Process
Workflow, Assessment, and Reassessment
Every piece of work flows through a consistent workflow: assess it honestly, execute through phases, reassess when new information arrives, and retrospect when complete. The process is lightweight — each assessment takes 2–5 minutes.
Initial Assessment
Evaluate the work: work mode, involvement, size, risk, confidence, spec quality. Takes 2–3 minutes.
Execute
Work through phases: Research, Specification, Implementation, Review. Both tracks active.
Reassess
When triggers fire (research done, issues found, phase change), update the assessment. Learning, not error.
Retrospective
At completion: was the spec good enough? What would we do differently? Feed learnings into future assessments.
Work Mode
Who leads the work?
Involvement
How much human attention does this need?
Size
How big is the scope?
Spec Quality
How well-defined is the work? Clear acceptance criteria, or vague intent?
Risk
How predictable is the outcome?
Confidence
How well do we understand the scope?
The quality of the specification determines the quality of the output. This is the most valuable assessment dimension.
| Quality | Indicators | What It Means |
|---|---|---|
| High | Clear acceptance criteria, testable conditions, defined scope | AI can execute with minimal iteration |
| Medium | General direction clear, details emerging | Expect collaborative refinement |
| Low | Vague intent, undefined scope | Research phase needed before specification |
The strongest predictor of delivery risk. If you're in a red zone, invest in specification before starting production.
| Size | Spec High | Spec Medium | Spec Low |
|---|---|---|---|
| S | Smooth, fast | Minor iteration | Quick to fix |
| M | Smooth | Some iteration expected | Research needed first |
| L | Smooth if specs exist | Expect surprises | Stop. Invest in spec first. |
| XL | Delivers if disciplined | Risk zone. Phase the delivery. | Do not start. Spec first. |
Assessments reflect current knowledge. As work progresses through phases, understanding deepens and assessments naturally refine. This is learning, not error.
| Phase | Typical Profile | Confidence | Spec Quality |
|---|---|---|---|
| Request | Based on description alone | Low — many unknowns | Low (intent only) |
| Research | Human or Human & AI, Intensive | Low → Medium — exploring | Low → Medium |
| Specification | Human & AI, Engaged | Medium — path clearer | Medium → High |
| Implementation | AI or Human & AI, Engaged | Medium → High — executing | High (spec is written) |
| Review / Testing | Human & AI, Engaged | High — validating | High (testing against spec) |
| Complete | Final assessment | High — retrospective | Evaluated (was the spec good enough?) |
Don't update constantly — reassess at these specific points. Skip updates for minor progress, routine implementation steps, and small issues quickly resolved.
Research complete
Work mode, size, confidence, spec quality
Discoveries change the picture
Spec approved
Involvement, confidence
Scope is now defined
First milestone
Risk, confidence
Reality validates or challenges plan
Issues discovered
Risk, confidence, size
New information
Phase transition
All dimensions
Natural checkpoint
Completion
Everything (retrospective)
Learning for future estimates
Assessment evolution is learning, not error. Frame changes honestly:
Complexity dropped
“Research found a simpler approach”
Complexity increased
“Deeper analysis revealed details to address”
Risk increased
“Testing surfaced considerations we’re now handling”
Confidence improved
“We have a clearer picture now”
Scope grew
“As we mapped this, we identified related needs”