OpenAI is reaching for a piece of plumbing that most people never see but every serious coding agent needs: a place to keep running once the laptop is shut. On June 11, 2026, the company said it had agreed to acquire Ona, the developer-tooling startup formerly known as Gitpod, and fold its team and technology into the group that builds Codex, OpenAI's AI coding assistant.
What Happened
OpenAI announced plans to buy Ona, officially registered as Gitpod GmbH, a German company whose platform lets AI coding agents run inside cloud-based sandboxes rather than on a developer's own machine. The financial terms were not disclosed, and the deal still requires regulatory approval before it closes. Ona's entire team, led by chief executive and co-founder Johannes Landgraf, is set to join OpenAI's Codex group once the transaction is finalized.
The logic of the purchase is rooted in a stubborn limitation of today's coding assistants. Developers typically run these agents on their local computers, which means that when the workstation is turned off, the agent's work stops with it. That makes multi-hour or multi-day tasks fragile. Ona's sandboxes stay online after a developer closes the lid, so an agent can keep working uninterrupted, draw on more compute than a single laptop offers, and finish jobs faster.
Security is the other half of Ona's pitch. Its sandboxes are deleted as soon as they are no longer needed, reducing the window in which sensitive code could be exposed, and administrators can block risky programs even when an attacker renames a file or buries it inside a script. The platform can also wall agents off from parts of the file system that hold credentials such as encryption keys, and it can cut outbound connections to suspicious servers. For OpenAI, which says Codex now has more than five million weekly users, that combination of persistence and control is exactly what enterprise customers have been asking for. The deal arrives three months after OpenAI bought Promptfoo to bolster the security features of its Frontier agent-building platform.
Why It Matters
The acquisition is a small headline with a large subtext: the competition in AI has moved past the model itself and into the runtime that an agent lives inside. A frontier model is no longer enough on its own. What increasingly decides whether an agent is useful in production is the surrounding scaffolding — where it executes, how long it can run, what it is allowed to touch, and how cleanly it shuts down.
By pulling Ona in-house, OpenAI gains an answer to that question that it controls end to end. Long-running agents that survive a closed laptop, spin up extra compute on demand, and operate inside policy-governed boundaries are the difference between a clever demo and software a company can actually deploy. The move also tracks a broader 2026 pattern in which the biggest players — chipmakers, cloud providers, and model labs alike — have spent the year competing over agent control planes rather than over benchmark scores.
There is a competitive edge to the timing as well. Rivals have been pressing hard into the enterprise developer market, and a credible, secure execution layer for Codex helps OpenAI defend ground where reliability and governance matter more than raw capability. Owning the sandbox, rather than renting it, lets OpenAI tune the entire loop from prompt to deployed code.
Reaction
For the millions of developers already leaning on Codex, the most tangible promise is simple: agents that do not quit when you close your computer. Engineers who have grown used to babysitting a coding assistant — keeping a terminal open, a session alive, a machine awake — stand to get back hours if the work can be handed off to a cloud sandbox that simply keeps going.
The deal has also been read through the lens of the broader contest among AI labs for the loyalty of professional developers. Coding has become one of the most commercially valuable proving grounds for AI, and each lab is racing to make its assistant the default tool inside engineering teams. Acquiring an established execution and orchestration layer, rather than building one from scratch, signals how urgently OpenAI wants to close any gap in that arena. Observers framed the Ona purchase as OpenAI answering rivals' enterprise push with infrastructure of its own.
What's Next
OpenAI says Ona will improve Codex's ability to handle long-running tasks that stretch across hours or days, and crucially, to let users step in to review an agent's work and provide input partway through. That human-in-the-loop checkpoint is a meaningful design choice: as agents take on bigger jobs, the ability to inspect and correct them mid-flight becomes as important as the autonomy itself.
Before any of that ships, the acquisition has to clear regulatory review, and the two companies have not put a public timeline on closing. Once it does, the Ona team folds into Codex, and the practical test will be how smoothly its sandbox technology integrates into a product already used by millions every week. Watch for Codex updates that emphasize durable, reviewable agent sessions — the clearest sign that Ona's engineering has landed.
Closing Thoughts
It is tempting to read an undisclosed-terms acquisition of a developer-tools startup as a footnote. But the Ona deal captures where AI is actually heading. The frontier is shifting from how smart a model is in a single reply to how dependably an agent can work on its own over time, safely, and under supervision when it counts.
That is a quieter ambition than a flashy new model launch, and arguably a more consequential one. The labs that win the next phase may be the ones that make autonomous agents boringly reliable — able to run for days, respect guardrails, and hand control back to a human at the right moment. OpenAI's move to own that layer suggests it sees the same future, and intends to build the floor it stands on.
한글 요약
오픈AI가 6월 11일 독일 개발자 도구 스타트업 오나(Ona, 옛 깃팟·Gitpod)를 인수한다고 발표했습니다. 인수 금액은 공개되지 않았으며 규제 승인 절차가 남아 있습니다. 오나는 AI 코딩 에이전트를 개발자의 PC가 아닌 클라우드 샌드박스에서 실행해, 노트북을 꺼도 작업이 중단되지 않고 며칠에 걸친 장기 작업까지 이어갈 수 있게 해 주는 기술을 보유하고 있습니다. 요하네스 란트그라프 CEO가 이끄는 오나 팀 전체가 오픈AI의 코덱스(Codex) 그룹에 합류합니다.
핵심은 보안과 지속성입니다. 오나 샌드박스는 필요가 끝나면 자동 삭제되고, 파일명을 바꾸거나 스크립트에 숨긴 악성 프로그램까지 차단하며, 암호화 키 같은 민감 정보 접근과 의심스러운 외부 서버 연결도 막습니다. 주당 500만 명이 사용하는 코덱스에 이 같은 '오래 작동하면서도 통제 가능한' 실행 환경을 더하려는 것입니다. 이는 모델 자체의 성능 경쟁을 넘어, 에이전트가 실제로 돌아가는 런타임을 누가 장악하느냐로 경쟁의 무게중심이 옮겨갔음을 보여 줍니다.
오픈AI는 오나를 통해 코덱스가 장기 작업을 더 잘 수행하고, 사용자가 중간에 작업을 검토하고 개입할 수 있도록 개선하겠다고 밝혔습니다. 인수가 마무리되면 관건은 수백만 명이 쓰는 제품에 오나의 샌드박스 기술이 얼마나 매끄럽게 통합되느냐입니다. 화려한 신모델 공개보다 조용하지만, AI의 방향이 '똑똑한 한 번의 답변'에서 '안정적으로 오래 일하는 에이전트'로 이동하고 있음을 보여 주는 상징적 거래입니다.
참고: OpenAI 발표, SiliconANGLE, InfoWorld (2026년 6월).