Planner Architect
§3 · Roles
Three human roles, one continuous validation function.
Planner, Builder, Reviewer, with PAT running alongside. These are responsibilities, not job titles. Their clarity is what makes continuous flow possible. Certifications determine who can take a task; teams choose who does, with per-artifact separation of duties.
Builder Executor
Flow driver. Supervises AI execution.
Reviewer Semantic Gatekeeper
Final semantic authority before release.
PAT Continuous Assurance
Async system-level validation.
Planner
Architect · §3.1
The semantic entry point of the pipeline.
Planners translate intent into constraints, scope, and domain-aligned direction. Their primary artifact is the Execution Plan, which guides all downstream AI and Builder activity.
Responsibilities
- →Interpret ambiguous intent and express it with crisp boundaries
- →Define architectural constraints, domain invariants, and integration surfaces
- →Collaborate with AI to produce the Execution Plan
- →Clarify what "good" looks like without dictating mechanical implementation
- →Ensure work fits cleanly within the domain without cross-team leakage
Does not
- ×Granular task breakdown
- ×Write scaffolding or boilerplate code
- ×Dictate step-by-step sequence of mechanical work
Human dimension A thinking job, not a production job. It draws engineers who enjoy systems reasoning, architectural judgment, and shaping the long-term coherence of a domain. AI doesn't replace the Planner. It makes the Planner more central, since their judgment governs the system.
Builder
Executor · §3.2
Owns execution. Supervises agentic AI.
Builders supervise AI as it generates code, tests, migrations, refactors, and artifacts aligned with the Execution Plan. They ensure fidelity to constraints and resolve ambiguity without slowing flow.
Responsibilities
- →Drive AI generation of code, tests, and system updates
- →Maintain scope and constraint alignment with the Execution Plan
- →Perform first-pass validation before sending work to review
- →Request clarification from the Planner when ambiguity appears
- →Keep the pipeline unblocked
Does not
- ×Make architectural decisions
- ×Expand the problem beyond defined scope
- ×Perform semantic approval
Human dimension Work shifts from mechanical typing to high-leverage orchestration. Builders apply engineering intuition, debugging skill, and contextual insight to keep AI output aligned. This role fits engineers who enjoy rapid iteration and continuous movement.
Reviewer
Semantic Gatekeeper · §3.3
The final semantic authority before release.
Reviewers make sure output matches intent, respects domain invariants, and is safe to ship. They are the only human at the merge boundary.
Responsibilities
- →Perform semantic review (meaning, correctness, alignment)
- →Validate architectural coherence and invariant integrity
- →Approve, reject, or request clarification
- →Oversee release flow across rings (Ring 0 → Ring 1 → general)
- →Trigger severity responses when issues surface (Sev1–Sev5)
Does not
- ×Rewrite code manually
- ×Perform mechanical / syntax / lint review
- ×Expand scope or re-plan
Human dimension Reviewers carry the highest cognitive load on the team. Their work looks more like editorial curation or architectural judgment than classical code review. It's a role defined by trust, not throughput.
PAT
Product Acceptance Testing · §3.4
Continuous, asynchronous system validation.
PAT operates outside the Planner-Builder-Reviewer loop. It checks correctness at the system level, not the artifact level.
Responsibilities
- →Exploratory testing probing domain and system-level edge cases
- →Regression verification and invariant stress-testing
- →Monitoring and surfacing defects from production rings
- →Certifying the system under evolving conditions
- →Feeding insights back into Planner / Reviewer judgment
Does not
- ×Gate releases or hold batches for sign-off
- ×Replace Reviewer judgment
- ×Stop flow except via Sev5
Human dimension Exploratory testing needs curiosity, intuition, and an eye for the unexpected. AI handles the permutations, but humans decide where exploration should go. Humans test ideas. AI tests permutations.
§3.5 · The GI Team
A fully formed GI team
A team owns a domain and is accountable for continuous flow, semantic clarity, and system integrity. It runs autonomously and interfaces with other teams only through explicit contracts and invariants.
A balanced GI team per domain
Reviewer capacity is the throughput ceiling. Sizing follows ratios, not fixed counts.
→ Reviewer capacity ≥ Builder capacity
→ ≥ 2 Planner-eligible per Domain
→ ~1 PAT per 3–5 active Builder seats
Planner
Architect
Senior · Staff · Principal
Executors
Builders
Mid-level to senior
Semantic gate
Reviewers
Senior ICs
Continuous assurance
PAT
Hybrid QA / engineering
Apprenticeship
Junior
Pairing + PAT participation
Role names describe modes of work, not job titles. A certified engineer can pick up Builder or Reviewer tasks on Tracks they hold the right cert for, subject to per-artifact separation of duties.