OpenAI has agreed to acquire Ona, a startup focused on infrastructure for AI agents, in a deal that is still subject to regulatory approval. The acquisition is aimed squarely at boosting Codex, OpenAI's AI-powered coding assistant, which has increasingly been used to handle complex, multi-step programming tasks.
The core problem Ona is designed to solve: AI coding agents need somewhere stable and secure to actually do work over an extended period. Running a quick code suggestion is one thing; autonomously managing a software project over hours or days requires a persistent, controlled environment where the agent can spin up processes, store state, and keep working without losing context. According to Pulse 2.0, Ona provides exactly that — secure, persistent agent workspaces.
InfoWorld frames the deal in broader terms, describing it as OpenAI buying Ona to help "rein in" AI agents — suggesting the technology also addresses the challenge of keeping autonomous systems predictable and contained as they take on longer-horizon tasks.
MSN reports that the acquisition could bring "big changes" to Codex by giving it access to the right kind of environment for these long-running agentic tasks — something the product has lacked as OpenAI pushes it further into autonomous software development territory.
The move reflects a wider industry race to make AI agents genuinely useful for real engineering work, not just one-off prompts. As AI assistants graduate from answering questions to independently writing, testing, and deploying code, the underlying infrastructure — where agents live and operate — becomes just as important as the models themselves. This acquisition signals that OpenAI is betting the future of Codex on agentic, long-duration workflows, and is willing to buy the plumbing to make it happen.