt might design an experiment, collect data, and analyze preliminary results.
Operations: A bolt might draft a runbook, simulate execution, and adjust procedures.