Ona joins OpenAI to deepen Codex infrastructure

Ona announced on June 11 that it has entered into an agreement to join OpenAI as part of the Codex team, subject to customary closing conditions, including required regulatory approvals. Until the transaction closes, the companies will remain separate, and Ona says it will keep supporting customers under existing commitments. The useful signal is not simply another AI acquisition. It is the kind of operating layer OpenAI wants to connect to Codex.

Ona, formerly known through Gitpod, describes its product as mission control for software engineering agents: customer-controlled cloud environments that keep state, tools, access and project context beyond a single laptop session. In plainer terms, it gives an agent a persistent workshop instead of a transient chat window or disposable terminal. In its announcement, Ona says weekly agent sessions have grown 13 times since the start of the year across demanding institutional customers, including a major bank, a large pharmaceutical company and a sovereign wealth fund.

For Codex, the rationale is straightforward. Coding assistants are moving from code completion toward longer workflows: reading an existing codebase, changing several files, running tests, opening a pull request and then resuming later with the same context. That shift depends less on broad claims about “agents” and more on infrastructure that can be trusted: isolation, permissions, audit trails, state recovery, repository access and integration with internal systems. Those are the specific building blocks Ona says it brings to enterprise work.

The announcement should still be read carefully. Ona does not disclose deal value or a firm closing date, and it explicitly notes the need for regulatory approvals and other closing conditions. What changes now is OpenAI’s stated product direction. Codex is being framed less as a coding helper tied to a user’s machine and more as a cloud work layer for software and, potentially, knowledge work. If the integration closes, competition in agentic AI will not turn only on model quality. It will also turn on who can let agents work inside controlled, persistent and verifiable environments without forcing enterprises to give up governance.