LoginSign Up
← Back to all sparks
C

Cursor

INFRA · APIS
Velocity6.3

Cursor is becoming a control plane for agent fleets — routed, governed, and driven from a phone.

agent-fleet-managementmodel-routingmobile-agentsenterprise-governancecode-review-automationmcp-distribution
◆Current state
Cursor has stopped shipping editor features and started shipping fleet management. The last two months added cloud VMs that run agents independently of the laptop, an iOS and now iPad client that launches and merges work from anywhere, Slack as a first-class agent entry point, and Bugbot as an automated reviewer in the pull-request path. Underneath all of it, Cursor Router now decides which model handles each request, with admins setting policy per team.
◆Where it's heading
Every surface added this quarter points the same way: the agent is the unit of work, and the IDE is only one of several places you supervise it from. The admin-facing releases — team marketplaces, distributed MCP servers, org-group access, router policy and model block lists — are assembling an enterprise control plane around that fleet. Cursor is competing less on editing quality and more on whether an organization can run, route, price, and audit many agents at once.
◆Prediction
The entries name their own next step: Bugbot's /review is explicitly marked "coming soon" to the CLI, and Cursor Router already spans desktop, web, iOS, CLI, and SDK — so expect review and routing to reach parity across every surface next. Finer-grained router policy, scoped per repo or per task rather than per team, is the natural follow-on to the admin controls just shipped.

◆Recent moves

  1. 14d ago

    Cursor for iPad adds full PR review and an agent inbox

    The mobile client picks up a second form factor and, more importantly, the rest of the review loop: comments, checks, approvals, and reviewer changes now happen on the device. Paired with the Inbox, this closes the gap the June iOS beta left open — supervising agents no longer means returning to a desktop. Bitbucket and Azure DevOps support quietly widens which teams can use any of it.

    View source ↗
  2. 21d ago

    Auto mode moves to Cursor Router with cost/intelligence modes

    ⚡ SPARK

    Model selection stops being a user preference and becomes a routed decision with a price dial attached. This is the commercial spine under the agent-fleet direction: once teams run many agents at once, per-request routing is what makes the volume affordable, and the admin controls make it governable.

    View source ↗
  3. 26d ago

    Cursor in Slack plans first and works across multiple repos

    Slack becomes a more serious agent surface rather than a notification relay: the agent states a plan before acting, reports status as it goes, and can target a named multi-repo environment instead of one default repository. Cross-channel reading and posting is the first sign of agents operating across a workspace rather than inside a single thread.

    View source ↗
  4. 1mo ago

    Side chats and searchable agent transcripts

    Side chats let a developer investigate a tangent without derailing a running agent, and each one is a durable conversation that can be pulled back into the main thread. Transcript search over a local index is the quieter half — it treats accumulated agent conversations as a corpus worth querying, which only matters once you are running many of them.

    View source ↗
  5. 1mo ago

    Team MCP servers distribute through team marketplaces

    Admins configure MCP servers once and push them to cloud agents, the agents window, the IDE, and the CLI, so members install approved integrations without touching server config. Restricting marketplaces to organization groups extends the same governance pattern Cursor has been applying to plugins and router policy.

    View source ↗
  6. 1mo ago

    Cloud agents come to iOS with remote control

    ⚡ SPARK

    This is the release that detached the agent from the workstation, and everything since — the iPad layout, the full review surface, the Inbox — has been filling in what a phone-first supervisor needs. Cloud VMs doing the execution is what makes the phone viable as a control surface rather than a viewer.

    View source ↗