If you've run a real long-horizon coding task through Codex — "migrate this module to TypeScript," "find and fix every SQL injection pattern in this repo," "modernize these 400 tests from Jest to Vitest" — you know the ceiling. The agent runs. Then you take a meeting, or your laptop sleeps, or the tab times out. Session ends. Work stops. You're not farther along than when you started; you're just staring at a half-refactored codebase.

OpenAI announced on June 11 that it is acquiring Ona, the German cloud startup most builders know as Gitpod. The acquisition is specifically designed to fix that ceiling.

Gitpod rebranded to Ona in late 2025 after pivoting from cloud development environments to AI agent orchestration. Ona provides secure, pre-configured cloud environments where agents can keep running after the developer disconnects. The execution happens inside the customer's own cloud infrastructure, not OpenAI's. That second property is the one enterprise security teams have been waiting for before greenlighting Codex for real production work. OpenAI provides the intelligence; you keep the data.

More than 5 million people use Codex each week, up 400% from earlier this year. That growth rate is what made this acquisition necessary. Session-bound architecture — agents that run as long as you're watching and stop when you're not — doesn't scale with that kind of usage growth. The gap between "Codex as a clever tab" and "Codex as a production workhorse" is the one you can't engineer past without changing the underlying execution model.

Terms of the deal weren't disclosed. The acquisition is subject to customary closing conditions.

Source spread

• OpenAI — "To acquire Ona" — [hype]. Company announcement; leads with the 5M weekly users and 400% growth figure; frames Ona as the natural next step in Codex's "longer-running tasks" roadmap.
• The Next Web — "OpenAI acquires Ona to run Codex agents inside the customer's own cloud" — [builder]. Best coverage of the Gitpod-to-Ona pivot and what the customer-controlled execution model actually means in practice.
• CNBC — "OpenAI to acquire Ona to support its AI coding assistant, Codex" — [builder]. Clean on the enterprise angle; emphasizes the agent-persistence use case.
• Bloomberg — "OpenAI to Acquire Cloud Platform Ona to Support AI Agents" — [builder]. First to report; notes terms undisclosed and deal subject to closing conditions.

Pros & cons

What's real:

• Persistent cloud execution unlocks the task class that actually matters for engineering teams. Large-scale refactors, automated security audits, multi-hour test migrations — these are what enterprises want from AI coding tools. Session-bound execution has been the hard stop, and it's been that way since Codex launched.
• Customer-controlled execution is the correct enterprise architecture. Running agents inside your own environment, with OpenAI handling only intelligence and orchestration, is the only model enterprises with real IP, compliance, or data residency requirements will accept. The alternative — an agent with access to your codebase running on OpenAI's infrastructure — is a non-starter for a significant portion of potential Codex customers.
• Gitpod worked with more than 2 million developers on production-hardened secure cloud environments. OpenAI is acquiring a proven infrastructure foundation, not a startup with a pitch deck and a prototype.
• The Gitpod-to-Ona pivot was the right read on where the market was going. OpenAI didn't need to convince Ona that AI agents were the future — Ona had already bet on it before this acquisition happened.

What deserves a side-eye:

• "Terms not disclosed" means we don't know whether this is a strategic platform acquisition or primarily a talent acquisition with incidental IP. Those have different timelines for when persistent execution actually ships in a form builders can use.
• The 400% Codex growth figure is OpenAI's number, using OpenAI's definition of "user," from an unspecified baseline. "Earlier this year" is not a specific comparison period. Don't treat this as independently verified growth data.
• Enterprise procurement doesn't move at startup speed. The security review and approval cycle for "AI agents running in our cloud" is measured in quarters. The integration can exist before the enterprise deals do, but not before the enterprise audits do.
• The deal hasn't closed yet. Nothing to build on until it does.

Samwise's take

What builders need to know

Further reading

• OpenAI — "To acquire Ona" — official announcement; primary source for the 5M weekly user figure and 400% growth
• The Next Web — "OpenAI acquires Ona to run Codex agents inside the customer's own cloud" — best on the Gitpod pivot and customer-controlled execution
• CNBC — "OpenAI to acquire Ona to support its AI coding assistant, Codex" — enterprise framing
• Bloomberg — "OpenAI to Acquire Cloud Platform Ona to Support AI Agents" — first report; terms undisclosed
• OpenAI — "Introducing Codex" — context on what Codex was before this acquisition