OpenAI unveils Frontier, an enterprise platform to build and manage AI agents

OpenAI has introduced Frontier a new enterprise platform to build, deploy, and manage AI agents that do real work inside organisations, according to the company. The launch underscores the shift from experimental pilots to production AI, with OpenAI positioning agents as “AI coworkers” that can be onboarded, given context, and governed much like employees.

What Frontier brings to the enterprise

Frontier is designed to consolidate how businesses operate multiple agents, providing shared context, role-specific permissions, and continuous feedback loops so performance improves on real tasks. OpenAI says the platform runs across local infrastructure, enterprise clouds, and OpenAI-hosted runtimes, prioritising low latency access to its latest models. The company adds that it is built on open standards so software teams and third-party builders can integrate agents within existing workflows.

• Business context, connecting systems of record like data warehouses, CRM tools, and internal apps to give agents durable institutional memory.
• Agent execution, a dependable environment for reasoning over data, working with files, running code, and coordinating multiple agents in parallel.
• Evaluation and optimisation, built-in monitoring and learning loops that show what is working and improve outcomes over time.
• Security and governance, enterprise IAM for agent identities, auditable actions, and compliance with standards such as SOC 2 Type II and ISO/IEC 27001, 27017, 27018, and 27701.

Early adopters include HP, Intuit, Oracle, State Farm, Thermo Fisher Scientific, and Uber, while existing customers such as BBVA, Cisco, and T‑Mobile have piloted elements of the approach. OpenAI is also working with a small cohort of Frontier Partners, including Abridge, Clay, Ambience, Decagon, Harvey, and Sierra, to co-design solutions and support deployments.

How does Frontier fit into existing enterprise stacks

OpenAI positions Frontier as a single pane of glass for “agent sprawl”, the fragmentation that occurs when multiple teams run siloed agents without shared data or controls. Industry coverage notes that Frontier is meant to unify management across agents from OpenAI and third parties, using open standards to plug into existing workflows. The company says agents can move between local, cloud, and OpenAI-hosted environments without forcing teams to rewrite processes.

Evidence from early deployments

OpenAI cites results from customers over the past few years. At a major semiconductor manufacturer, agents reportedly cut chip optimisation work from six weeks to one day. A global investment firm saw agents deployed end to end across sales processes, opening up over 90 percent more time for salespeople. A large energy producer recorded up to a five percent increase in output, which the company says translated to over a billion in additional revenue.

State Farm’s digital leader Joe Park said pairing the Frontier platform and OpenAI’s deployment expertise with employees is helping the insurer accelerate AI capabilities to better serve customers. OpenAI says Frontier is available to a limited set of customers now, with broader availability in the coming months.