Lectures

Tommy finds a bug in his published Snake game that the agent who assembled it himself cannot see. Alex explains: the one who built the thing cannot see it from the outside — you need someone completely fresh.

Alex untangles the confusion of three words: the main agent (Claude Code itself) and the subagent — a separate worker with a fresh head of its own, who lives in a file and is spun up for a narrow task.

Code review is separate work, not finishing up. Alex explains why another person’s eye sees better: it is not about smarts, but about a clean context, not clogged by your own build.

What goes into the subagent file, how to check that the worker is up, and how to delegate a task to it in one go. And there is also a bonus: you can give the agent a task by voice.

Tommy goes through the whole path himself from start to finish: he adds the subagent role, delegates the review, gets the list of findings with the bug first, chooses what to take on, adds his own changes, and turns in the result.

Three level articles in one list: how code review works and how to hand it to an agent, why a subagent gets its own context window and read-only access, and how to brief a reviewer and then check the report.