OpenAI just admitted, in public, what every agent builder already knew: the bottleneck isn't model intelligence anymore. It's infrastructure. OpenAI has agreed to acquire Ona, a cloud infrastructure company (formerly Gitpod), to give Codex persistent environments that keep running after you close your laptop. Announced June 11, still pending regulatory approval. Why this matters more than it looks: → Most AI coding agents today work in session-based bursts — context fills up, the session times out, you start over. Fine for quick fixes. Useless for multi-day codebase migrations or modernization projects. → Codex usage has exploded: 5M+ weekly users in June, up from ~3M in April. Knowledge workers outside engineering now make up ~20% of that base, growing at 3x the rate of the core dev audience. → On the Ona side, enterprise agent usage grew 13x in 2026, with deployments at a major U.S. bank and multiple pharma and sovereign-wealth clients already in production. → The key detail: Ona's "customer-controlled execution" model lets agents run inside an organization's own cloud, with OpenAI providing the model and orchestration layer. That's the difference that gets compliance teams to sign off. This is also OpenAI's sixth acquisition in roughly a year — io, Software Applications, Torch, Promptfoo, now Ona — each one filling a gap that pure model capability can't solve alone. The bigger signal: as agents start running for hours or days with standing access to credentials and infrastructure, the real competition shifts from "whose model is smartest" to "whose agent can run reliably, securely, and continuously without a human babysitting the session." Whoever owns that execution layer gets the enterprise contracts. Nothing changes for Codex users this week. But if you're building or buying agent infrastructure, this is the layer to watch in the second half of 2026. https://lnkd.in/dZ__R_Uz #AI #OpenAI #Codex #AIAgents #EnterpriseAI #CloudInfrastructure