VORCL
← All courses
Intermediate · 7 lessons

Code Review in the AI Era: Trust, Verify, Merge

AI writes code faster than your team can read it. A course on reviewing diffs you did not write: a failure taxonomy of «almost right» code, security checks, agent reviewers, tests as the trust boundary, and a team auto-merge policy.

Start →
  1. 01Almost Right: a Failure Taxonomy of AI CodeWhy AI code compiles, passes tests, and still breaks production: confident API hallucinations, copy-pasted legacy patterns, and silent contract changes — with a checklist for each class.45 min
  2. 02Reviewing Diffs You Did Not WriteStrategy instead of line-by-line reading: three passes over the diff, reading from tests and contracts, red flags, and questions to the authoring agent instead of heroically reading two thousand lines.45 min
  3. 03Security Review of Generated CodeSQL injection from concatenation, missing permission checks, hardcoded secrets, and invented packages: where generated code is insecure by default and how to raise a SAST gate in CI.50 min
  4. 04Agents Reviewing AgentsAnother agent does the first review pass: structured findings, hallucination control for the reviewer, precision calibration, and a CI pipeline where humans only see escalations.45 min
  5. 05Tests as the Trust BoundaryWhy tests matter more than the implementation in an AI PR: tautological tests, the red-green proof, mutation testing, and what exactly the trust boundary locks in.45 min
  6. 06Team Policy: What Merges AutomaticallyChange classes by blast radius: what merges automatically on green CI, where a human is mandatory, how to codify the policy with CODEOWNERS and rulesets, and which metrics to steer it by.40 min
  7. 07Capstone: Triaging a Queue of Ten AI PRsThe final run: a queue of ten AI-authored PRs, ninety minutes, and a triage protocol — auto-merge, quick review, deep review, or reject and regenerate.60 min

Sources

Code Review in the AI Era: Trust, Verify, Merge · VORCL