A recent critique challenges the notion that coding agents—autonomous LLM-based systems capable of reading, writing, and repairing software—can entirely replace human code review in the software development pipeline.
Code Review Is More Than Just Automated Detection, Argues Critique of AI Substitution Claim
The argument against full automation rests on the premise that traditional code review provides value beyond simple defect detection or style enforcement. Critics argue that human reviewers perform several unique functions that current AI agents struggle to replicate:
- Assessment of Comprehensibility and Intent: A human reviewer can signal when code is too complex or its intent is unclear. While an LLM can process the code, it may fail to signal "legitimate human incomprehension."
- Identifying Absence: Human experts can detect what is missing, such as missing error handling in a changed API contract—a task where LLMs are prone to "absence blindness."
- Contextual Knowledge: Reviewers bring operational context from outside the repository, such as recent service incidents, upcoming dependency deprecations, or legal requirements.
- Social and Governance Roles: Code review serves as a coordination and sensemaking process. Human reviewers bring "skin in the game," providing a level of personal accountability and social coordination that an agent cannot emulate.
The critique concludes that viewing code review solely as a detection process ignores its fundamental role in governance and human-centric adaptation to unplanned circumstances.
Sources
- There is more to code review than (automatable) detection （Hacker News Frontpage, 2026-09-26）