Coding Agent Code Review Loop
Overview
A workflow for letting agents produce code while preserving reviewer comprehension, test evidence, rollback paths, and issue-specific context.
Conference Context
- Software-factory and coding-agent sessions frame review as a loop across prompt, diff, tests, trace, and user impact.
- Several talks separate code generation from the discipline needed to decide whether generated code should land.
- The conference graph links coding agents to context engines, PR analysis, quality gates, and team adoption.
Implementation Pattern
- Start with a task contract that names files, tests, owner, and done criteria.
- Have the agent produce a diff plus a short evidence bundle rather than only a prose summary.
- Run deterministic checks and an LLM review against policy-specific failure modes.
- Ask the human reviewer to inspect the riskiest changed behavior, not every generated token.
- Record what passed, what failed, and which reviewer decision changed the policy.
Evidence
- coding agents - Topic synthesis
- software factories - Topic synthesis
- what makes a codebase agent ready - Question layer
- 2026 06 29 daksh gupta what we learned by analyzing 1m ai generated prs - Official schedule
- 2026 06 29 itamar friedman the last human code review building trust in ai generated code - Official schedule
- 2026 06 30 alex volkov the z l continuum should ai engineers still read code - Official schedule
Evidence Boundary
This is a reusable workflow synthesized from the linked conference evidence. Treat it as a recommended implementation pattern, not as a direct quote from any single talk.