OpenAI Acquires Ona to Let Codex Run Long-Running Agents Inside Customer Clouds

Historically, tech giants have bought infrastructure specialists to cross the last mile into risk-averse enterprises—Microsoft’s 2018 $7.5 bn GitHub deal and IBM’s 2019 $34 bn Red Hat buy are parallels where ownership of trusted developer platforms unlocked corporate budgets. OpenAI’s move signals a similar pivot: large-language-model vendors must now solve for governance, data residency and auditability, not just raw intelligence. By keeping agents inside customer-controlled VPCs, Ona’s model tries to reconcile 1990s-era perimeter-security instincts with 2020s cloud-native workflows—a tension likely to define AI deployment for decades. If autonomous software engineers do become as ubiquitous as human ones, this moment—when execution moved from a user’s laptop to an auditable, policy-enforced cloud sandbox—could be remembered as the step that made regulators, insurers and boards comfortable letting code write code. Alternatively, if history echoes the expert-system bust of the late 1980s, the acquisition may read as another exuberant bet; but on a 100-year horizon, the steady march is toward delegating more cognitive labor to machines, and securing the execution environment is a prerequisite each cycle.