OpenAI has introduced Frontier, a platform it says helps enterprises build, deploy and manage AI agents that can “do real work.”
According to OpenAI, the launch is a response to what it calls an “AI opportunity gap” between what models can do and what organizations can reliably deploy across their existing systems and governance.
The firm said HP, Intuit, Oracle, State Farm, Thermo Fisher and Uber are among the first to adopt Frontier. It also said customers including BBVA, Cisco and T-Mobile have already piloted the platform’s approach.
OpenAI’s pitch: close the gap between pilots and production
In its announcement, OpenAI said enterprises already struggle with “disconnected systems and governance spread across clouds, data platforms and applications.” It argued that as agents proliferate, each can become “isolated in what it can see and do,” sometimes adding complexity instead of reducing it.
It added that Frontier is meant to provide an end-to-end way to “build, deploy, and manage agents.” The company compares agent management to how organizations onboard and manage employees: onboarding, institutional knowledge, feedback and access boundaries.
Shared business context: a “semantic layer” across enterprise systems
OpenAI said Frontier connects siloed data warehouses, CRM systems, ticketing tools and internal applications to provide agents with a shared “business context.”
The company described this as a “semantic layer for the enterprise” that agents can reference to understand information flows, decision points and outcomes.
On its Frontier product page, OpenAI positioned “Business Context” as a layer that links enterprise systems so agents can work with the same information people do and build “durable institutional memory over time.”
An execution environment plus evaluation loops
OpenAI said Frontier provides an “agent execution environment” where agents can work with files, run code and use tools. It also said agents can build “memories,” turning prior interactions into context that improves performance over time.
OpenAI also emphasized built-in “evaluation and optimization loops” intended to show what works and what does not, so agents can improve with experience and remain useful as workflows change.
Identity, controls and observability aimed at enterprise governance
OpenAI’s Frontier page says enterprises can assign agent identities and scope access “to exactly what each task requires,” tying the concept to enterprise IAM.
The same page says agent actions are “visible and auditable” and that built-in monitoring and “detailed logs” can provide traceability and control across the platform.
“Open standards” and a partner ecosystem
OpenAI said Frontier is “built on open standards” and is designed so software teams can “plug in and build agents” that benefit from the same shared context and controls. In the announcement, OpenAI argued this matters because many agent apps fail when they lack context, face complex permissions or require one-off integrations.
OpenAI also introduced a Frontier Partners cohort including Abridge, Clay, Ambience, Decagon, Harvey and Sierra and said it plans to expand the program over time.
Competitive pressure: control planes are converging
While OpenAI is pitching Frontier as a cross-system management layer, Microsoft has separately positioned Agent 365 as “the control plane for AI agents,” emphasizing unified observability for agent fleets through telemetry, dashboards and alerts.
Anthropic is taking a different route with Claude Cowork, which it describes as “Claude Code for the rest of your work,” positioning it as a desktop agent that can work inside a user-selected folder and carry out tasks beyond coding.
Anthropic says Cowork is available in research preview for Team and Enterprise plans, extending its agent tooling beyond coding workflows.
Salesforce is also positioning its Agentforce product as an enterprise agent platform that can work across business systems using governed data and controls.
Madhav Thattai, EVP and GM of Salesforce AI, describes Agentforce as an “operating system for humans and agents” and emphasizes shared business context, deterministic controls, observability into actions and orchestration across workflows.
Pricing and availability
OpenAI said Frontier is available to a limited set of customers today, with broader availability coming “over the next few months.” OpenAI did not publish pricing in its Frontier announcement post.