16:35Y Combinator
A concrete senior-engineering walkthrough rather than a panel: Holtz compares Claude Code and Codex, orchestrates several coding agents in parallel, and marks "slop-free zones" where generated code is not allowed to land unreviewed. The setup embodies the article's rules — repository context, small reviewable patches, and the human staying architect while the agents do the typing.
What you should get from this: Set up a multi-agent coding workflow with explicit review boundaries so generated changes stay small, tested and owned.
Watch or know first: Experience reviewing pull requests and running at least one AI coding assistant against a real repository.
AI Expert note: Holtz builds Conductor, one of the tools shown, so treat the tool choices as one practitioner's stack rather than a ranking. The durable part is the discipline: parallel agents, human architecture ownership and zones where slop cannot land.