- Published on
Quoting Addy Osmani, Morgan Westlee Lunt, and Qing Wang
- Authors
- Name
- DeepHorus
In a 16 October 2026 post, Addy Osmani, Morgan Westlee Lunt, and Qing Wang of the Claude Code team argued that “did a human read the code?” is an imprecise proxy for risk. The better question, they wrote, is what could go wrong with a change and which check—automated or human—should catch it. Their comments describe how code review is being reorganized around Claude.
Previously, review was a gate at the end of the road. Engineers wrote code, opened a pull request, and waited for a colleague. That model concentrated scrutiny at one late stage and made human attention a bottleneck.
Now review runs continuously, the authors said, more like spellcheck while writing. Most changes have been reviewed and fixed by Claude at least once before a human is asked to look. This catches routine mistakes early and saves team attention for higher-level questions.
The shift matters because it changes both timing and purpose. Automated review is not simply replacing humans; it is triaging work. Routine errors are handled before human review begins, allowing reviewers to focus on design, risk, and context.
The authors frame the change around precision. A human may be the right check for some changes. For others, an automated pass may be faster and more consistent.
The Claude Code team’s account reflects broader debates about AI in software engineering. It suggests review is becoming a continuous layer rather than a final gate. That could reduce delays and improve consistency, but it could also require teams to redefine human oversight.
For now, the authors present continuous review as a practical workflow. Claude performs an initial pass, and humans handle higher-level scrutiny. The approach aims to match checks to risks, and its success will depend on how well teams calibrate that division of labor.
