Cursor introduces Agent SDK

Cursor has released an Agent SDK for building software agents as code. These agents can respond to events in a team’s workflow, such as a new pull request, and follow policies the team can inspect and change. Cursor says teams can deploy agents to Cursor’s infrastructure with one command, or run them on their own infrastructure. The announcement drew 64 points and 67 comments on Hacker News.

The practical appeal is control: an agent’s instructions, tools, service connections, and tests sit together in a software project, so changes can go through version control and review. Cursor gives the example of an internal agent that checks pull requests against the company’s recorded lessons. The company says its team built it in about a day, and that it now flags conflicts and cites the relevant notes. That’s Cursor’s example, not yet proof of what the SDK will deliver for other businesses. Its operational value still needs to be assessed.

For a business considering agents, the useful question is whether there’s a recurring workflow with clear rules and a way to check the results. This is the kind of thing we build into an AI command centre, where teams can keep track of the tools and workflows they’re putting to work. Cursor’s announcement also describes built-in evaluations and event replay, and says the SDK is generally available.