OpenAI just announced it plans to acquire Ona, a company focused on secure, persistent cloud environments. The goal is straightforward: make Codex capable of running AI agents that don’t lose their place when a task takes hours, days, or even weeks. If you run a service business and you’ve been experimenting with AI agents, this is the infrastructure move you should pay attention to.
Right now, most AI agents hit a wall. They can do quick tasks just fine. But ask one to monitor a compliance workflow over several days, or manage a multi-step research project, and it falls apart. It forgets context. It loses its compute environment. It basically has to start from scratch every time it wakes up. Ona is built to fix exactly that problem.
What happened
- OpenAI announced plans to acquire Ona, a company specializing in secure, persistent cloud environments.
- The acquisition is aimed at expanding Codex, OpenAI’s developer-facing product, with infrastructure that lets AI agents run continuously across long timeframes.
- The target use case is enterprise workflows that require agents to maintain state, memory, and compute access over hours or days, not just seconds.
- This signals OpenAI is moving beyond chat-style AI and investing heavily in agent infrastructure (the plumbing that lets AI agents actually do sustained work in the real world).
Why persistent environments matter for AI agents
Think of it this way. Today’s AI agents are like contractors who show up, do one task, then forget your name and the job site address. Persistent cloud environments give those contractors a permanent office, a filing cabinet, and a key to the building. They can pick up where they left off.
That’s a big deal for any workflow that takes more than a few minutes. And in service businesses, most real work does.
5 things service business operators should know about this acquisition
- Long-running agents are about to get real. Until now, building an AI agent that could run a multi-day task (like auditing a quarter’s worth of contracts or reconciling months of financial data) required duct-taping together custom infrastructure. Ona’s persistent environments built into Codex could make this dramatically easier.
- Memory is the bottleneck, not intelligence. The models are already smart enough for most business tasks. What’s been missing is the ability for agents to remember what they were doing, hold onto files, and maintain access to tools over time. This acquisition targets that gap directly.
- Security is part of the pitch. OpenAI specifically called out “secure” persistent environments. For industries like law, healthcare, and finance, where data handling rules are strict, this matters. Agents that lose state also lose audit trails. Persistent environments can fix that.
- Codex becomes an agent platform, not just a coding tool. Codex started as a code generation product. With Ona’s infrastructure baked in, it’s evolving into something bigger: a platform where developers build agents that actually run in the cloud, do real work, and stay alive between sessions.
- The competitive landscape just shifted. Google, Anthropic, and Microsoft are all pushing agent capabilities. OpenAI acquiring dedicated infrastructure for persistent agents is a clear bet that the next wave of AI value comes from agents that work autonomously over long periods, not just chatbots that answer questions.
The hot take
This is the most important AI acquisition of the year so far, and most people will sleep on it because it’s infrastructure. Not flashy. No new model. No viral demo. But here’s the thing: the company that solves persistent agent compute wins the enterprise market. Period. Every serious business workflow takes longer than a single API call. Every one. OpenAI clearly understands that the “make the model smarter” phase is giving way to the “make the agent actually useful in production” phase. Ona is a bet that plumbing matters more than parlor tricks. They’re right.
The Agency OS play
If you run any kind of service business, this week is a good time to audit your workflows for “long-running” candidates. Pull up a whiteboard (or a doc, whatever works) and list every process in your business that takes more than a day from start to finish. Compliance reviews. Client onboarding. Monthly reporting. Vendor vetting. Research projects. These are the workflows that persistent AI agents will handle first, and you want to know which ones are yours before the tools fully land.
Start experimenting with Codex now if you haven’t already. You don’t need to wait for the Ona integration to go live. Get familiar with how Codex handles tasks, how it structures agent workflows, and where it currently breaks down on longer jobs. That firsthand experience will put you months ahead of competitors who wait for the polished product launch. Build small. A three-step agent that gathers data, processes it, and drafts a summary is enough to learn the patterns.
Finally, think about what “persistent” means for your clients. If you serve law firms, imagine an agent that monitors regulatory changes across jurisdictions for weeks and flags relevant updates. If you work with ecommerce brands, picture an agent that tracks competitor pricing daily and adjusts recommendations. If you’re in finance, think about an agent that reconciles transactions across accounts over an entire quarter without losing context. The specific workflow matters less than the habit of asking: what would my business look like if an AI agent could work on something for days without stopping? Start designing for that future now. The infrastructure is coming faster than you think.