OpenAI announced Wednesday it is acquiring Ona, a cloud execution startup formerly known as Gitpod, in a move that signals what the next phase of AI deployment actually looks like: agents that keep working long after you stop watching them.

The deal, disclosed June 11, closes a gap that has frustrated enterprise AI deployments for the past year. OpenAI’s Codex can already write code, analyse data, and orchestrate multi-step workflows. What it has lacked is a persistent, secure place to do that work when no human is actively supervising the session. Ona solves that problem.

What Ona Actually Does

Ona provides pre-configured cloud environments stocked with the tools, systems, and context an AI agent needs to operate independently. When Codex receives a task — audit this codebase, refactor this module, run this analysis pipeline — it no longer needs your laptop to be open for the job to finish. The work happens in Ona’s cloud environment, continues across hours or days if needed, and hands results back when complete.

That might sound like a minor infrastructure detail. It is not.

The practical limit on current AI agents is that they are session-bound. Start a task, step away, and the session times out. That constraint forces a human to babysit every meaningful agent deployment, which defeats much of the productivity argument. Ona removes that constraint entirely.

Ona’s enterprise clients before the acquisition included a major US bank, European pharmaceutical companies, and Asian sovereign wealth funds — exactly the industries where compliance, auditability, and continuous operation matter most. The platform includes access controls, audit trails that log every query and external call, and isolated execution environments that prevent agent actions from contaminating production systems.

The Codex Context

More than 5 million people now use Codex weekly — a figure that represents roughly 400% growth since the start of 2026. That user base spans individual developers automating personal workflows all the way to enterprise teams deploying Codex across engineering departments.

Until now, Codex’s most ambitious capabilities have been constrained by the session problem. A developer could start a complex refactor, leave for a meeting, and return to a timed-out session. An enterprise team could queue up a large analysis run and watch it fail midway because the connection dropped.

With Ona’s infrastructure integrated into Codex, those scenarios change. Multi-day tasks become practical. Agent deployment at scale becomes an infrastructure problem rather than a human oversight problem.

OpenAI confirmed that financial terms of the deal were not made public. The acquisition is subject to customary regulatory approvals and closing conditions.

Why This Matters Beyond Coding

The Ona acquisition is being framed as a Codex infrastructure play, and it is. But the implications run wider.

The core capability — a secure, persistent, auditable cloud environment where an AI agent can operate without continuous human supervision — is exactly what enterprise AI deployments need across every function, not just software development.

Think about what this infrastructure enables: a finance agent that runs overnight reconciliation and delivers a report at 7am. A compliance agent that monitors contract databases continuously without a human triggering each check. A customer operations agent that handles escalations in parallel across time zones without shift handoffs.

This is not theoretical. Ona’s existing enterprise clients were already deploying exactly these patterns before the OpenAI acquisition. The deal accelerates availability to the much larger Codex customer base.

What This Means for Business

The agentic infrastructure gap is closing fast. The biggest barrier to enterprise AI agent deployment has not been model capability — it has been infrastructure: secure execution environments, audit trails, persistent context, human escalation paths. Acquisitions like Ona show the major AI providers are methodically filling those gaps.

“AI agent” is becoming a meaningful enterprise concept. Twelve months ago, “AI agent” in enterprise contexts meant a chatbot with some automation bolted on. Today it means a persistent, auditable autonomous workflow with a defined scope, access controls, and a reliable execution environment. That distinction matters when you are building business processes around it.

The build vs buy calculation is shifting again. For organisations that have been waiting to deploy AI agents because the infrastructure felt immature, the maturity curve is accelerating. The window for building proprietary agent infrastructure is narrowing as the major platforms build it in.

Governance features are becoming table stakes. Ona’s audit trails and access controls are not differentiators — they are standard expectations. Any AI agent infrastructure you evaluate or build should start from the assumption that audit trails, scoped access, and human override mechanisms are non-negotiable.

The pace of AI agent infrastructure development has moved faster than most enterprise IT planning cycles. Understanding what the stack looks like today — and where it is heading — is becoming a core strategic competency, not just a technology decision.

---

Enterprise DNA’s Omni Ops service helps businesses design and deploy AI agent workflows that fit their actual operations — not theoretical best practices. Start the conversation with Sam’s team about what persistent AI agents could do inside your business.