How to read code

Sean Goedecke, a software engineer, argues that code written by AI agents still has to be read by people, and shows how to read a large diff in quick passes.

Sean Goedecke is a software engineer who writes essays about engineering practice. He argues that code is hard to read from front to back because it is ordered for the computer, usually read as a set of changes, and structurally complex. He describes a faster way to read it and rejects the claim that AI agents or AI reviewers remove the need to read code.

• He reads a large diff out of order, first tracing an important path through the functions. He then follows how each function or piece of data is used, and only then reads end to end.
• In one case, an AI agent turned a small change into a three-thousand-line diff to fix a race condition that was harmless by design.
• He says such errors are rarely bugs: the code does what the agent intended, but it does not match the company's technical values.
• He says an AI reviewer cannot replace a human reader for the same reason, though it can help if its output is also read carefully.

The article offers engineers a method for reviewing agent-written changes, not a rule about who reviews them. Treating a diff that grows far beyond its task as a reason for a closer human read is an inference and is not the author's claim.

Derived by Working Surface from the article; more in the Playbook. Source line: I routinely find massive errors in AI-generated code.

8 October 2026: as AI agents produce code and prototypes faster than people can check them, teams must decide where human checks and decisions sit instead of letting review quietly lapse.