OpenAI unveils Frontier to deploy AI coworkers at scale

OpenAI has launched Frontier, a platform for enterprises to build, deploy, and manage AI agents across their organisations.

Frontier is described as an end-to-end layer for running so-called AI coworkers. It is intended to address barriers that can prevent organisations from moving beyond pilots, including fragmented governance and disconnected systems.

The company cited internal survey data suggesting a shift in day-to-day work: 75% of enterprise workers reported that AI helped them complete tasks they could not do before. It also said demand is coming from a broad set of business functions, not only technical teams.

Examples in manufacturing, finance, and energy were presented. At a major manufacturer, it said agents reduced production optimisation work from six weeks to one day. A global investment company, it said, used agents across its sales process and freed up more than 90% of salespeople's time for customer engagement. At a large energy producer, it said agents increased output by up to 5%, which it linked to more than a billion in additional revenue.

Deployment gap

OpenAI argues that enterprises now face an "AI opportunity gap" because agents often run in isolated environments. Many organisations operate across multiple clouds, data platforms, and applications. In that structure, OpenAI said, each new agent can add complexity if it lacks context and cannot access the systems needed to complete work.

OpenAI also linked the deployment challenge to the pace of product releases in the AI market. It said it ships something new roughly every three days, and that the pace is increasing. Teams, it added, are still developing the operational knowledge required to run agents in production environments.

Frontier centres on shared context, identity and access controls, and evaluation of agent performance. OpenAI said the platform is designed around familiar workplace elements, including onboarding, learning with feedback, and clear permissions and boundaries.

How Frontier works

Frontier connects data sources and business systems, including data warehouses, CRM tools, ticketing platforms, and internal applications. The goal, OpenAI said, is for agents to share context across the business rather than operate as stand-alone tools.

The platform adds a semantic layer that agents can reference when interacting with these systems. OpenAI said this improves how agents interpret and act on business information.

Agents built on Frontier can also access a computer and tools, OpenAI said, allowing them to plan and take actions in workflows rather than only generate text responses.

Frontier supports different runtime environments. OpenAI said AI coworkers can run locally, on enterprise cloud infrastructure, or in OpenAI-hosted runtimes. It also said Frontier includes evaluation and optimisation features that allow agents to learn from experience and improve over time.

Security and governance are also central to the product's positioning. OpenAI said each AI coworker operates within explicit permissions and boundaries, and that enterprises can apply identity and access management controls through the platform.

Integrations and interfaces

OpenAI said Frontier integrates with systems that enterprises already use, avoiding new formats and reducing the need to replace agents and applications already deployed.

Agents built with Frontier can be accessed through different interfaces, including ChatGPT, workflows with Atlas, and existing business applications, according to OpenAI.

Several large organisations have started work with Frontier, OpenAI said. Early adopters named by the company include HP, Intuit, Oracle, State Farm, Thermo Fisher, and Uber. Dozens more, it added, are piloting the platform.

Services and partners

Alongside the platform, OpenAI said it pairs Forward Deployed Engineers with enterprise teams to share best practices for building and running agents in production environments. It added that this collaboration informs ongoing research and model development.

OpenAI also positioned Frontier as part of a wider ecosystem, saying the platform is built on open standards that allow software teams to connect their own tools and agents while benefiting from shared context.

In addition, a group of Frontier Partners will work with it on solutions and deployments. Named partners include Abridge, Clay, Ambience Healthcare, Decagon, Harvey, and Sierra.

Frontier is available to a limited set of customers, with broader availability expected in the coming months.