OpenAI is making its most aggressive move into the enterprise world yet with the launch of Frontier, an enterprise platform for building, deploying, and managing AI agents that can run other software such as Salesforce and Workday.
Frontier appears to be OpenAI’s goal to become an “enterprise operating system,” providing an integrated platform for building agents that can interact with apps, run workflows, and make decisions. In a blog post announcing the new platform, OpenAI said Frontier connects databases, business systems of record such as customer relationship management software and human resources resources, ticketing tools, and other internal applications, allowing AI agents to run processes on these systems.
The company describes Frontier as “an enterprise-ready semantic layer that all AI colleagues can reference to operate and communicate effectively.” He said human employees can work on the same platform, so both humans and AI can access all the same data and tools with similar access controls and security provisions.
OpenAI has signed a number of prominent Fortune 500 companies as early Frontier customers, including Intuit, State Farm, Thermo Fisher, and Uber.
Frontier debut following the movement of human enterprises
OpenAI’s Frontier debut follows a series of moves by rival Anthropic to make it easier for enterprise customers to use other business software to build agents or launch bespoke software to run enterprise workflows. Last month, Anthropic debuted Claude Cowork, which allows users to use Claude AI models in an agent-like manner across popular business software. And this week, Anthropic released an open-source plugin for Cowork that targets tasks in specific areas of expertise, such as legal work and marketing.
Anthropic and OpenAI’s joint deployment of a new agent AI system for enterprises surprised investors in traditional large enterprise SaaS companies such as Salesforce, ServiceNow, Workday, SAP, and Microsoft. The concern is that AI-native startups like OpenAI and Anthropic will increasingly disintermediate the relationship between large SaaS providers and their customers, eliminating the need for those customers to upgrade to AI agent products offered by the SaaS giants themselves. That could hurt the growth prospects of these SaaS companies.
In some cases, you may not need this SaaS software at all. For example, if Frontier agents can perform sales workflows without a human logging into Salesforce, the “per-seat” licensing fees that currently underpin the SaaS economy could lose their legitimacy.
Fidji Simo, CEO of Applications at OpenAI, said in a media briefing that when he was CEO of Instacart, providing his team with access to the best AI tools required evaluating hundreds of different software vendors and incorporating those tools into enterprise workflows, a complex and time-consuming process. “We spent months integrating each of the tools we selected,” she added. “We didn’t even really get what we wanted, because each tool was good for one use case, but they weren’t integrated and didn’t work with each other, so we were just reinforcing the silos.”
Instead, he said he dreamed of one platform where he could create and manage all of his organization’s agents. “Now that I’m here at OpenAI, all the CEOs are asking me, where is this going? My answer is humans and AI working together on one platform.”
Could Frontier replace other AI agent platforms?
Simo argued that the platform is intended to embrace existing enterprise software vendors, not exclude them. She calls Frontier “a recognition that we’re not going to build everything ourselves, we’re going to build it in partnership with the ecosystem, and we accept the fact that companies are going to need different partners.”
For some software companies, he said, the frontier could become an important distribution channel, a way to “bring companies into larger companies and for larger companies to introduce these foreign solutions without further fragmenting their systems.”
But companies like Salesforce are betting their future on AI agent platforms. Salesforce’s multibillion-dollar “Agentforce” initiative envisions businesses building fleets of autonomous agents that reside directly within their CRM software. Microsoft’s Copilot Agent is designed to do the same thing across Microsoft 365 products. These companies are betting that their customers will want agents deeply embedded in their “systems of record” (where the data actually resides) rather than OpenAI’s generalist agent sitting on top of every system.
This isn’t OpenAI’s first foray into the enterprise, but it does mark a philosophical shift. When the company launched “ChatGPT Enterprise” in 2023, the pitch was strictly about empowering human employees. OpenAI currently offers agents focused on workflow automation, logging into applications, performing tasks, and managing tasks with little human interaction.
