Two OpenAI moves in one week: buying Ona, giving Codex to open source, after the same prize
In one week OpenAI bought cloud-execution company Ona to complete Codex's runtime, and started handing Codex free to the most influential open source maintainers. Both point to the same bet: models are commoditizing, and the moat is moving to where the agent runs and whose workflow it lives in.
Summary
In a single week, OpenAI made two moves on Codex.
On June 11 it announced it would acquire Ona, folding the company’s secure cloud execution and orchestration technology into the Codex ecosystem. OpenAI’s stated reasoning: as Codex grows more capable, its most valuable work is stretching from minutes to hours or days, and people should not stay tied to the machine where a task began. Ona provides secure, persistent environments where an agent can keep accessing the tools, systems, and context it needs inside a customer’s own cloud until the work is done. OpenAI cites two figures: more than 5 million people use Codex each week, up 400 percent from earlier this year, and Ona has previously served 2 million developers. The deal is still subject to customary closing conditions and regulatory approval, and the two stay independent until close.
Almost in parallel, OpenAI launched Codex for Open Source, a support program for the maintainers behind critical open source software. Selected maintainers receive 6 months of ChatGPT Pro (including Codex), API credits for the project, and conditional access to Codex Security, aimed at easing the load of review, issue triage, and release management. Applications are reviewed on a rolling basis, and only the chosen are notified.
Read separately, one is an enterprise acquisition and the other is developer goodwill. Read together, they are two faces of the same move: OpenAI completes the cloud-execution layer Codex was missing on the way up, and pushes Codex free into the open source workflow on the way down. The prize is not a few more model calls. It is the default entry point for writing code with AI.
The move
Start with Ona. Codex already looked complete as a product, a model plus a CLI plus cloud tasks, but it was missing the one piece that matters most once an agent runs for a long time: where it actually runs.
For a patch task that takes a few minutes, the runtime is irrelevant. But OpenAI has staked its own narrative on work that lasts hours to days, and that runs straight into a wall. No enterprise will let an outside agent touch its source code, credentials, and build pipeline for days inside an environment it cannot see or control. What they want is the agent running in a cloud they have approved, with their hands on how permissions are scoped, how credentials are limited, how activity is logged, and how changes move through review. A customer-controlled cloud execution environment is exactly what Ona spent years building. OpenAI brings the model and the orchestration intelligence, Ona brings the ground the agent stands on, and only together does Codex become something an enterprise will trust in production.
Now Codex for Open Source. It looks like charity. It is mindshare. Open source maintainers are a pressure point in the software world: the projects they keep alive are depended on by millions of developers, and the daily work they do, PR review, issue triage, release management, is precisely where a coding agent can insert itself. Rather than buy ads, OpenAI drops the tool straight into the most tiring part of these people’s workflow and makes Codex the one they reach for without thinking. Once the maintenance flow of critical projects starts carrying Codex by default, the vast downstream of developers who depend on those projects quietly comes to treat Codex as the tool you are simply supposed to have.
Put the two together: shore up the foundation so Codex can enter production, capture mindshare so Codex becomes the open source default. Same target.
The real motive
The surface motives are in the announcements. Ona frees agents from a single device and helps more organizations deploy in production. Codex for OSS eases maintainer load. Both are true, and neither is the deepest layer.
The deepest layer: the model itself is commoditizing, and the moat has to move.
Over the past two years the capability gap between frontier models has narrowed fast, nowhere more than in code, where no leader can claim a generational edge. When the model stops being scarce, what decides a coding agent’s real reach is no longer how many points it leads by on a benchmark. It is two far stickier things: where its agent runs, and whose workflow it lives in. Win those, and model scores fade into the background, because users can no longer easily leave.
Buying Ona grabs the first one. A long-running agent needs a secure, customer-controlled, persistent place to run, and a model plus a CLI cannot supply that. Building it slowly in-house was one option. OpenAI instead bought a company that had built it over years and arrived with 2 million developers’ worth of usage, closing the gap in one step. Codex for OSS grabs the second. Giving it free to the most influential maintainers drives a stake into developer mindshare and the PR workflow, and it drives it at a pressure point that moves the whole dependency chain.
One detail worth naming: Ona’s co-founder and CEO is Johannes Landgraf. The label hardly matters, but the move really is a land grab. With an acquisition and a subsidy at once, OpenAI is fencing off both the ground the agent runs on and the developer’s mind, while the competitive window is still open.
Who is threatened
The most directly exposed are startups building developer-agent infrastructure. Companies like Ona, offering secure cloud execution, agent orchestration, and controlled runtime, are exactly the kind that get absorbed. The signal is cold: this layer is no longer a neutral market a startup can grow into on its own, it is strategic ground that model vendors have decided they must hold themselves. A good cloud-execution startup’s most likely outcome is not becoming a platform, it is being bought by some model vendor. The window for head-on competition is narrowing.
Next, other coding agents, especially those that are mostly a CLI plus a model and have not yet built out the where-it-runs and whose-workflow layers. By shoring up the foundation and grabbing mindshare at the same time, OpenAI raises the bar from “is the model good” to “does the whole chain exist.” Products that are a thin skin over a model will find it harder and harder to explain what they charge for.
Then there are Ona’s existing customers and ecosystem. OpenAI says the two stay independent until close, but any neutral infrastructure acquired by a frontier model vendor has its neutrality called into question. Customers who chose Ona precisely because it would connect to any model now have to weigh whether it will tilt steadily toward Codex.
As for direct rivals like Anthropic’s Claude Code, the threat is a different shape: OpenAI has pushed the front line into the open source maintainer’s workflow, the frontier where developer default habits are won. Whoever becomes the standard in critical projects first captures the mindshare of the entire downstream dependency chain.
What to ignore
Do not skim past the Ona deal as “OpenAI bought another IDE tool.” The weight sits in the piece it fills, the hardest and most valuable part of the agent economy: keeping a long-running agent working inside an environment the enterprise is willing to trust. Running under control inside a customer’s approved cloud is itself a capability, and its commercial value is plainly higher than an equally smart agent that cannot get inside the security boundary. Underrate this move and you misread where the whole coding-agent contest is heading.
But do not treat Codex for OSS as pure charity either. The perk is real, and six months of ChatGPT Pro is real money to a maintainer. It also carries clear lock-in: when the free window ends and you return to paid, your review, triage, and release flow may already have grown around Codex, and migrating out has a cost. Maintainers should price this before applying. The effort saved is now, the workflow you bind is long-term.
Finally, do not get pulled along by deal-price or user-count comparisons. The announcements give only a few numbers: Codex above 5 million weekly users, up 400 percent, and Ona having served 2 million developers. No acquisition price is disclosed. Any sharper figure or market-share comparison has no primary basis right now and should be ignored. The real story sits in the two pieces of ground OpenAI just fenced off, not in the price.
FAQ
What is OpenAI actually buying with Ona?
Cloud execution and orchestration technology. Ona's work was moving software development off local machines and into secure, reproducible cloud environments; by OpenAI's account it has served 2 million developers. What OpenAI wants is a persistent place for a long-running agent like Codex to keep working inside a customer's own cloud, so the task continues even after the laptop is closed. The IDE itself is beside the point. OpenAI already had the model and the CLI. The missing layer was where the agent runs, and who controls permissions and logging. That is the layer Ona fills.
Is Codex for Open Source free, and what are the conditions?
Free for selected maintainers, but time-boxed and gated. Those chosen get 6 months of ChatGPT Pro (which includes Codex), API credits for the project, and conditional access to Codex Security. The bar: you must be a primary or core maintainer of an active open source project with meaningful usage, broad adoption, or clear ecosystem importance, and your GitHub repo and profile must be public. Applications are reviewed on a rolling basis. Worth pricing in: after six months you return to paid, and by then your review, triage, and release flow may already be wired around Codex.
Codex or Claude Code, how should I choose now?
If you are an enterprise that cares about an agent running under your control inside your own cloud, the Ona acquisition clearly strengthens that line for OpenAI, and it is worth watching what it ships into. If you are an open source maintainer with an influential project, the six free months of Codex for OSS are real money, but price in the lock-in. In the short term, do not pick on model scores alone. Pick on how hard it is to migrate out once the agent is embedded in your workflow. Both vendors are racing to own the workflow, so favor the one whose agent behavior is more transparent and whose exit cost is lower.