OpenAI to acquire Ona
OpenAI
OpenAI is buying Ona, a startup that builds cloud dev environments, to beef up Codex. The plan is letting AI agents run long tasks securely inside company systems, not just quick code snippets.
Based on reporting by OpenAI — read the original for the full story.
Summary, retelling and take written by AI under human oversight; images are AI-generated illustrations. How we work · Report an error
OpenAI has confirmed it's acquiring Ona, a smaller outfit that specializes in secure, persistent cloud environments for software development. The deal folds Ona's tech directly into Codex, OpenAI's coding-focused AI product, and the stated goal is straightforward: give AI agents a stable place to live and work inside enterprise systems, not just a sandbox that resets every session.
That distinction matters more than it sounds. Most AI coding assistants today operate in short bursts — you ask, it answers, the context vanishes. Ona's infrastructure is built around persistence, meaning an agent can keep a workspace, credentials, and state over hours or days rather than minutes. Pair that with Codex's ability to write and execute code, and you get something closer to a junior engineer who doesn't forget what they were doing when you step away for lunch.
OpenAI has been pushing Codex hard as its wedge into enterprise software teams, competing directly with GitHub Copilot and various agentic coding startups. Buying rather than building the persistent-environment layer is a familiar move for the company at this stage — it did something similar when it scooped up smaller teams to speed up product gaps rather than engineer them from scratch. Ona's technology, in particular, seems chosen for one job: making it safe for a company to let an autonomous agent touch real infrastructure without babysitting it constantly.
No financial terms were disclosed, and OpenAI hasn't detailed a timeline for when Ona's tools show up inside Codex proper. But the direction is clear enough. The company wants Codex to stop feeling like a chatbot that happens to write code and start feeling like a system that can be handed a multi-day project and left alone — with enough guardrails that IT departments don't have a collective panic attack.
My take — AI-written commentary, not fact-checked reporting
Every acquisition like this is really OpenAI admitting that 'agent' still means something fragile and short-lived under the hood, and they need someone else's plumbing to fix it. I'd bet this pattern repeats: labs building flashy models, then buying boring infrastructure companies to make the flash actually usable. It's a tell that the hard part of AI right now isn't intelligence, it's plumbing.
Read more about this at: OpenAI