Guarding critical thinking in review

Beyond Vibe Coding with Addy Osmani

Topic 9 of 11

Published Oct 29, 2025 · Clip 48:09–56:07 (7:58)

The host confesses he accepts AI suggestions more readily over time and worries the same complacency could hit code reviews ('LGTM'). Addy suggests deliberately not using AI on certain days or features to preserve critical thinking, warning you'll still need to solve problems yourself when the model fails repeatedly. He treats agent code-review approvals as signals, not replacements for human judgment, and stays intentional about solving some tasks without AI. They discuss Claude Code's new explanatory and learning modes, and Addy advises newcomers to use the LLM to understand the codebase before prompting features into existence. The host notes companies seeing faster onboarding with AI as a 'trusted person 24/7', and Addy hopes AI-as-a-learning-tool becomes standard for concepts, frameworks, and porting features across codebases.