OpenAI acquires Ona to make Codex persistent

OpenAI is buying Ona, a small cloud startup whose specialty is giving AI agents a secure, persistent place to work, and folding the team into Codex, the company’s coding and automation assistant. OpenAI announced the deal on 11 June 2026; the business reasoning was unpacked the same day by The Financial Express. Financial terms were not disclosed; the deal is subject to customary closing conditions and regulatory approvals.

Codex has stopped being a developer toy

The acquisition lands at a moment when Codex is being used at a scale that goes well beyond its original developer audience. OpenAI said more than 5 million people now use Codex every week — a 400% increase on the start of the year, and up from roughly 3 million in April. The shape of the work is changing too. A coding assistant that finishes a task in 90 seconds can run on a developer’s laptop. An assistant that takes 18 hours to refactor a customer-handling workflow cannot.

5Mpeople now use Codex every week, up from 3 million in April — a 400% jump since the start of 2026.

In OpenAI’s telling, the most valuable work Codex now does unfolds over hours or days, not minutes — and that work needs somewhere to live when the user is not in the room.

What Ona actually does

Ona sells secure cloud environments in which an AI agent can pick up tools, files and credentials, and keep working even after the human closes the laptop. The agents run inside a customer’s own cloud account, with the customer’s own controls over access, logs and review. In OpenAI’s framing, that is the missing piece that has stopped enterprises from putting Codex into production rather than pilots. Once the deal closes, the Ona team will join the Codex group inside OpenAI.

“Agents need more than intelligence; they need a trusted workspace. We built Ona to give agents cloud environments with the context, control and collaboration enterprises require.” — Johannes Landgraf, Ona CEO, in a LinkedIn post quoted by The Financial Express

Thibault Sottiaux, OpenAI’s Core Products Lead, made the same point from the buyer’s side: enterprises want powerful agents that can do real work while meeting the security and control requirements of their environments, he said, and Ona will help Codex meet those standards.

Why the deal matters beyond OpenAI

Read narrowly, this is a tuck-in acquisition — a small team joining a much larger one. Read in context, it is part of a wider consolidation of the agent stack by a handful of well-funded model companies:

• OpenAI is buying its way into the agent runtime — the bit between the model and your files — rather than relying on partners for it.
• It is the latest in a string of deals. OpenAI bought cybersecurity startup Promptfoo in March, healthcare technology startup Torch earlier in the year for about $60 million, the maker of the Mac interface Sky in October, and the Jony Ive hardware startup io in May 2025 for more than $6 billion.
• An IPO is on the horizon. OpenAI recently confidentially filed paperwork with the US Securities and Exchange Commission, days after rival Anthropic submitted its own filing.

For a small UK firm, the practical question is what happens to the tools you already use when one vendor owns the model, the runtime, the cloud workspace and the audit logs.

What to do with this

A few things are worth doing in the next fortnight, before the deal closes and the product roadmap hardens.

• Audit where Codex already runs in your business. Pull a list of the workflows you have built or are piloting on Codex, and note which ones currently run on someone’s laptop versus in a managed environment. The laptop ones are the first candidates to be moved to Ona-style persistent runtimes once those arrive in Codex.
• Decide what you would do if your data had to live in a US cloud account. Ona’s whole pitch is that the agent runs inside your tenancy. If that is something your compliance team needs, push OpenAI for written confirmation of the data-residency story before you commit a production workflow to it. If you are a regulated firm — a practice handling client files, for instance — this is the conversation to have now, not at renewal.
• Compare on the same axis. The Claude Fable 5 release and the Microsoft Agent Framework going GA are two of the alternative agent stacks worth running a small pilot against. The point of a pilot is not to declare a winner; it is to learn what your team actually needs from a runtime so that, when Codex’s persistent mode lands, you can tell whether it is the right fit.
• Keep an eye on pricing, not just features. Persistent agents run for hours or days, which is a fundamentally different cost shape from a 90-second chat. Our primer on usage-based agentic pricing walks through what to budget for; pair it with the free-tier roundup if you are still at the exploration stage rather than production.

The headline is that Codex is being turned into a platform you deploy, not a tool you open. For a small UK team, that is a real upgrade — and a real reason to be deliberate about which workflows you let it own.