Skip to content

Human Review Requirements

When to Use

When you need to understand what "human review" actually means for AI-generated code — the minimum standard before you can honestly say you've reviewed the code.

Decision: What Review IS vs IS NOT

Human Review IS Human Review IS NOT
Understanding the problem being solved Running the code and seeing if it works
Reading every line and knowing why it's there Scanning code quickly for obvious errors
Verifying API calls against Drupal documentation Trusting AI's assertion that "this is correct"
Testing edge cases and error conditions Running it once with test data
Questioning whether the approach is simplest Accepting the first solution that works
Being able to defend every technical choice Saying "the AI chose this approach"

Decision: Review Depth by Disclosure Level

Disclosure Level Minimum Review Standard
AI Assisted Code Read all code, verify API usage, run tests, ensure you understand everything
AI Generated Code All of the above + verify every API call exists, test edge cases thoroughly, question the approach
Vibe Coded All of the above + consider rewriting from scratch with understanding. If you can't review at the "AI Generated" level, don't submit.

Pattern: The "Senior Developer" Test

Ask yourself these questions about the code:

  1. "What does this code do?" — You should be able to explain in your own words, not the AI's words
  2. "Why this approach?" — You should know the alternatives and why this one was chosen
  3. "What happens if [input] is [edge case]?" — You should know the behavior for unexpected inputs
  4. "What would a malicious user try?" — You should have considered security implications
  5. "Is there a simpler way?" — The AI frequently over-engineers. Apply the "isn't CSS simpler?" test
  6. "What breaks if this code is wrong?" — You should understand the blast radius

If you can answer all six confidently, you've reviewed the code.

Common Mistakes

  • Confusing "it works" with "it's correct" — All three solutions in the laollita.es example worked. Only one was right. Working code can still create unnecessary technical debt and maintenance burden.
  • Trusting AI's confidence — AI presents solutions as "production ready" or "this is done, perfect!" It never says "I'm confused, maybe we should start over." You must be the skeptic.
  • Reviewing AI code less carefully than human code — AI code deserves MORE scrutiny because it may contain subtle issues that look correct but aren't (hallucinated APIs, deprecated patterns, unnecessary complexity)
  • Not starting fresh when confused — If you've been going back and forth with AI and things aren't making sense, start a new session. Context poisoning is real.

See Also