OpenAI acquires Ona: enterprise AI agents in the cloud

The operation in a nutshell: who is Ona and why does OpenAI want her

The June 11, 2026, OpenAI has made public its intention to acquire Ona, a startup focused on creating secure and persistent cloud environments. The announcement comes at a time when competition in generative AI is intensifying. Therefore, the move is not surprising to those who follow the sector closely.

She is not a language model company. On the contrary, she deals with infrastructure: execution environments that maintain state, context, and security over time. This is exactly what current AI agents lack to operate in real enterprise scenarios. In fact, most agents today operate in stateless sessions, without persistent memory between one execution and the next.

Integration will take place within Codex, OpenAI's AI coding platform. However, the implications go far beyond writing code. Ona's persistent environments enable agents capable of performing complex tasks over long time horizons, with secure access to corporate resources.

Timeline of a strategic acquisition

To understand the significance of the operation, it's helpful to place it in a logical sequence. Over the past eighteen months, OpenAI has accelerated its transition from a model provider to a platform for AI agents. First came Operator, then the expansion of Codex as a standalone tool. Thus, the acquisition of Ona represents the third act of this strategy.

She had developed technology to isolate and protect agent execution environments. Specifically, she ensured that each agent operated in a separate context, with controlled access to company data. This addresses one of the most critical problems in enterprise AI: security in managing sensitive information.

Similarly to what happened with other acquisitions in the sector, OpenAI is not just acquiring technology. It is acquiring a team with specific expertise in cloud security and distributed environment orchestration. Therefore, the transaction has value beyond the code.

Winners and Losers in the Enterprise Agent Market

Who benefits the most from this acquisition? First and foremost, OpenAI consolidates its position in the enterprise segment, where competition with Microsoft, Google, and Anthropic is most intense. Furthermore, Codex becomes a more credible tool for enterprise IT teams seeking secure automation.

Potential losers include startups that were building persistence and security layers on top of OpenAI's APIs. With Ona integrated natively, some of that middleware market could shrink. Conversely, those who bet on multi-model architectures might find new arguments for diversification.

For user companies, the picture is more nuanced. On one hand, they will have access to more capable and secure agents. On the other hand, dependence on a single vendor increases. This is a trade-off that every organization will have to carefully evaluate, as also highlighted by recent analyses of Gartner on enterprise AI.

A persistent environment, for someone who isn't a developer, refers to a setup or system that remains the same and consistent over time, even after you close and reopen it. Think of it like this: * **A regular notebook vs. a digital note-taking app:** If you write in a physical notebook, your notes are always there, in the same place. That's a persistent environment for your notes. If you use a digital app that automatically saves your work, closing and reopening the app will show you the same notes. * **A freshly reset computer vs. your own computer:** When you first get a computer, it's in a “default” state. If you install programs, change settings, or save files, your computer becomes "persistent" with your customizations. When you turn it on again, everything you did is still there. * **A game you're playing:** When you save your progress in a video game, that saved state creates a persistent environment. The next time you play, you can pick up right where you left off, with all your characters, items, and progress intact. In essence, when something is "persistent," it means its state or configuration doesn't disappear when you stop interacting with it. It's saved and will be there the next time you access it.

The technical concept deserves an accessible explanation. A traditional AI agent receives an input, processes a response, and then forgets everything. It's like a consultant who starts from scratch every day, with no memory of previous sessions. Therefore, it cannot handle complex projects that develop over time.

An agent with a persistent environment, on the other hand, maintains context between sessions. It can monitor a process for days, update documents, interact with external systems, and pick up where it left off. This is the leap that makes AI truly useful for structured enterprise workflows.

Furthermore, the security component is fundamental. Ona's environments ensure that each agent operates in an isolated space with defined permissions. Consequently, the risk of unauthorized access to sensitive data is significantly reduced. For SMEs managing customer data or financial information, this is a non-negotiable requirement.

SHM Studio reading: real opportunity or premature hype?

We of SHM Studio We are following the evolution of generative AI with a pragmatic approach. The acquisition of Ona is genuinely significant. However, it is important to distinguish between what is available today and what will be available in twelve or twenty-four months.

Today, most Italian SMEs have not yet adopted first-generation AI agents. Therefore, talking about enterprise persistent agents may seem premature. In reality, the right time to understand these technologies is first that they become mainstream, not later. Those who arrive late pay the price of forced adaptation.

The most immediate implications concern IT teams and digital managers of medium-sized businesses. In particular, those who manage repetitive workflows, automated reporting, or integration between different systems should begin evaluating agentive automation scenarios. AI solutions that we're accompanying in SMEs clients are already evolving in this direction.

Next moves: what to monitor in the coming months

The integration of Ona into Codex will not be immediate. Technological acquisitions require time to produce stable and documented products. Therefore, it is realistic to expect a roadmap to develop between late 2026 and 2027.

There are three signals to monitor closely. First, the evolution of Codex APIs: if OpenAI releases endpoints for managing persistent environments, it will be a signal that integration is mature. Second, the competitive response from Google DeepMind and Anthropic, who are developing similar solutions according to MIT Technology Review. Third, adoption by European system integrators, who often anticipate demand from mid-tier SMEs.

For companies that are building their own digital strategy, this is the time to map internal processes that could benefit from agentive automation. This isn't about investing in immature technology today. It's about being prepared when that maturity arrives.

Operational implications for Italian B2B SMEs

Italian B2B SMEs have specific characteristics that make AI agents particularly relevant. Many manage hybrid processes, with legacy systems, Excel spreadsheets, and cloud tools that do not communicate with each other. Therefore, an agent capable of operating persistently across these environments could significantly reduce operating costs.

The most immediate use cases include: automated customer reporting management, continuous monitoring of KPIs across multiple platforms, and integration between CRM and billing systems. Furthermore, in B2B retail, persistent agents could handle automatic inventory reordering or sales performance analysis by category.

Who is responsible for SEO e content will find applications in agents that monitor positioning over time and suggest contextual optimizations. Likewise, those who manage Google Ads campaigns o LinkedIn campaign It will benefit from agents that continuously optimize budget allocations without manual intervention.

In any case, the prerequisite is to have a structured digital presence and documented processes. AI agents amplify what exists. They do not replace a lack of solid digital foundations. Therefore, those who have not yet completed their basic digital transformation should start there, before thinking about agents.

The construction site still open: what remains unresolved

Despite this, open questions remain that warrant critical attention. The first concerns data governance. Persistent agents access company information over time: who controls what they see, and for how long? European data protection regulations, particularly GDPR, impose precise constraints that American platforms have yet to fully demonstrate compliance with in agentive scenarios.

The second issue concerns the transparency of decisions. An agent operating autonomously for days on a business workflow makes decisions. Who is responsible for them? Finally, there's the issue of vendor lock-in: the more a company integrates OpenAI agents into its processes, the more difficult it becomes to change platforms in the future.

These are not reasons not to adopt technology. They are, however, reasons to do so with awareness and with the support of partners who understand both the Italian technological and regulatory context. To delve deeper into these topics or to evaluate specific scenarios for your company, it is possible Contact the SHM Studio team to explore the in-depth articles on our blog.