OpenAI launches Frontier, an enterprise agent platform

OpenAI launched Frontier, an enterprise platform for building, deploying and managing AI agents that the company pitched as treating agents like employees rather than isolated tools: each agent gets an onboarding process, feedback loops and improvement cycles modelled on how organisations manage human staff. The platform includes auditing, explicit permissions and logged actions for every agent interaction, aimed at enterprises that had been running scattered AI pilots without a consistent way to deploy, monitor or govern them.

Notably, Frontier was described as model-agnostic at the agent layer, able to run agents built on models from OpenAI, Google, Microsoft and Anthropic as well as custom-built agents, rather than restricting enterprises to OpenAI’s own models. TechCrunch reported HP, Intuit, Oracle, State Farm, Thermo Fisher and Uber as confirmed early users, with BBVA, Cisco and T-Mobile named as running pilots.

The launch positioned OpenAI more directly against enterprise software incumbents such as Microsoft, Salesforce and ServiceNow, which had each built their own agent-orchestration layers over the preceding two years, and reflected a broader shift among frontier labs from selling raw model access toward selling the surrounding infrastructure — permissions, monitoring, deployment tooling — that large organisations said they actually needed to put agents into production.