It's 2 a.m. The developer shut her laptop hours ago. Somewhere in a data center, a process she kicked off before dinner is still running — reading her codebase, writing functions, fixing the bug she described in a single sentence. She's asleep. The work isn't.

That scenario is what OpenAI just paid an undisclosed sum to own.

On June 11, OpenAI announced it's acquiring Ona — the company most engineers still know by its old name, Gitpod. Founded in Kiel, Germany, back in 2020, it spent years selling cloud development environments before quietly rebranding to Ona in late 2025 and rebuilding the entire product around AI agents. Now it belongs to the maker of ChatGPT.

Here's what Ona actually does. It spins up secure, pre-loaded cloud environments — stocked with your tools, your repositories, your context — where an AI agent can grind through a multi-step coding job for hours, sometimes days. The human who started it can log out, close the laptop, drive home. The agent keeps committing.

OpenAI is bolting this straight onto Codex, its coding assistant. And Codex is no longer a side experiment. More than 5 million people use it every week, a figure OpenAI says is up 400% since the start of the year. The Ona team folds into the Codex team. Terms weren't disclosed, and the deal still needs regulatory sign-off.

The timing tells you everything. Anthropic has spent the past year eating the enterprise coding market alive, and OpenAI knows it. This isn't a company expanding from comfort. It's a company buying the infrastructure it didn't build fast enough, because the rival across town got there first.

And the headline feature — agents running inside your own cloud, untethered from any human at the keyboard — is the part nobody wants to admit is also the risk. One trade outlet framed the deal, accurately, as OpenAI buying technology to "rein in" its own agents. Sit with that phrasing. You don't need to rein in software that waits for permission.

My Opinion

I'll be blunt: the feature everyone's cheering is the one that should make engineering leaders nervous. An autonomous agent making commits to your codebase at 3 a.m. with zero humans watching is not a productivity win by default — it's a liability you've agreed to sleep through. The selling point and the failure mode are the same sentence.

Here's what bugs me about the bigger pattern. OpenAI keeps buying instead of building. Sora got killed. The companionship products got shelved. Now, staring down Anthropic's grip on developers, the move is to acquire a German startup's plumbing rather than ship its own. Nothing wrong with a sharp acquisition — Ona's tech is genuinely good — but a string of them starts to look less like strategy and more like a company spending its way out of second place.

Still, I won't pretend this doesn't matter. Persistent, cloud-native agents are where coding is actually headed, and running them inside the customer's own security perimeter is the right architectural call — the only one enterprises will tolerate. Whoever nails the boring parts — permissions, audit logs, an instant kill switch for a runaway agent — owns the next five years of software development. OpenAI just bought itself a head start on the boring parts. The teams that learn to supervise these agents will outlast the ones that simply let them run.

Author: Yahor Kamarou (Mark) / www.humai.blog / 13 Jun 2026