Review is the new bottleneck
balancing loop · 4 steps · Trace on the map →
The cycle
- Agents write most new codeAI writes most new code
- More changes to reviewCode review at scale
- Reviewers become the constraint
- Shipping slows to review speedCoding agents as products
What drives it
Code output scales with compute, but review still scales with people, so software is the first place the verification bottleneck bites.
What breaks it
AI review and stronger automated tests relieve it, if teams trust them.
What to watch
Review time per change. Share of AI-written changes merged without human review.