Maya: Hey builders! Here's what's on the radar today on Cursor HQ. Gemini 4 Argon landed in Cursor, Origin opened its app marketplace, Grok Bot hands code changes to cloud agents, Cursor Rules steer the agent, and Projects coordinate sub-agents. I'm Maya.

James: And I'm James. Let's get into it.

Maya: So Cursor announced on October thirteenth that Gemini 4 Argon, a Google model, is now available in Cursor. Cursor says it scores fifty-two point eight percent on CursorBench at default effort, which it describes as on par with Anthropic's Claude Opus five point five.

James: That comparison is the headline, Maya, but here's what matters for budgeting. Pricing is two dollars per million input tokens and ten dollars per million output tokens. Input is your prompt and code context, output is what the model writes back.

Maya: Exactly right, James, and one honest caveat. CursorBench is Cursor's own benchmark, and that figure comes from Cursor's announcement. They did not say how the two models compare on cost or speed, and they did not describe how to select the model.

James: Next up, Cursor's Origin, its code hosting platform, now integrates with more than twenty apps, and the marketplace opened to public submissions on October thirteenth. Per the changelog, any team building developer tools can now publish an app.

Maya: The launch apps cover six areas, James. Deployments like Vercel and Netlify, CI checks like CircleCI, code review including CodeRabbit, security with Semgrep and Socket, Sentry for monitoring, and Linear for project management. You connect your repositories to tools you already use.

James: That's the pitch, and the publishing path is fixed. You create an app in your namespace, build against the Origin API with an app JWT and installation tokens, receive webhooks, then make it public and submit with a slug, tags and a demo video.

Maya: Right, and Cursor reviews each submission before listing it. CodeRabbit says it is a launch partner, bringing reviews directly to pull requests hosted on Origin. Cursor also points builders to its publishing guide and the Origin API reference.

James: Now this one is a genuinely different loop. Grok Bot, a chat assistant, connects to your Cursor account and passes coding requests to Cursor Cloud Agents. The agent writes the change in your GitHub repository and opens a pull request for your approval.

Maya: And that pull request includes a live preview link, James, with nothing merged until the user approves. The whole thing runs without an open laptop. Users describing the setup say the bots handle the workflow and the human is left with review.

James: Supabase published a worked example of the same chain. Cursor runs against the Supabase backend through the Supabase MCP server, so it can query the database, inspect the schema, read logs and deploy Edge Functions. Supabase says it is joining the SpaceXAI Partner Network.

Maya: And because requests come from people not reading the code, Supabase suggests a Cursor rule that runs its get_advisors scan on every schema change. It flags tables missing Row Level Security or unindexed foreign keys, so a comments feature ships with RLS enabled.

James: That rule tip is a perfect bridge, Maya, because Cursor Rules are the whole next story. A rule is a set of instructions for the agent, and when you apply it, Cursor adds it at the start of the context.

Maya: Exactly, and the context is everything the AI sees when it starts, so a rule is among the first things it reads before it acts. A rule can explain the project, the style you want, or the architecture you want to follow.

James: And you can scope them, Maya. Write a rule for the frontend, the part people see and click, and a separate rule for the backend, the part behind the scenes that stores data. The material covers the idea, so check Cursor's rules documentation for exact steps.

Maya: Last one. Cursor's Projects feature replaces a pile of chat windows with one workspace for a feature or a full app. You state the goal, a coordinator agent plans the tasks and assigns them to sub-agents, much like a project manager.

James: To create one you find Projects, click plus, name it, link a GitHub repository and choose a model. The agents run in the cloud, so several run in parallel and shutting down your computer does not stop the work. They share the Project's files too.

Maya: And a Project can listen to Slack, pull requests and scheduled tasks, dispatching work automatically. If you want to skip the coordinator, right-click and choose New Agent to run a task directly, and that agent still shares the Project's context.

Maya: That's your Cursor HQ daily brief. Check cursorhq.ai for the full stories. Keep building.