AI for developers: assistants, agents and the review that follows

A coding assistant that writes a function in four seconds is only useful if the function passes review, passes tests and doesn't quietly add a dependency nobody asked for. We judge developer tools on that standard.

This section covers code assistants, agent frameworks, model APIs, evaluation and cost. We look at what a tool can read in your repository, what it sends to a vendor, and how its changes show up in a diff you can actually review.

Agents get extra scrutiny. An agent that runs ten steps on its own can make ten mistakes on its own, so we always ask where the checkpoints are.

Next on the list: getting started with AI for developers, and beginner advice for teams adding an assistant to code review.