coding-agents

Security teams blocked Cursor, so agents move on-prem

Promtime

coding-agents

Coder has moved its Coder Agent Relay service into private preview with SpaceXAI as launch partner, aimed at the dozens of customers whose security teams had blocked Cursor outright. Coder CEO Rob Whiteley described the deployment problem and the usage imbalance behind it in an interview with The New Stack.

At a glance

  • Each Coder workspace can start a Cursor worker that opens an outbound connection to Cursor, keeping tool execution and the repository checkout on infrastructure the customer controls.
  • Whiteley says 1% of Coder's engineers account for 40% of its token spend, which he frames as uneven adoption rather than overspending, and calls token dollars the wrong productivity metric.
  • Banking, life sciences, defense, aerospace and government teams gain compliance reporting from per-run logs of what an agent accessed, executed, changed and was blocked from doing, without reconstruction by hand.

The gap Whiteley describes is not about budget. Coding agents have concentrated in the hands of a small group of engineers, and in regulated sectors that concentration is enforced from the outside by security policy rather than by skill. Agent Relay reads as an attempt to convert a blanket ban into a configuration choice, which likely matters more to Coder's addressable market than to any individual developer's workflow.

Cursor keeps inference and planning while tool calls run on customer machines

Cursor's acquisition by SpaceX closed on August 14, and its Cloud Agents can now run inside Coder workspaces on a regulated organization's own infrastructure. Cursor has supported self-hosted agents since March, and Whiteley says many customers were already wiring the two products together themselves.

Cursor continues to run the agent loop, including inference and planning, while tool calls execute in Coder environments on the customer's network, so source code, secrets and internal services stay on machines the organization controls. Relevant code and tool output are sent to Cursor for cloud-hosted reasoning, and model selection and inference remain with the cloud provider.

Platform teams provision and scale those workspaces the way they already manage developer environments. Agent environments are sandboxed, ephemeral and scoped to a single task, and a prompt injection pushing an agent toward unauthorized resources is blocked at the environment layer rather than left to the model to refuse.

Dozens of customers blocked Cursor over an architectural mismatch

Whiteley says the common case is the worst one. "The worst-case is the common case: blocking Cursor outright," he told The New Stack, adding that dozens of Coder customers have security teams that did so. He attributes the bans to an architectural mismatch rather than a flaw in Cursor, saying such enterprises need complete control over the environment in which an agent runs.

Whiteley describes teams that adopted Cursor, saw productivity gains, then lost it when adoption reached the security team. He says some of those customers reversed the policy with a mandate allowing Cursor as long as agents run in Coder, sparing developers the alternative of being told to use Citrix.

1% of Coder's engineers account for 40% of its token spend

The imbalance Whiteley points to inside his own company is the argument he makes for the partnership. He says the issue is not the amount spent but uneven adoption, and a widening gap between agentic haves and have-nots, which the industry answered with what he calls the ill-fated tokenmaxxing fad.

At Coder, 1% of my engineers are responsible for 40% of my token spend. The problem isn't how much they're spending; it's the uneven adoption of the technology. There's a widening gap between the (agentic) haves and have-nots.

Whiteley calls encouraging engineers to maximize AI usage the right sentiment with the wrong metric, and describes dollars spent on tokens as a red herring that can be gamified. He says both companies independently arrived at the need for closer integration, and that the customer should never be the systems integrator.

Who gets Agent Relay first

Coder Agent Relay for Cursor is in private preview with design partners, and no general availability date or pricing has been announced. SpaceXAI is the first of several partnerships Coder says it is building around the argument that the best AI coding tools should run inside any organization, on any infrastructure, without giving up control. The other partners have not been named.

Comments

No comments yet. Be the first.

Join the conversation

Sign in with Google to leave a comment. Your name and avatar come from your Google profile, and the comment appears after moderation.

We only use your name and avatar from Google. We never store your email address.