Stanford's CS146S, a ten-week course on modern software development, changed its assignments after its first run showed AI coding agents could complete much of the original work in a single prompt. Mihail Eric, who teaches the course and leads AI at Monaco, said the first run had nine assignments and a capstone. In one exercise, students ran an AI code reviewer on a repository and evaluated its feedback. By the next run, he found students could one-shot many of those exercises, and when he asked a room of sophomores and juniors how they coded, everyone he surveyed already used an agent. The central change is that part of the coursework now requires students to contribute a patch to an open-source project, which the project's maintainer reviews. Eric argues this shifts the skill from producing code to judging whether a change belongs in the project, understanding why existing code behaves as it does, and anticipating what else the change might break. At Monaco, the engineering team evaluated AI code-review products and kept using Devin, but built its own workspaces where background agents can run the product, reproduce failures and test patches. Eric's rule of thumb, as quoted in the episode, is “You buy to learn, and you build to keep.” The company's internal system, Hal, lets engineers ask it to investigate a failure and open a pull request, and lets sales staff ask how the product works. The show notes cite a Faros study finding that teams with high AI adoption merged 98% more pull requests while their review time was 91% longer. For Australian businesses, the episode changes little directly. It announces no new regulation, and the costs it describes depend on tooling choices it does not specify. The practical lesson is about review capacity: faster agent-written output can add to review queues, so teams adopting coding agents should plan for that workload. If agents are given access to production systems or customer data, as Monaco's workspaces are, obligations under the Privacy Act 1988 may apply to how personal information is handled, so access controls and logging deserve attention. The one idea worth trying now is the explain-diff exercise the episode describes, which turns an agent-written change into a walkthrough and quizzes the author on it before merge. For businesses without software teams, there is nothing here to act on.