OpenAI Buys Ona to Power Long-Running Codex Agents
Key insights
- OpenAI is acquiring Ona, formerly Gitpod GmbH, to give Codex agents cloud sandboxes that keep running after a developer's machine shuts down.
- Ona's infrastructure enables Codex to handle multi-hour or multi-day engineering tasks, such as codebase refactors and test runs, without interruption.
- CEO Johannes Landgraf and the full Ona team will join OpenAI's Codex group once regulatory approvals are finalized; deal terms were not disclosed.
Why this matters
Persistent cloud sandboxes remove the compute ceiling that has constrained AI coding agents to the limits of a developer's session length and local hardware, opening the door to engineering tasks that would previously require constant supervision. The acquisition signals that OpenAI sees long-running agentic infrastructure, not model capability alone, as the next competitive moat against Anthropic's growing developer and enterprise base. For technical leaders and founders building on AI coding tools, the shift toward multi-day autonomous agents raises the stakes on task specification and output review, since errors now compound over unattended runs before a human can intervene.
Summary
OpenAI is acquiring Ona (formerly Gitpod GmbH), a German startup that builds cloud sandboxes for AI agents. Ona's sandboxes stay live after a developer shuts down their machine, giving agents access to compute a laptop can't match.
The acquisition targets Codex, enabling multi-day tasks, like codebase refactors and test runs, without interruption.
Essentially: (OpenAI, Ona) are building the persistence layer for unattended AI engineering work.
- Sandboxes outlive the local session, removing the hardware ceiling for agents.
- CEO Johannes Landgraf and the Ona team join OpenAI's Codex group pending regulatory approval.
- Financial terms were not disclosed.
Long-running autonomous agents, not faster code generation, are the real competitive front between OpenAI and Anthropic.
Potential risks and opportunities
Risks
- Anthropic could accelerate its own persistent-sandbox infrastructure for Claude's developer tools, tightening the window of OpenAI's first-mover advantage with unattended Codex agents.
- Multi-day autonomous agents running unattended increase the blast radius of a misspecified task, exposing enterprise Codex users to significant unreviewed code changes before a human can review output.
- Key Gitpod-era engineers face integration risk during the regulatory-approval lag period, potentially slowing Codex's roadmap into late 2026 if the team attrits post-acquisition.
Opportunities
- Cloud infrastructure providers could see increased enterprise demand as persistent agent sandboxes running for days require substantially more compute than session-based tools.
- Enterprises investing now in roles focused on task specification and output evaluation gain direct leverage as Codex handles unattended multi-day coding work at scale.
- Anthropic faces pressure to match OpenAI's persistent sandbox capability, likely accelerating its own agentic infrastructure investment for developer-facing tools in the second half of 2026.
What we don't know yet
- Financial terms of the acquisition were not disclosed, leaving the deal's valuation unknown to the public.
- Whether Ona's sandbox technology will be available to third-party developers via API, or remain exclusive to OpenAI's Codex, is unaddressed in the announcement.
- The timeline for regulatory approval completion is unspecified, leaving the integration schedule unclear as of June 2026.
Originally reported by techtimes.com
Read the original article →Original headline: OpenAI Acquires Cloud-Agent Sandbox Platform Ona to Power Long-Running Autonomous Codex Sessions