OpenAI announced on June 11 that it has agreed to acquire Ona, the German startup previously known as Gitpod, with the entire team folding into the Codex effort. Ona's platform runs AI agents inside pre-configured cloud sandboxes with access controls and audit trails. OpenAI did not disclose financial terms, and said the transaction is contingent on standard closing requirements.
The point of the deal is duration. Codex today runs in a developer's editor and stops when the editor stops — useful for autocomplete and small refactors, far less useful for a multi-hour migration or a long test loop. Ona's infrastructure is built to keep agents alive in the cloud after the human walks away, which is the specific gap between today's coding copilots and the autonomous workers OpenAI keeps advertising in its IPO materials.
This fits a broader race to move AI coding from inline suggestions to long-running agents that can read a ticket, write the code, run the tests, and open a pull request. GitHub launched its own background agents earlier this year, Anthropic shipped Claude Code's cloud sandboxes, and Cognition's Devin has rebuilt itself around the same primitive. Owning a managed sandbox provider, instead of renting one, removes a layer of cost and a layer of supply risk for OpenAI.
Takeaway for learners: when an AI tool moves from "assistant you talk to" to "worker you check on later," the bottleneck becomes infrastructure, not intelligence. The companies winning this round are the ones that solve where the agent runs, what it can touch, and how you audit it after — skills that look more like devops than prompt engineering.