On June 10, 2026, OpenAI announced it was acquiring Ona, a small startup that builds secure cloud execution environments for AI agents. The deal is about one thing: fixing a gap that kept Codex useful only for short, simple tasks.
Before this, Codex ran in isolated containers with no internet access, no database connectivity, and no ability to keep working once your laptop closed. You could ask it to fix a bug or write a feature. Anything bigger required you to stay present and stitch the pieces together yourself. That limitation made Codex a decent developer aid but a weak enterprise tool.
Ona changes the infrastructure underneath all of that.
The biggest shift: agents that run while you sleep
With Ona’s persistent cloud environments, Codex agents can work for hours or days without interruption. You close your laptop at night, and the agent keeps going. File state, memory, and environment variables carry across the entire workflow. No restarting from scratch, no lost context.
This matters more than it sounds. A multi-stage task like deploying a full feature with database migration, API integration, testing, and documentation used to require a developer managing each stage manually. Now six coordinated agents can handle the whole thing overnight. Developer review time drops from hours to about fifteen minutes.
Your data stays on your infrastructure
One concern enterprises have with AI tools is data control. Ona addresses it directly: the cloud environments run on the customer’s own infrastructure, whether that’s AWS, Azure, GCP, or a private cloud. OpenAI does not host your data. You manage your own security boundaries, access controls, and compliance policies.
This matters for organizations subject to SOC 2, GDPR, or HIPAA. Customer-controlled infrastructure means audit trails are yours, access is role-based, and nothing passes through a third-party host you didn’t choose.
What Codex can do now that it couldn’t before
Tool access was previously blocked entirely. Codex had no way to reach external APIs, read or write to databases, or run command-line tools. After Ona, agents work inside a governed tool access layer. APIs are whitelisted. Database permissions are role-based. Every interaction gets logged.
This opens up use cases well beyond coding. Agents can now generate technical documentation from actual code, pull data from multiple sources to produce business reports, manage and update knowledge bases, and prepare compliance audit materials. These are tasks that previously required a human working full days.
Who should pay attention
If you run an engineering team, the most immediate payoff is large-scale work you’d otherwise schedule across days. Codebase refactoring across multiple repositories, database migrations with live testing, full feature deployments from analysis through documentation — these now run autonomously. The developer reviews the output rather than doing the work step by step.
If you lead an enterprise IT or operations function, the longer-term angle is broader automation. IT incident response, data pipeline maintenance, customer support ticket handling — these are on the roadmap as Codex expands beyond software engineering.
No other platform in this space currently offers persistent, customer-controlled cloud environments with full multi-agent coordination. GitHub Copilot, Amazon CodeWhisperer, and Microsoft Copilot all cap out at minutes-long tasks and require your laptop to stay on. That gap gives OpenAI a real lead, though competitors will eventually respond.
Quick verdict: This acquisition is meaningful, not just incremental. Ona solves the exact infrastructure problem that kept Codex from being useful at enterprise scale. If you’ve written off AI coding tools as good only for autocomplete, this changes the calculation. The jump from single-task assistant to overnight autonomous agent is a practical one now, not a theoretical one.