Recurring activities: define, delegate, build and verify
DEFINE, DELEGATE, BUILD and VERIFY are recurring activities within the workflow: clarify the intended result, authorize bounded work, build within scope and evaluate evidence. They describe how the human directs agent work, not another workflow, separate methodology level or four additional phases.
AI Architect Mastery's AI Driven Development Methodology has one canonical workflow: PRD → PLAN → TASK → IMPLEMENTATION. The Human as Architect, AI as Builder principle keeps responsibility human-owned throughout it.
How the actions relate to the Core
- DEFINE
- Establish intent, constraints and acceptance in the PRD; shape a reviewable approach in the PLAN.
- DELEGATE
- Authorize a TASK with responsibility, selected context, scope, permissions and expected evidence.
- BUILD
- Execute the bounded assignment within IMPLEMENTATION, recording what changed and what was observed.
- VERIFY
- Inspect the change and test the claim. A human checkpoint decides whether the result is acceptable or needs further work.
Why this view matters
These activities describe decisions that must not disappear inside an agent session. They can recur as the work develops; they do not rename the workflow phases one-to-one. Verification and review obligations are already part of IMPLEMENTATION and its acceptance model.
Constructed example: an export boundary
Define that only the record owner may export. Delegate the authorization change with the existing interface preserved. Build the narrow correction. Verify allowed and denied requests, then let the human evaluate whether the evidence satisfies the requirement.
When the result reveals a wrong assumption
VERIFY is not an instruction to keep patching until a test turns green. If the acceptance expectation is missing, return to the appropriate earlier artifact. A passing local test cannot validate an undefined requirement.