OpenAI has introduced Frontier, a new enterprise platform designed to help organizations build, deploy, and manage AI agents that operate across real business systems with shared context and governance.

The announcement addresses what OpenAI describes as an “AI opportunity gap”: the growing distance between what advanced AI models are capable of and what enterprises can reliably deploy in production. While AI experimentation has accelerated across organizations, many teams struggle to turn increasingly powerful models into dependable, organization-wide systems.

Frontier shifts the focus from model intelligence to execution, emphasizing how AI agents are onboarded, evaluated, governed, and integrated into existing enterprise workflows.

AI is already changing how work gets done across enterprises, not just within technical teams. OpenAI points to internal and customer data indicating that 75% of enterprise workers report using AI to complete tasks they previously discussed but could not execute, increasing pressure on organizations to move from experimentation to real deployment.

According to OpenAI, what holds many teams back is not model capability but operational reality: enterprises already run on fragmented environments spanning multiple clouds, data platforms, applications, and governance models. As AI agents proliferate across organizations, that fragmentation becomes more visible and, in many cases, more costly.

Agents are often deployed in isolation, with limited access to data, tools, or organizational context, as organizations test their capabilities. As a result, each new agent can add complexity instead of reducing it, because it lacks the context needed to do its job well. OpenAI argues that this structural problem is now the primary blocker preventing enterprises from scaling agents beyond pilots.

OpenAI says it has seen this dynamic across more than one million businesses. In one example, AI agents reduced production optimization work at a major manufacturer from six weeks to one day. A global investment firm deployed agents across its sales process, freeing up more than 90% additional time for sales teams to spend with customers. At a large energy producer, agents helped increase output by up to 5%, which OpenAI says translated into more than $1 billion in additional revenue.

The challenge is compounded by the pace of AI development. As AI agents become more capable, the gap between what models can technically do and what enterprises can actually deploy has continued to widen. OpenAI says this gap is not driven by technology alone. Many teams are still developing the operational knowledge required to move agents from early pilots into real production work, even as AI capabilities improve at an accelerating pace.

OpenAI notes that new capabilities ship roughly every few days, making it difficult for enterprises to balance experimentation with control, especially in regulated or mission-critical environments. As a result, organizations face growing pressure to operationalize AI now, as the distance between early adopters and slower-moving enterprises continues to expand.

OpenAI says it learned that enterprises don’t simply need better tools for isolated tasks. Instead, teams need support moving AI agents from early experiments into everyday business use.

At the core of Frontier is a simple idea: AI agents need many of the same foundations as human employees, including shared context, onboarding, feedback-driven learning, and clearly defined permissions. That structure allows agents to move beyond isolated tasks and operate across an organization.

The platform was built by examining how enterprises already scale people, not software. That insight shaped Frontier’s design, with AI agents treated more like employees—onboarded, trained, evaluated, and governed within clear boundaries.

For AI coworkers to operate effectively at scale, OpenAI says several conditions need to be in place:

OpenAI says Frontier is designed to keep AI coworkers operating within clear and enforceable boundaries. Each AI coworker has its own identity, along with explicit permissions and guardrails that define what it can access and what actions it is allowed to take.

This structure allows enterprises to use AI agents confidently in sensitive or regulated environments, where security, compliance, and oversight are critical. OpenAI says enterprise-grade security and governance are built into Frontier by design, enabling organizations to scale AI use without losing control.

Rather than replacing existing infrastructure, Frontier is designed to work across the systems enterprises already use, including environments spread across multiple clouds. Teams can bring together existing data, AI models, and applications where they already live, without forcing replatforming.

Frontier relies on open standards to integrate existing applications, avoiding the need for new data formats or abandoning agents and applications that are already deployed.

Once deployed, AI coworkers can operate across different environments without requiring teams to change how work already gets done. Agents can run in local systems, within enterprise cloud infrastructure, or in OpenAI-hosted environments, depending on organizational needs.

For work that requires fast response times, Frontier is designed to prioritize low-latency access to OpenAI’s models, helping ensure interactions remain quick and consistent as agents are used in real operational workflows.

This approach allows AI coworkers to be accessible through multiple interfaces rather than confined to a single application. Agents can work alongside people wherever work happens—whether through ChatGPT, workflow tools like Atlas, or inside existing business applications—regardless of whether those agents are built in-house, provided by OpenAI, or integrated from other vendors.

OpenAI says Frontier is already being adopted by large enterprises across multiple industries, including manufacturing, financial services, transportation, insurance, and life sciences. Early adopters include HP, Intuit, Oracle, State Farm, Thermo Fisher Scientific, and Uber, with additional pilots underway at companies such as BBVA, Cisco, and T-Mobile.

Executives from several of these organizations emphasized the importance of pairing AI capability with governance, trust, and operational reliability.

At State Farm, Executive Vice President and Chief Digital Information Officer Joe Park said:

“Partnering with OpenAI helps us give thousands of State Farm agents and employees better tools to serve our customers. By pairing OpenAI’s Frontier platform and deployment expertise with our people, we’re accelerating our AI capabilities and finding new ways to help millions plan ahead, protect what matters most, and recover faster when the unexpected happens.”

Oracle Cloud Infrastructure Executive Vice President Greg Pavlik highlighted Frontier’s role in enabling AI use cases that operate across the business, saying:

“Our partnership with OpenAI continues to expand, helping enterprises unlock what’s possible with data and AI. With OpenAI Frontier, we have a strong platform to continue to introduce innovative AI use cases that work across the business.”

At Uber, Chief Technology Officer Praveen Neppalli Naga pointed to the importance of combining AI with human judgment at scale:

“At Uber, AI is already saving our engineers time and helping to power our products, but that is just the beginning. Our focus is on pairing the best of human judgment with the latest models to deliver real, measurable value across the company. We’re excited to work with OpenAI to find new ways to embed AI more deeply into our operations at true enterprise scale.”

Thermo Fisher Scientific Chairman, President, and CEO Marc Casper emphasized the platform’s potential impact on scientific research:

“Our collaboration with OpenAI is about driving science forward. By combining our deep life sciences expertise with OpenAI’s Frontier platform, we’re working to help our customers—scientists and researchers—advance their important work and deliver new medicines to patients faster.”

OpenAI emphasizes that closing the AI opportunity gap is not purely a technical challenge. Alongside Frontier, the company pairs enterprises with Forward Deployed Engineers (FDEs) who work directly with customer teams to develop best practices for deploying agents in production.

OpenAI says its forward-deployed engineers give enterprise teams a direct connection to OpenAI’s research organization. As companies deploy AI agents in real operational environments, OpenAI says it learns not only how to improve systems built around its models, but also how the models themselves need to evolve to better support enterprise work.

This creates a feedback loop in which real-world business use informs not only how systems are deployed and governed, while also shaping how agents and underlying models evolve as they move from pilots into production.

Because Frontier is built on open standards, OpenAI says third-party developers can build agents and applications that tap into the same shared enterprise context.

OpenAI argues this matters because many agent applications fail today due to a lack of context. Data is often scattered across systems, permissions are complex, and each new integration becomes a one-off project. Without access to shared business context, agents struggle to operate reliably inside real workflows.

Frontier is designed to make that context accessible with appropriate controls, allowing applications to work inside enterprise workflows from day one. For organizations, OpenAI says this can reduce integration overhead and speed deployment, avoiding the need to rebuild context and permissions each time a new agent or application is introduced.

The company is launching the platform with a small group of AI-native partners, including Abridge, Clay, Ambience, Decagon, Harvey, and Sierra. OpenAI says these partners are committing to deeper integration with Frontier, working closely with the company to understand customer needs, design enterprise-ready solutions, and support deployment. OpenAI plans to expand the partner ecosystem over time to include additional builders focused on enterprise AI.

OpenAI says Frontier is currently available to a limited group of enterprise customers, with broader availability expected over the coming months. The company says interested organizations can explore access through their existing OpenAI enterprise contacts.

Q: What problem is Frontier meant to solve?
A: Frontier addresses the gap between AI capability and enterprise deployment. While models have become more powerful, organizations struggle to operationalize them across real workflows, systems, and teams. Frontier focuses on execution, governance, and integration rather than model intelligence alone.

Q: How does Frontier give agents business context?
A: Frontier connects siloed systems—such as data warehouses, CRMs, ticketing tools, and internal applications—into a shared semantic layer. This allows AI agents to understand how information flows, where decisions occur, and what outcomes matter, similar to how human employees learn institutional knowledge.

Q: What does “AI coworkers” mean in practice?
A: AI coworkers are agents that can operate across tools and interfaces, not just within a single application. They can interact through ChatGPT, workflow systems like Atlas, or embedded business applications, while maintaining consistent context and permissions.

Q: How does Frontier handle evaluation and quality control?
A: Frontier includes built-in evaluation and optimization loops that allow both human managers and AI agents to understand what’s working and what isn’t. Over time, agents learn from real-world feedback, improving performance on tasks that matter most to the business.

Q: Is Frontier suitable for regulated environments?
A: OpenAI says enterprise security and governance are built into Frontier by default. Each agent has a defined identity, explicit permissions, and guardrails, enabling deployment in sensitive or regulated settings without sacrificing oversight.

As AI models continue to improve rapidly, the limiting factor for enterprises is no longer intelligence—it’s execution. OpenAI Frontier reflects a shift in enterprise AI strategy toward operational readiness, governance, and system-wide integration rather than isolated experiments.

Who should care:
Enterprise leaders responsible for AI strategy, IT and data executives managing complex systems, and business leaders under pressure to deliver measurable AI outcomes should pay close attention. Frontier directly addresses a common enterprise challenge: deploying AI agents safely and consistently across real workflows without increasing operational risk.

Why it matters now:
The gap between early AI adopters and slower-moving organizations is widening as AI capabilities advance faster than most enterprises can operationalize them. As new models and agent capabilities ship at an accelerating pace, organizations without shared context, evaluation processes, and governance structures risk falling behind—not because the technology is unavailable, but because they lack the execution frameworks needed to deploy AI reliably at scale. For enterprise leaders, the challenge is no longer whether AI is powerful enough, but whether their organizations are structured to keep up.

What decision this affects:
Frontier highlights a strategic choice facing enterprises: whether to continue treating AI as a set of disconnected pilots or to invest in an execution layer that allows agents to operate as dependable coworkers across the business. That decision affects how quickly organizations can move from experimentation to sustained, organization-wide AI impact.

How enterprises answer that question will influence not only how quickly they scale AI, but how reliably those systems can be trusted as they become part of everyday work.

Editor’s Note: This article was created by Alicia Shapiro, CMO of AiNews.com, with writing, image, and idea-generation support from ChatGPT, an AI assistant. However, the final perspective and editorial choices are solely Alicia Shapiro’s. Special thanks to ChatGPT for assistance with research and editorial support in crafting this article.