A trust deficit is the gap between how quickly AI produces code and how quickly engineers can verify that the code deserves to exist. When generation outpaces evaluation, unverified code accumulates risk rather than eliminating work. section 8.1.1
When evaluation becomes the bottleneck
A developer who once produced 200 meaningful lines per day can generate 200 lines in an hour with AI. A reviewer may still thoughtfully examine only 400 lines per hour, leaving as many as 1,600 generated lines per day competing for limited evaluation capacity. section 8.1.1
Manage the deficit by strengthening verification rather than reviewing faster or trusting the generator more:
- Use tests designed around edge cases, specifications, properties, contracts, and intended behavior rather than relying on coverage alone. section 8.9
- Combine mutation testing, characterization testing, formal methods, and runtime observability to expose weak tests, collateral damage, and production failures. section 8.9
- Match verification investment to system criticality; regeneration may cost less than debugging, while QA becomes a core engineering competency. section 8.9
