Developer catalogs seven recurring limits after two months of AI code review

On 2026-11-12 at 23:00, the AI section of DEV Community published a retrospective written by a developer, titled "After Two Months of Running AI Code Review, I Found 7 Recurring Limits." The author built a custom multi-Agent pipeline and ran AI code review on real pull requests for two months, then summarized seven recurring weaknesses: unverified fixes, weaker reviews on large PRs, anchoring on detailed specs, softer repeat passes, instruction overload, inconsistent rule enforcement, and missing intent.

On the question of why these limits appear, the author notes that existing research on code generation and long-context LLMs can explain only part of the pattern and does not account for the full pipeline. The piece also observes that AI review still feels like a creature of mystery: on one side it catches bugs, while on the other it can make things worse and hand the author work that has to be redone.

At this stage, the retrospective stops at the seven limit categories and at the observation that prior research falls short of covering the end-to-end flow. No follow-up developments, responses from tool vendors, or independent confirmations are included in the available material, so the account currently rests on this single developer's two-month experience with the self-built pipeline.