Role Split AI Coding Workflow Checklist

A role split AI coding workflow assigns different tools to different parts of the software task. The point is not to collect models. The point is to reduce mismatch: one tool for repository edits, one for architecture risk, one for bulk transformation, one for source checking, and one for fast drafts.

Start with the task class. Is the work a code change, explanation, test repair, migration plan, log triage, documentation update, UI copy pass, or source-backed answer? If the task is a real repository edit, use a tool that can inspect files, run commands, and verify changes. If the task is conceptual, use a reasoning model and require a short decision record.

Define the handoff contract. Every route should return the same minimum fields: what changed, what evidence was checked, what remains uncertain, and how to verify. Without a shared contract, role split becomes a pile of disconnected outputs.

Keep one default path for small tasks. A role split workflow should not force routing for every typo or tiny question. Use the default until a trigger appears: large context, high cost, private data, failing tests, external sources, or production risk.

Review the routing table monthly. Models change, prices change, and failure modes change. A tool that was best for code search last month may be less useful after the repository gains better tests or docs. The workflow should evolve from measured task outcomes, not model hype.

Role Split AI Coding Workflow Checklist

// COMMENTS

ON THIS PAGE