OpenAI and Cursor converge on agent coordinators, but disagree on who runs them

On September 10, OpenAI and Cursor exposed the same architecture independently. OpenAI opened its Agents API in public beta with the harness that powers Codex, and Cursor launched Projects, where a coordinator agent directs thousands of subagents. In both, the coordinator understands the overall objective and manages the work while specialized agents execute individual pieces.

The pattern is not new: AWS Bedrock AgentCore reached general availability in October 2025, and Anthropic's Claude Managed Agents entered public beta in April 2026.

Why the single-agent loop breaks down

A coding agent works in a loop: it observes the repository, chooses the next step, calls a tool and examines the result. That is enough for a small task, not for a large migration: the pressure lands on the context window. "A large context doesn't only include everything correct or important; it also includes a lot of throwaway information," Hilliary Lipsig, a senior principal site reliability engineer at Red Hat, told The New Stack. A 2026 study of Claude Opus 4.6, GPT-5.4 and Gemini 3.1 Pro found the models missed a dangerous action buried in a long transcript two to 30 times more often once it followed 800,000 tokens of benign activity.

The coordinator is a control plane, not another coding agent

Once work is divided, the coordinator no longer writes code: it holds the global task, manages dependencies and decides how execution proceeds. Anthropic documented the same pattern as orchestrator-subagent architecture: in a June 2025 writeup it reported that a Claude Opus 4 lead agent with Claude Sonnet 4 subagents outperformed single-agent Opus 4 by 90.2% on its internal research evaluation, at roughly 15 times the token cost.

Parallelism also brings distributed-systems failure modes. Cursor moved its cloud-agent execution loop to Temporal, pushing reliability past two 9s; Temporal now handles 50 million Cursor actions a day across 7 million unique workflows. "Durable execution isn't a nice-to-have here. It's the difference between a system you can operate and one you can only demo," Lipsig said.

Where OpenAI and Cursor differ

The convergence does not make the two systems interchangeable. OpenAI exposes the harness through an API, and it is open source, so teams can inspect the coordinator logic. The Agents API supports U.S. data residency but not Zero Data Retention.

Cursor packages more of the surrounding workflow: Projects combines the coordinator, cloud execution, shared project context and a developer-facing workflow in one environment. An API leaves more decisions to developers; an integrated platform answers more of them on their behalf. Neither approach removes the engineering problems, it changes where they are implemented and who operates them.