Over the past year, writing code became almost effortless.
You give an AI agent a prompt, it writes 300 lines across four files, and you open a pull request five minutes later.
But here’s the problem nobody warned us about: reviewing code is now the biggest bottleneck in engineering.
When a human writes 300 lines, they usually know:
- Why they chose that data structure.
- What edge cases broke during manual testing.
- Why line 42 has a weird check.
When an AI generates 300 lines, the author often didn't write it, didn't step through it in a debugger, and just verified that the happy path works.
That pushes the entire cognitive burden onto the reviewer:
- The code looks clean and well-formatted, which creates a false sense of trust.
- Subtle bugs, concurrency issues, and stale assumptions get buried under layers of polite comments.
- PR queues are twice as long because dumping code is free, but reading code still takes real human brain cycles.
Speeding up code generation without speeding up our ability to verify it doesn't make teams faster—it just turns senior engineers into full-time human linters.
How is your team handling the flood of AI-generated PRs?
