OpenAI Acquires Ona: Codex Just Got a Persistent Brain
One Acquisition, One Very Clear Signal
On 11 June 2026, OpenAI announced it will acquire Ona, a secure cloud execution and orchestration platform, to expand its Codex ecosystem. The deal is pending standard regulatory approvals. Until closing, both companies remain separate.
If you've been treating Codex as a clever code assistant, this acquisition should change that frame entirely. OpenAI is building something much bigger: an agent that works through the night, with or without you.
What Ona Actually Does
Ona has spent years helping developers move software work from local machines into reproducible, secure cloud environments. It counts 2 million developers across its platform and already shares customers with OpenAI.
That infrastructure, persistent workspaces, scoped credentials, activity logging, secure execution contexts, is exactly what an enterprise agent needs to operate without a human babysitting it. It's not glamorous. It's the plumbing. And the plumbing is what makes agentic AI actually deployable at scale.
The Shift From Sessions to Sustained Work
Codex currently serves more than 5 million weekly users, up 400% from earlier in 2026, according to OpenAI's official acquisition announcement. That growth tells you adoption is real. But the next problem is duration.
Right now, most AI tasks are bounded by a session. You open a tool, run a task, close the laptop. The agent stops. With Ona's infrastructure inside Codex, an agent can continue working inside your cloud environment after the session ends, checking progress, making decisions, moving work through review stages, all without you staying connected.
That's the architectural shift. From synchronous, user-triggered tasks to asynchronous, agent-driven workflows.
What This Means for Your Ops Stack
The practical implication is that your AI architecture decisions just got more consequential. A persistent agent needs governance, not just a prompt. Think about what it can access, how credentials are scoped, how its activity is logged, and how outputs are reviewed before they affect live systems.
Ona's model keeps agents operating inside your own cloud environment. OpenAI provides the intelligence and orchestration; you retain control over the infrastructure, the data boundaries, and the security perimeter. That separation matters enormously for enterprise procurement, compliance, and audit teams.
| Dimension | Session-Based AI (Current) | Persistent Agent (Post-Ona) |
|---|---|---|
| Task duration | Minutes, tied to active session | Hours or days, runs asynchronously |
| Human involvement | Continuous oversight required | Check-in model: review, redirect, approve |
| Infrastructure | Local device or single-session cloud | Customer-controlled persistent cloud env |
| Security model | Per-request permissions | Scoped credentials, full activity logging |
| Use cases | One-shot tasks, drafting, analysis | Production workflows, test runs, migrations |
The Question Growth Teams Should Be Asking Right Now
Which of your current workflows depend on a human staying in the loop purely because the AI can't persist between steps? Content pipelines, SEO audits, data pulls, QA cycles: these are all candidates for persistent agent automation once this infrastructure is in place.
The teams who will move fastest are those already thinking about AI as a workflow layer, not a chat interface. If you're still evaluating how AI fits into your broader attribution and measurement stack, the piece on deploying OpenAI models via Amazon Bedrock for enterprise growth teams is a useful parallel read on operationalising these decisions.
And if you're thinking about how AI-generated output surfaces in search, our breakdown of GA4's automated tracking of ChatGPT and Perplexity traffic is directly relevant to measuring what persistent agents produce.
The Concrete Action
Audit your current AI use cases for session dependency. Any task that stalls because a human has to re-initiate it is a candidate for persistent agent deployment. Map those workflows now, before your competitors do, and brief your security and infrastructure teams on Ona's customer-controlled execution model so procurement isn't the bottleneck when this ships.
Persistent agents are not a future concept. They are the next production release.
Tags