OpenAI’s Frontier wants to manage your AI agents – it could upend enterprise software, too

ZDNET’s key takeaways

• OpenAI’s Frontier emulates Palantir’s use of forward-deployed AI engineers.
• Frontier promises to also handle agent security features.
• Like Claude Cowork, Frontier threatens the traditional software industry.

OpenAI, which to date has made most of its money from consumer users of ChatGPT, on Thursday took a page from Palantir’s playbook, aiming to move deeper into enterprise sales. Palantir is arguably the most successful enterprise AI software company, with revenue of more than $4 billion annually from government and business clients.

OpenAI unveiled Frontier, a framework for deploying enterprise artificial intelligence agents; the company said the offering will help companies overcome impediments to deploying agents within organizations.

Also: True agentic AI is years away – here’s why and how we get there

“We’re introducing Frontier, a new platform that helps enterprises build, deploy, and manage AI agents that can do real work,” said OpenAI in its press release. The framework, it said, “gives agents the same skills people need to succeed at work: shared context, onboarding, hands-on learning with feedback, and clear permissions and boundaries.”

(Disclosure: Ziff Davis, ZDNET’s parent company, filed an April 2025 lawsuit against OpenAI, alleging it infringed Ziff Davis copyrights in training and operating its AI systems.)

Borrowing Palantir’s use of forward-deployed engineers

While details are scant, the Frontier approach appears to lean heavily on “forward-deployed engineers,” teams that work on-site with customers to tweak the software to an organization’s specific processes and to incorporate what’s learned into ongoing software development.

“We pair OpenAI forward-deployed engineers (FDEs) with your teams, working side by side to help you develop the best practices to build and run agents in production,” said OpenAI. “The FDEs also give teams a direct connection to OpenAI Research.” According to OpenAI:

As you deploy agents, we learn not just how to improve your systems around the model. We also learn how the models themselves need to evolve to be more useful for your work. That feedback loop, from your business problem to deployment to research and back, helps both sides move faster.

Palantir founder and CEO Alex Karp has featured forward-deployed engineers as a key asset that helps the company to integrate its software with customers’ processes. “The reason you end up with forward-deployed engineers, especially in the beginning, is you have to extend the product,” he said during a keynote address in September.

The large language model, such as OpenAI’s GPT, said Karp, “is a raw material that has to be processed,” and is often “over-hyped” as the savior of businesses. It takes engineering to make it real in deployment, he indicated.

Just this week, Palantir touted a partnership with Accenture that will “deploy Palantir’s forward-deployed engineers alongside 2,000+ Palantir-skilled Accenture professionals to accelerate AIP [Palantir AI platform] deployment across healthcare, telecommunications, manufacturing, consumer goods, and financial services.”

Also: Enterprises are not prepared for a world of malicious AI agents

OpenAI’s move from being merely an LLM provider to selling a platform for agent deployment will doubtless bring it into conflict with Palantir, C3.ai, and other commercial software firms that have spent years deploying AI with an industry focus.

Frontier is currently being used by a “limited set of customers,” said OpenAI, including HP, Inc., Intuit, Oracle, Thermo Fisher, and Uber. It will become available more broadly  “over the next few months,” the company said.

Talking up ontologies and agentic security

Frontier also appears to borrow from Palantir’s approach to learning the specifics of industries. Frontier, said OpenAI, will connect various data sources with what it describes as “a semantic layer for the enterprise that all AI coworkers can reference to operate and communicate effectively.”

That echoes the ontologies that Palantir has promoted as a unique asset, offering a framework for how AI understands the fundamental business terms of an industry.

Also: While Google and OpenAI battle for model dominance, Anthropic is quietly winning the enterprise AI race

At the same time, OpenAI is clearly moving into the domain of cybersecurity and authorization. The Frontier framework identifies and authorizes AI agents, OpenAI said.

“Frontier makes sure AI coworkers operate within clear boundaries. Each AI coworker has its own identity, with explicit permissions and guardrails. That makes it possible to use them confidently in sensitive and regulated environments.”

Traditional cybersecurity vendors such as Palo Alto Networks have told enterprises that they will be the ones to help identify, authenticate, and grant permissions and place restrictions on AI agents.

Also: Claude Cowork automates complex tasks for you now – at your own risk

Threatening traditional software sales

Frontier is both an opportunity and a threat for existing commercial software firms such as Palantir and Palo Alto Networks. On the one hand, it is highly likely that OpenAI will need to partner with those firms to close its own experience gap in enterprise sales. That would require OpenAI to bring those vendors into its deals to make up for what Frontier lacks.

On the other hand, Frontier is similar to another program that has cast a shadow over traditional software, the Claude Cowork program from Anthropic. Like Cowork, Frontier could make traditional software packages less important than large language models, which will serve as the front-end user interface for years to come.

Also: I let Anthropic’s Claude Cowork loose on my files, and it was both brilliant and scary

By replacing the user interface of business, Frontier and Cowork could make every commercial software package less important as a choice for business owners. That prospect has lead to a sell-off in recent weeks of software company stocks, which Bloomberg has labeled the “SaaSpocalypse.”