Skip to content
Today
Sections

Build Independently confirmed2 publishers 3 min read Published Updated

Coder registers your self-hosted workspaces as a compute pool in Cursor's scheduler

Agent Relay keeps Cursor's reasoning loop in Cursor's cloud and lands the tool calls in workspaces you already govern, which turns the security review's question from whether the agent is allowed into what its planner gets sent.

The Engineer · Build desk

Send a correction

Illustration accompanying Coder registers your self-hosted workspaces as a compute pool in Cursor's scheduler
Generated illustration

What happened

  • Coder launched Agent Relay, which runs cloud-hosted coding agents inside self-hosted Coder workspaces rather than the agent vendor's cloud, and named Cursor as its first integration partner.
  • The integration hangs off the agent provider's own compute pools, with Coder registering itself as one of them so agent work can be dispatched to infrastructure the customer already operates.
  • Coder's own assessment of 100 engineering organizations put 70 percent on infrastructure never designed for agents and only 31 percent at organization-wide AI governance.

Why it matters

  • What it costs cost The warm start is prebuilt capacity, so the platform team pays for workspaces sitting ready for a scheduler it does not operate to ask for them.
  • Who has to decide decision With the planner remote and the effectors local, a security review has to stop arguing about tool approval and start specifying what the outbound link is permitted to carry.
  • What it sets up precedent If the integration surface is the vendor's own compute-pool interface, any agent vendor exposing one becomes relayable, and governance is bought once instead of rebuilt for each new agent a team adopts.
  • What doesn't add up contradiction Coder's two posts credit different launch partners, so a buyer reading the vendor's own materials cannot tell which entity actually signed for this.

Look at the direction of the connection. The provider sends a resource request into Coder, Agent Relay launches a workspace, using a prebuild for a warm start, applies the correct template, and the provider's daemon inside that workspace opens a secure link back out to its external service [11]. Each Coder workspace can start a Cursor worker that dials outbound [4]. Nothing requires an inbound hole, which the network team will like, and the flow carrying agent traffic to the vendor is the one that ends up on the egress allowlist rather than under inspection. Coder says Agent Firewall, RBAC and audit logging apply automatically because the work happens inside a workspace, so there is no separate security model to build for agent traffic [6].

The split is planner remote, effectors local. Cursor continues to run the agent loop, including inference and planning, while tool calls execute in Coder environments on the customer's network, which is how Coder gets to say source code, secrets and internal services stay on machines the customer controls [3]. A planner still needs context to plan with. Both posts describe the perimeter in terms of where tool calls execute, not in terms of what the loop is sent to reason over. For the perimeter claim to hold up in a bank's review, one of two things has to be true: the remote loop plans without file contents, or the context it receives is scoped and recorded on the way out. Neither post says which.

The attribution work is the part I would pay for. Agent Relay resolves each incoming request against the organization's existing identity provider, so an agent's commits, tool calls and file access carry the same attribution as a human developer's [9]. Coder's own AI Maturity Model assessment of 100 engineering organizations found 70 percent running agents on infrastructure never designed to support them and only 31 percent at organization-wide governance, usually because nobody can say which human an agent's actions trace back to [19]. That leaves 69 of the 100 without it, and a 39-point gap between the two figures [21]. Coder assessed those organizations itself, so treat the numbers as the shape of the problem rather than its size.

The most persuasive paragraph in the launch post is the one about what failed. Coder's earlier self-hosted work used AgentAPI, an approach similar to today's Agent Client protocol, to run agent CLIs inside workspaces behind a common interface; it worked, but AgentAPI never gained traction as a standard, and wrapping CLI agents added friction when recreating their experience in a web app [14]. Registering as a compute pool in someone else's scheduler avoids re-implementing someone else's product surface, and it divides the work honestly: Coder supplies the machine, the vendor supplies the loop. Coder Agents remains the other path, running its reasoning loop in the Coder control plane while code, credentials and execution stay in customer infrastructure [15].

Both posts reach for Gartner. Forty percent of enterprise applications carrying task-specific agents by the end of 2026, up from less than 5 percent in 2025 [13], is more than eightfold in about a year [18], and that transfers to your capacity plan only if your applications get counted the way Gartner counts a task-specific agent feature. If your constraint is where commands run and which human owns them, this is the right tradeoff, and the plumbing looks well made. If your constraint is what the model sees, the relay moves execution and leaves that question where it was.

What to watch

  • Whether Coder documents what the Cursor worker sends outbound, and whether Agent Firewall can inspect that link rather than only allowlist the endpoint.
  • Whether a second agent vendor exposes a compute-pool interface Coder can register into, which is the real test of the any-agent claim.
  • Whether the truncated global financial institution compliance figure appears in the whitepaper with a stated baseline.

Clarity's read

What the record supports and how the coverage leans. The claims behind it follow.

Reality

Evidence35
Adoption10
Hype gap+40
Incentives85
Confidence60
Why these scores

Claim ledger

Ranked by verification strength, evidence, and original report placement.

  1. [1]

    Coder introduced Agent Relay, launching with Cursor as its first integration partner, giving enterprises a way to run cloud-hosted coding agents inside secure, self-hosted Coder workspaces instead of the agent vendor's cloud.

    Reported Supported Source: coder.com launch postView cited source
  2. [2]

    Coder frames the stalling question as not whether the agent is good enough but where it actually executes; by default a cloud-hosted coding agent reads the codebase, runs commands and writes files inside infrastructure the vendor controls.

    Reported Supported Source: coder.comView cited source
  3. [3]

    Cursor continues to run the agent loop, including inference and planning, while tool calls execute in Coder environments on the customer's network, so that source code, secrets and internal services stay on machines the customer controls.

    Reported SupportedView cited source
  4. [4]

    Each Coder workspace can start a Cursor worker that opens an outbound connection to Cursor.

    Reported SupportedView cited source
  5. [5]

    A second coder.com post announces Coder Agent Relay with SpaceXAI as its launch partner, while its own text refers to Cursor Cloud Agents running inside Coder workspaces and to SpaceXAI's agentic experience.

    Reported SupportedView cited source
  6. [6]

    Coder states that because the work happens inside a Coder workspace, Agent Firewall, RBAC and audit logging apply automatically, with no separate security model to build for agent traffic.

    Reported Supported Source: coder.comView cited source
  7. [7]

    Rob Whiteley, CEO of Coder, said enterprises never rejected AI agents but rejected the deployment model, and that the model was never the bottleneck.

    Reported Supported Source: Rob Whiteley, CEO of CoderView cited source
  8. [8]

    Platform teams can provision and scale the Cursor-running workspaces the same way they already manage developer environments, deploying Cloud Agents next to private code and custom hardware.

    Reported SupportedView cited source
  9. [9]

    Agent Relay resolves each incoming request against the organization's existing identity provider, so an agent's commits, tool calls and file access carry the same attribution as a human developer's.

    Reported SupportedView cited source
  10. [10]

    Agent Relay connects the Coder control plane to an external cloud-hosted agent provider where the agent's reasoning loop runs; the provider lets an organization configure compute pools for where its agent work executes, and Coder registers itself as one of those pools.

    Reported SupportedView cited source
  11. [11]

    When a developer starts a session, the agent provider sends a resource request, and Agent Relay launches a workspace, using prebuilt workspaces for a warm start, applies the correct template, and opens a secure connection between the agent provider's daemon inside the workspace and its external service.

    Reported SupportedView cited source
  12. [12]

    The two coder.com posts name the launch partner differently: one calls Cursor the first integration partner, the other names SpaceXAI as launch partner.

    Derived Supported
  13. [13]

    Gartner projects that 40% of enterprise applications will feature task-specific AI agents by the end of 2026, up from less than 5% in 2025.

    Reported Supported Source: Gartner, cited by CoderView cited source
  14. [14]

    Coder's early work on self-hosted agent infrastructure used AgentAPI, an approach similar to today's Agent Client protocol, to run agent CLIs inside Coder workspaces behind a common user interface; it worked, but AgentAPI never gained traction as a standard and wrapping CLI-based agents introduced friction when recreating their experience in a web app.

    Reported SupportedView cited source
  15. [15]

    Coder also built Coder Agents, which runs its reasoning loop in the Coder control plane while keeping code, credentials and execution inside the customer's infrastructure.

    Reported SupportedView cited source
  16. [16]

    Coder argues enterprises are running several agents from different vendors, that standardizing infrastructure around one agent does not help when the next team adopts a different one, and that rebuilding governance for every new agent does not scale.

    Reported Supported Source: coder.comView cited source
  17. [17]

    Gartner projects that 80% of enterprise software engineers will need to upskill for generative AI by 2027.

    Reported Supported Source: Gartner, cited by CoderView cited source
  18. [18]

    The projected move from under 5% in 2025 to 40% by end-2026 implies more than an eightfold increase in roughly one year.

    Derived Supported
  19. [19]

    Coder's AI Maturity Model assessment of 100 engineering organizations found that 70% are running AI agents in infrastructure that was never designed to support them, and only 31% have reached organization-wide AI governance, usually because nobody can say which human an agent's actions trace back to.

    Reported Insufficient Source: Coder's AI Maturity Model assessmentView cited source
  20. [20]

    Coder's post begins to describe one global financial institution using its identity and audit controls to cut compliance work, but the sentence is cut off in the supplied text and no figure is given.

    Reported InsufficientView cited source
  21. [21]

    Of the 100 organizations Coder assessed, 69 have not reached organization-wide AI governance, and the gap between the share on unsuitable infrastructure and the share with organization-wide governance is 39 percentage points.

    Derived Insufficient

Sources

2 independent publishers whose own reporting we read for this story.

  1. coder.com

    2 articles · September 2, 2026

    Coder and SpaceXAI Collaborate to Bring Agentic Coding Inside the World’s Most Regulated Enterprises
  2. thenewstack.io

    1 article · September 4, 2026

    “1% of my engineers are responsible for 40% of token spend”: Why Coder and SpaceXAI want to give developers nice things

Topics and entities

Topics

Entities