Theo Browne separates the importance of code from the percentage of AI-generated lines a developer reads. Production-critical changes still need careful verification, while temporary experiments, debugging tools and alternative implementations can be discarded after answering a specific question.
Theo Browne illustrates this distinction with hypothetical code-volume ratios, custom lints, load tests and small tools built around a product. These are examples of how he uses agents, not measured evidence that larger generated codebases are automatically safer or more productive.
Theo Browne emphasizes API and SDK contracts as important review points and describes using weaker agents to expose confusing interfaces. The argument is about choosing the right scrutiny for each artifact, not weakening production review or treating AI-generated tests as inherently correct.
Watch on YouTube
