Automation complacency
Automation complacency is the drop in attention people pay to a task once an automated aid usually gets it right, so the aid's misses go unnoticed; in code review it is skimming a pull request because CI or an AI reviewer has already passed it.
Why it matters
The term comes from human factors research on pilots and operators. Parasuraman and Manzey's 2010 review in Human Factors found that complacency appears when people have several tasks competing for attention, affects novices and experts alike, and is not overcome by simple practice. A reviewer with a queue of pull requests and a green check from a bot is in exactly that situation.
An AI reviewer that catches most bugs makes the remaining ones harder to see, because the reviewer stops expecting them. On DiffDojo's own benchmark, AI models caught between 86% and 98% of planted bugs overall (September 2026), and independent 2026 benchmarks put GitHub Copilot's recall on real bugs at roughly 25% to 34%. Whatever the number, the bugs the bot misses are the ones a complacent human approves.
Sources
- Parasuraman and Manzey (2010), Human Factors 52(3)
- DiffDojo living benchmark
- Copilot recall in 2026 benchmarks
See also
- Rubber-stamping
- Review recall
- Worked example: Hallucinated API: axios has no retries option, so nothing ever retries in Node, an AI-written change that looks configured and does nothing