My AI‑written plan went through every check I could think of. The gaps were in what I never said.

Description and use

An AI‑generated migration plan created through a multi‑agent workflow that passed 51 checks and resolved all findings.

Problem

Generate a comprehensive, accurate architecture plan for a large production system change using AI agents, while ensuring all requirements, code references, and potential gaps are identified and addressed.

Outcome

A detailed, AI‑generated migration plan that passed 51 cross‑checks, with all findings addressed and a transparent record of revisions.

Workflow

• 1. Ask Claude Code which model should write the plan; choose costly model for planning, regular model for orchestration and review.
• 2. Create a brief that only trusts executable code, records exact user input, and includes a read‑back for clarification.
• 3. Run a pipeline: brief check (2 checkers), architects (3 costly model agents from different angles), judges (2 costly model agents), synthesis (costly model), cross‑check (3 regular model checkers), revision (costly model).
• 4. Address truncation by rebuilding from transcript; produce 13 milestones and 32 open questions.
• 5. Perform second pass: provide costly model with missing requirements, have it reassess and rewrite plan if needed.
• 6. Final revision and documentation of findings with reasons.

Evidence

The plan was 22,000 words long. Twelve agents had worked on it for close to two hours.

Source

It added one rule I would have forgotten: the plan's author should not review its own plan.

Source

The judges spot‑checked the architects' references. They ran 71 checks on file:line claims… 58 held exactly as written. The other 13 came back qualified…

Source