OpenAI launches Frontier to manage your AI coworkers

Most big companies are running into the same awkward wall with AI: the models keep getting smarter, but reliable AI teammates are still nowhere to be found.

Pilot projects are everywhere, but the moment an agent needs to touch a messy CRM, pull context from scattered data warehouses, and actually complete an end-to-end task across multiple clouds, the whole setup falls apart.

OpenAI just launched Frontier, a new enterprise platform built to help businesses build, deploy, and manage AI agents that can actually do real work.

Heavyweights like HP, Intuit, Oracle, State Farm, Thermo Fisher, and Uber are among the first to sign on, while BBVA, Cisco, and T-Mobile have already been testing the underlying approach.

Onboarding AI like human hires

Instead of treating agents like isolated software widgets, OpenAI is framing Frontier around how companies scale human employees.

New hires need shared company context, access to tools, clear permissions, and feedback loops to understand what good work looks like. Frontier attempts to bundle all of that into a unified layer.

It connects fragmented data across clouds, ticketing systems, and enterprise apps, acting as a shared semantic layer so agents understand how the business actually functions. From there, AI coworkers can run code, interact with files, and build up memories over time within secure sandboxes.

Crucially, each agent gets an explicit identity and strict permission boundaries, which matters a lot for teams deploying in regulated spaces.

Borrowing Palantir's playbook

Technology alone isn't going to bridge the enterprise gap, and OpenAI knows it.

In a clear nod to Palantir's famous playbook, OpenAI is deploying Forward Deployed Engineers (FDEs) to work side-by-side inside customer teams. These engineers will help build and stabilize production agents on the ground.

Even better for OpenAI, these FDEs act as a direct pipeline back to OpenAI Research. Real enterprise bottlenecks will funnel straight back into training the next generation of models.

OpenAI is also bringing in a tight circle of Frontier Partners, including AI-native startups like Harvey, Sierra, Clay, Decagon, Ambience, and Abridge, to plug specialized workflows into the platform.

Frontier is rolling out to a limited group of enterprise customers today, with broader access slated for the coming months.

Why it matters

We think Frontier represents a massive pivot in OpenAI's enterprise playbook. Selling raw model tokens is a commodity race; owning the orchestration layer and enterprise context is where real defensibility lives.

By sending Forward Deployed Engineers into customer offices, OpenAI is admitting that raw intelligence isn't enough. Enterprises need hand-holding to wire models into messy legacy tech stacks. It is a smart, high-touch move that turns OpenAI from an API vendor into an indispensable enterprise systems partner.

For enterprise operators, Frontier could finally provide an escape route from endless, fragmented proof-of-concept purgatory. Being able to run agents across hybrid clouds without ripping out existing databases is a huge plus.

However, the risks are just as real. Handing OpenAI the keys to your internal semantic layer, execution environments, and workflow orchestration creates immense platform lock-in. It also puts OpenAI on a direct collision course with enterprise software giants like Salesforce and its biggest backer, Microsoft, both of whom want to be the default brain of corporate workflows.