OpenAI launches Frontier, an AI agent platform that can reshape enterprise software

Frontier appears to be OpenAI’s attempt to become an “enterprise operating system,” offering a unified platform for building agents that can move between applications, execute workflows, and make decisions. In a blog post Announcing the new platform, OpenAI says Frontier can connect databases and business systems of record for things like CRM software, HR, ticketing tools, and other internal applications, and then allow AI agents to run processes across these systems.

The company described Frontier as “an enterprise semantic layer that all AI coworkers can refer to to work and communicate effectively.” Human employees can work on the same platform, so both humans and AI can access the same data and tools, with similar access controls and security provisions, she said.

OpenAI has tapped a number of well-known Fortune 500 companies as initial clients for Frontier, including Intuit, State Farm, Thermo Fisher, and Uber.

Frontier’s debut follows the movements of humanitarian projects

OpenAI’s Frontier debut comes on the heels of a series of moves by rival Anthropic to also make it easier for enterprise customers to build agents that use other business software and run corporate workflows, as well as create custom software. Last month, Anthropic debuted Claude Cowork, which allows users to use the Claude AI model in an agentic manner across popular business software. And this week, Anthropic launched open source plugins for Cowork targeting tasks in specific professional sectors, such as legal work or marketing.

The joint rollout of Anthropic’s new agent AI systems and OpenAI for enterprises has spooked investors in traditional large enterprise SaaS companies, such as Salesforce, ServiceNow, Workday, SAP and Microsoft. The concern is that AI-native startups, such as OpenAI and Anthropic, will increasingly disintermediate the relationship that large SaaS providers have with their customers and will avoid the need for those customers to upgrade to the AI ​​agent offerings offered by the SaaS giants themselves. This could dampen the growth prospects of these SaaS companies.

In some cases, it may replace the need to have this SaaS software at all. For example, if a Frontier agent can execute sales workflows without a human logging into Salesforce, the “per seat” licensing fees that currently power the SaaS economy may lose their justification.

Fidji Simo, chief applications officer at OpenAI, said in a press conference that when she was CEO of Instacart, giving her teams access to the best AI tools meant having to evaluate hundreds of different software vendors and then go through a complex, time-consuming effort to integrate those tools into the organization’s workflow. “We spent months integrating each of the tools we chose,” she said, adding, “We didn’t even get what we really wanted, because each tool was good for one use case, but they weren’t integrated or talking to each other, so we were just reinforcing silos upon silos.”

Instead, she said she dreams of a single platform to create and manage all of the organization’s agents. “Now that I work at OpenAI, every CEO asks me, ‘Where is this all going?’ I tell them, ‘It’s about collaboration between humans and AI on one platform.’”

Can Frontier replace other AI agent platforms?

Simo insisted that the platform aims to embrace enterprise software vendors, not displace them. She describes Frontier as “an acknowledgment that we’re not going to build everything ourselves, we’re going to work with the building ecosystem alongside them, and we embrace the fact that companies are going to need a lot of different partners.”

For some software companies, Frontier could become an important distribution channel — a way “to get it into the big companies, and for the big companies to adopt these foreign solutions without fragmenting their systems further,” she said.

But companies like Salesforce have bet their future on AI agent platforms. Salesforce’s $1 billion “Agentforce” initiative envisions companies building fleets of independent agents that live directly inside their CRM software. Microsoft’s Copilot agents are designed to do the same thing across Microsoft 365 products. These companies are betting that customers will need agents deeply integrated into their “systems of record” — where the data actually resides — rather than a generic OpenAI agent sitting on top of each system.

This isn’t OpenAI’s first foray into this area, but it signals a philosophical shift. When the company launched “ChatGPT Enterprise” in 2023, the offering was strictly about empowering human employees. OpenAI now offers agents that are more concerned with automating workflows, such as logging into applications, executing tasks, and managing tasks without significant human intervention.