The strongest AI operator is not necessarily the strongest operator for every stage of work.
An operator with high leverage may be perfect for generation and wrong for review. An operator with high persistence may be perfect for construction and wrong for discovery. Workflow fit is not a personality trait — it is a learned mapping between operating patterns and the stages of your company's actual work.
See Diagnose Design a PilotTwelve stages. One pipeline.
Work is not a single act. It is a sequence of stages, each with different demands on the AI operator. The canonical model has twelve stages. Not every workflow uses all twelve — but every workflow can be mapped onto this scaffold.
Discovery
Identifying the problem space. High exploration, low structure. Rewards broad context and fast iteration.
Research
Gathering and synthesizing evidence. Rewards high leverage — extracting signal from large input volumes.
Framing
Defining the problem. Rewards context construction and precision. Low output, high structure.
Planning
Sequencing the work. Rewards decomposition and persistence. Plans must survive across sessions.
Generation
Producing first drafts. Rewards high yield and high leverage. Output volume matters, but so does quality.
Construction
Building the artifact. Rewards high persistence, context reuse, and low variance. The build-heavy stage.
Iteration
Refining the artifact. Rewards stability and consistent technique. Small, precise changes across sessions.
Review
Evaluating the artifact. Rewards verification-heavy patterns and critical context construction.
Validation
Testing against requirements. Rewards precision and thoroughness. Low tolerance for volatility.
Integration
Combining components. Rewards multi-stage consistency and context carry-forward across modules.
Delivery
Shipping the result. Rewards reliability and documentation. Context must be transferable.
Maintenance
Sustaining the result. Rewards stability and low volatility over long windows. Consistency over brilliance.
Which operators fit which stages.
Workflow fit is a mapping between operating patterns and workflow stages. It is learned from observations and company-defined outcomes — not assigned from personality labels. Below is an illustrative fit table for a software development workflow.
| Stage | Top Fit Operators | Signal |
|---|---|---|
| Research | 014, 022, 041 | high leverage / stable |
| Architecture | 008, 031 | high persistence |
| Build | 003, 012, 031 | high yield / low variance |
| Review | 026, 040 | stable / verification-heavy |
| Integration | 012, 026 | multi-stage consistency |
All fit scores on the website must be clearly labeled illustrative until validated. Fit is learned from your company's own outcomes — not from a universal ranking.
Fit heatmap — operators × stages
An illustrative heatmap showing operator fit scores across workflow stages. Darker cells indicate stronger fit. This is synthetic data for demonstration.
Heatmap scores are illustrative. Real fit scores are learned from your company's defined outcomes and validated against re-measurement.
What workflow fit is for.
Learn which operating patterns, operators, tools, and models perform best at which stages of the company's own workflow.
Workflow fit is not a ranking exercise. It is a learning system. The goal is to discover the mapping between operating patterns and stage outcomes that is specific to your company — then use that mapping to assign, train, and intervene with precision.
Do not assign from personality labels.
It is tempting to assign stage fit from generic personality labels — "she's creative, put her in discovery" or "he's detail-oriented, put him in review." This is the wrong approach. Personality labels are not measurements. They do not predict how an operator actually operates AI.
Wrong
Assign stage fit from personality labels, self-report, or generic role descriptions. "Creative people go in discovery."
hypothesisRight
Learn fit from observations + company-defined outcomes. Measure how the operator actually operates AI at each stage, then map the pattern to the outcome.
measuredDo not assign stage fit from generic personality labels. Learn fit from observations + company-defined outcomes.
Your workflow is not the canonical model. That is fine.
The twelve-stage canonical model is a scaffold, not a mandate. Every company's workflow is different. Companies can define custom stages — with custom names, custom demands, and custom outcome definitions. The fit system learns against your stages, not ours.
Define your stages
Name the stages of your actual workflow. They do not need to match the canonical twelve. They need to match how your company works.
Define your outcomes
For each stage, define what a good outcome looks like. This is the validation target. Without it, fit cannot be learned — only guessed.
Learn the mapping
The system observes operators working at each stage, measures their operating patterns, and learns which patterns predict your defined outcomes.
Custom workflows are the norm, not the exception. The canonical model exists to provide a shared vocabulary — not to impose a universal process.
Fit is learned. Interventions are tested. Results are re-measured.
Once you know which operators fit which stages, the next step is developing interventions — and verifying that they work.
See Develop Verification