Image: Cursor
Why it mattersTeams that bought a code-review bot get the bot the vendor chose. The Agent SDK moves that choice to the team. A reviewer that enforces your own rules replaces one that enforces someone else's.
A team that bought a code-review bot got the bot the vendor chose. Cursor released the Agent SDK on 20 October to move that choice to the team. Cursor calls it generally available. The post reached the front page of Hacker News with 73 points at the time of writing.
The Agent SDK is the framework Cursor uses to build its own managed agents, including Bugbot, Security Reviewer and Rollouts. The company says teams can now use the same framework for their own work. An agent is defined as a small software project. Prompts sit in Markdown files. Tools are TypeScript functions. Connections to outside services live in code too. Changes to any of these go through the same version control and review as the rest of the code.
What one command deploys
A developer runs one command to deploy to Cursor's infrastructure. The command builds the agent, gives it a URL and starts it. Secrets and integrations go through the dashboard. Logs, evals and the local playground stay available. Teams that prefer their own infrastructure can run an agent as a single Node process in a container.
Cursor says the SDK includes evals and event replay from the start. A failing agent run can be replayed step by step. Evals catch a regression before it reaches production. For complicated jobs, an agent can hand a task to another agent with its own tools and sessions. It can also split the work across subagents that run in parallel.
One worked example and one named customer
Cursor describes an internal team that wanted every pull request checked against a growing knowledge base of past lessons. Using the Agent SDK, the team built a reviewer called kb-reviewer in about a day. The agent now runs on every pull request in that team's main code. It flags changes that contradict a recorded lesson. It cites the note behind each finding. It then asks the engineer to reconcile the discrepancy.
The PGA Tour is Cursor's named outside user. Michael Gyarmathy, Principal Engineer at PGA America, is quoted in the post. He says the SDK runs every part of what he calls the team's software factory. He says the built-in evals let the team iterate on new ideas and keep existing capabilities reliable.
A code-review agent is Cursor's own first example. Each team has its own standards for what deserves attention and who should review it. An agent built with the SDK can respond to a pull request event from GitHub, GitLab or Bitbucket. It can judge the risk of a change. It can route the change to the right team. It can approve the ones that match the team's policy.
Source
- Introducing the Agent SDK, Cursor
- Hacker News thread, 73 points
This item was written by an AI system from the linked source. Reveneau is responsible for what it publishes.
Get AI News in your inbox
New developer tools, model and agent releases, and how teams are actually using them to release software. Short, and only when there is something worth reading.