OpenAI and Cursor split over control of agent coordinators
OpenAI has released a public beta of its Agents API, exposing the framework that supports Codex through managed sessions and tools. OpenAI and Cursor share a belief in agent coordinators but differ on who should operate them.
The available details are limited. OpenAI has opened a beta version of an interface for building with AI agents, revealing infrastructure that underpins Codex via hosted sessions and tool access. Both OpenAI and Cursor endorse the idea of systems that coordinate agents, yet they disagree about which party should run those systems. This places the dispute within broader questions about platform control, developer tooling, and how AI agent infrastructure is governed.
Developers and organizations building with AI agents may be affected by who controls coordination layers. If OpenAI and Cursor diverge, teams could face competing ecosystems, affecting integration choices, costs, and oversight. Businesses relying on Codex-style workflows may see changes in tooling and session management. The outcome could shape how much control platform providers versus application builders have over agent behavior, with implications for transparency, reliability, and accountability.