CtrlK
OpenAI Frontier is an enterprise platform for building, deploying, and managing AI agents at scale. Launched on February 5, 2026, the platform enables organizations to treat AI agents as "AI coworkers" that can be onboarded, assigned identities, granted scoped permissions, and continuously evaluated for performance. Frontier connects to an organization's existing systems (CRM platforms, data warehouses, ticketing tools, and internal applications) to give every agent shared business context, and it supports agents built by OpenAI, by the enterprise itself, or by third-party providers including Google, Microsoft, and Anthropic.
At launch, early adopters included Uber, State Farm, Intuit, Thermo Fisher Scientific, HP, and Oracle, with dozens of additional organizations such as BBVA, Cisco, and T-Mobile participating in pilot programs. On February 23, 2026, OpenAI announced the Frontier Alliances program: multiyear partnerships with Accenture, Boston Consulting Group (BCG), Capgemini, and McKinsey & Company to accelerate enterprise deployment.
OpenAI's annual recurring revenue (ARR) topped $20 billion in 2025, tripling from $6 billion the previous year. The company's API business grew faster than its consumer ChatGPT product, and enterprise customers accounted for roughly 40% of total revenue. CFO Sarah Friar stated in a January 2026 blog post that the company would make 2026 its year of "practical adoption," writing that "the priority is closing the gap between what AI now makes possible and how people, companies, and countries are using it day to day."
Despite rapid growth in enterprise AI spending, many organizations struggled to move beyond isolated pilot projects. According to a Menlo Ventures survey from December 2025, while enterprise spending on large language models had grown substantially, most deployments remained limited to single-use cases. Frontier was designed to address this "opportunity gap" by providing a unified orchestration layer that could coordinate multiple agents across an organization's entire technology stack.
Sam Altman, OpenAI's CEO, framed the launch on X (formerly Twitter): "The companies that succeed in the future are going to make very heavy use of AI. People will manage teams of agents to do very complex things. Today we are launching Frontier, a new platform to enable these companies." He also told reporters that enterprise growth would be a "huge focus" for OpenAI going forward, and acknowledged that AI was making software easier to create, noting it would be "quite bad for some software companies" while emphasizing that many software providers have value propositions beyond their code.
OpenAI describes Frontier as a "semantic layer for the enterprise." Rather than replacing existing software systems, Frontier acts as an intelligence layer that stitches together disparate data sources and applications. The platform is built on four core pillars.
Frontier connects siloed data warehouses, CRM systems, ticketing tools, and internal applications so that every AI agent operating within the platform accesses the same institutional knowledge. This shared semantic layer builds what OpenAI calls "durable institutional memory" over time. As agents complete tasks and interact with enterprise data, they develop a richer understanding of the organization's processes, terminology, and workflows.
The business context layer is designed to solve a common problem in enterprise AI deployments: agents that operate in isolation, each with access to only a fragment of the information needed to complete complex tasks. By centralizing context, Frontier allows agents to collaborate across departments and systems.
The agent execution environment enables AI agents to apply model intelligence to real business situations. Agents can work with files, run code, browse the web, use external tools, and reason over structured and unstructured data. The execution environment supports parallel agent operations, meaning multiple agents can tackle different parts of a complex workflow simultaneously.
Frontier provides multiple runtime options: agents can run locally within an enterprise's own infrastructure, in cloud environments, or on OpenAI-hosted runtimes. This flexibility is designed to accommodate different security requirements and data residency regulations.
Built-in evaluation and optimization loops track agent performance across tasks. The system captures feedback on what works and what does not, then uses this data to refine agent behavior, adjust responses, and improve future performance. OpenAI compares this to the feedback cycles that human employees go through: just as a new hire improves with on-the-job experience and managerial feedback, Frontier agents are designed to learn from corrections and get better at their assigned roles over time.
The evaluation framework provides dashboards and detailed logs that give administrators visibility into agent actions, success rates, and failure modes.
Enterprise Identity and Access Management (IAM) applies across the full workforce of human employees and AI coworkers. Every Frontier agent is assigned a digital identity with specifically scoped permissions, analogous to a human employee's badge and clearance level. Agents can act on behalf of users without over-permissioning or added risk.
The governance layer provides comprehensive audit trails. CIOs can review agent actions with the same granularity as human user logs, supporting compliance with standards such as GDPR and internal policies. The platform meets leading security certifications including SOC 2 Type II, ISO/IEC 27001, ISO/IEC 27017, ISO/IEC 27018, ISO/IEC 27701, and CSA STAR.
| Feature | Description |
|---|---|
| Agent Builder | Build custom AI agents that connect to enterprise systems and execute multi-step workflows. Supports both no-code and code-based agent creation. |
| Multi-Model Support | Operate agents built on OpenAI models alongside agents from Google, Microsoft, Anthropic, and custom-built agents. Vendor-agnostic management from a single platform. |
| Agent Identity Management | Assign each agent a unique digital identity with scoped permissions through integration with existing enterprise IAM systems. |
| Shared Business Context | Unified semantic layer connecting data warehouses, CRM platforms, ticketing systems, and internal applications. |
| Agent Onboarding | Structured onboarding process for new agents, similar to onboarding a human employee with documentation, permissions, and feedback loops. |
| Evaluation Dashboards | Built-in monitoring and analytics showing agent performance, task completion rates, and failure modes. |
| Audit and Compliance Logging | Detailed action logs enabling regulatory compliance and internal governance review. |
| Forward Deployed Engineers | OpenAI engineers embedded directly within customer teams to help design architectures, operationalize governance, and establish repeatable patterns. |
| Parallel Agent Execution | Multiple agents can work simultaneously on different parts of a complex workflow. |
| Flexible Runtime Options | Agents can run on-premises, in enterprise cloud environments, or on OpenAI-hosted infrastructure. |
One of Frontier's most notable design decisions is its support for agents built on models from competing providers. The platform can orchestrate agents powered by OpenAI's own models (such as GPT-4 and the o-series reasoning models) alongside agents built on Google Gemini, Microsoft's AI models, Anthropic Claude, and models developed in-house by the enterprise.
This vendor-agnostic approach reflects a strategic bet by OpenAI that enterprises will increasingly operate in a multi-model environment, where different tasks may be best served by different models. Rather than forcing customers to commit exclusively to OpenAI's own models, Frontier positions itself as the coordination and governance layer that sits above any individual model provider.
From a practical standpoint, each third-party agent integrated into Frontier receives the same identity management, permission scoping, and audit trail capabilities as an OpenAI-native agent. This means that an organization could run a Claude-based agent for document analysis alongside a GPT-4-based agent for code generation, with both operating under the same security and governance framework.
However, some analysts have noted ambiguity about the depth of third-party integration. While Frontier supports third-party agents, the extent to which enterprises can freely mix and match models within complex agent workflows without friction remains an area that will become clearer as the platform matures.
Frontier is not an upgrade to or replacement for ChatGPT Enterprise. The two products serve fundamentally different purposes and are designed to be complementary.
| Dimension | ChatGPT Enterprise | OpenAI Frontier |
|---|---|---|
| Primary Purpose | Conversational AI assistant for individual employees and teams | Platform for building, deploying, and managing autonomous AI agents |
| User Interaction | Chat-based interface; users interact via natural language prompts | Agent-based; agents operate autonomously across enterprise systems |
| Data Integration | Connectors to Slack, Google Drive, SharePoint, GitHub, and other sources for context-aware chat | Deep integration with CRM systems, data warehouses, ticketing tools, and internal applications as a semantic layer |
| Autonomy Level | Responds to user requests in real time; limited autonomous action | Agents execute multi-step tasks independently, run code, use tools, and coordinate with other agents |
| Identity Model | User accounts tied to enterprise SSO | Agent IAM: each AI agent receives its own digital identity with scoped permissions |
| Governance | Admin console for usage management and data controls | Full audit trails, compliance logging, and agent-level governance aligned with SOC 2, ISO 27001, and GDPR |
| Multi-Model | OpenAI models only | Supports OpenAI, Google, Microsoft, Anthropic, and custom models |
| Support Model | Standard enterprise support | Forward Deployed Engineers embedded in customer teams |
| Target Use Case | Employee productivity, Q&A, document drafting, brainstorming | Cross-system workflow automation, agent orchestration, enterprise-scale AI operations |
OpenAI has emphasized that organizations can run both products simultaneously. ChatGPT Enterprise serves as a productivity tool for day-to-day employee tasks, while Frontier handles complex, cross-system automation that requires autonomous agent behavior.
Frontier launched with a select group of enterprise customers, with broader availability planned for later in 2026.
| Customer | Industry | Status |
|---|---|---|
| Uber | Transportation / Technology | Early adopter |
| State Farm | Insurance | Early adopter |
| Intuit | Financial Software | Early adopter |
| Thermo Fisher Scientific | Life Sciences | Early adopter |
| HP | Technology / Hardware | Early adopter |
| Oracle | Enterprise Software | Early adopter |
| BBVA | Banking / Financial Services | Pilot program |
| Cisco | Networking / Technology | Pilot program |
| T-Mobile | Telecommunications | Pilot program |
These early adopters span a range of industries, from financial services and insurance to life sciences and telecommunications. The diversity of the initial customer base reflects OpenAI's ambition to position Frontier as a horizontal platform applicable across sectors rather than a solution tailored to any single industry.
At launch, OpenAI's Chief Revenue Officer declined to discuss specific pricing details. The platform follows a custom enterprise sales model where pricing is determined by factors such as the number of agents deployed, data volume, API usage, deployment environment, and the level of Forward Deployed Engineer support required.
On February 23, 2026, eighteen days after the platform launch, OpenAI announced the Frontier Alliances program: multiyear partnerships with four of the world's largest consulting firms. The program was designed to help enterprises move beyond AI pilot projects and into production-scale agent deployments.
| Partner | Primary Role | Focus Area |
|---|---|---|
| McKinsey & Company | Strategy and transformation | AI strategy, operating model redesign, change management, executive adoption |
| Boston Consulting Group (BCG) | Strategy and transformation | Strategy development, operating model transformation, sustained impact measurement |
| Accenture | Implementation and integration | System integration, wiring Frontier into enterprise data and applications, security, reliability |
| Capgemini | Implementation and integration | Technical deployment, enterprise system integration, secure and reliable agent operation |
McKinsey and BCG handle the strategic work: helping C-suite executives determine where to start, how to restructure operating models for an agent-driven workforce, and how to drive organization-wide adoption. Accenture and Capgemini take on the technical integration, connecting Frontier to the systems and data that enterprises actually run on.
All four firms are building dedicated practice groups that will be certified on OpenAI technology. These teams receive roadmap insight, access to technical resources, and direct engagement with OpenAI's product and research teams. The consulting firms work alongside OpenAI's Forward Deployed Engineers, combining OpenAI's research expertise with the consulting firms' deep transformation experience and global delivery capabilities.
OpenAI declined to share the financial terms of the partnerships, though CNBC reported that the deals were multiyear in nature.
The Frontier Alliances program addresses a well-documented challenge in enterprise technology: the gap between platform capability and organizational readiness. Many enterprises have the technical infrastructure to deploy AI agents but lack the strategic frameworks, change management processes, and implementation expertise to do so at scale. By partnering with firms that already have deep relationships with Fortune 500 C-suites, OpenAI gains a distribution channel and implementation capability that would take years to build internally.
The Enterprise Frontier Program is OpenAI's hands-on engagement model for Frontier customers. It pairs OpenAI Forward Deployed Engineers (FDEs) with customer teams to work side by side on architecture design, governance operationalization, and production deployment.
FDEs are not traditional customer support staff. They are engineers with deep technical expertise who embed directly within customer organizations. Their responsibilities include designing agent architectures tailored to the customer's specific systems and workflows, helping operationalize governance frameworks, establishing repeatable patterns that the customer's own team can own and extend, and providing a direct connection to OpenAI Research so that customer feedback can influence product development.
The FDE model draws inspiration from similar programs at companies like Palantir, which pioneered the use of forward-deployed engineers as a go-to-market strategy for complex enterprise software.
On March 9, 2026, approximately one month after Frontier's launch, OpenAI announced its intention to acquire Promptfoo, a cybersecurity startup founded in 2024 by Ian Webster and Michael D'Angelo. Promptfoo developed open-source tools for testing security vulnerabilities in large language models and agentic workflows. The company reported that its products were used by more than 25% of Fortune 500 companies.
Once finalized, Promptfoo's technology will be integrated directly into the Frontier platform to provide automated red-teaming of agent workflows, security evaluation of agentic processes, runtime monitoring for risks and compliance needs, and ongoing vulnerability assessment. OpenAI stated that it would continue building Promptfoo's popular open-source project. The Promptfoo team will join OpenAI. Financial terms of the acquisition were not disclosed.
The acquisition underscores OpenAI's recognition that as AI agents take on more autonomous roles within enterprises, the security surface area expands significantly. Agents that can run code, access enterprise data, and interact with external systems present new categories of risk that traditional cybersecurity tools are not designed to address.
Frontier entered a competitive market. Several major technology companies had already launched or announced their own enterprise agent platforms by early 2026.
| Dimension | OpenAI Frontier | Anthropic Claude Cowork | Microsoft Copilot + Agent 365 | Salesforce Agentforce |
|---|---|---|---|---|
| Launch Date | February 5, 2026 | January 2026 (research preview) | Copilot Wave 3 / Agent 365 (March 2026) | October 2024 (Agentforce 1.0); expanded 2025 |
| Architecture | Centralized orchestration layer ("semantic layer") across all enterprise systems | Personal AI coworker operating through Claude Desktop for macOS | Embedded within Microsoft 365 ecosystem; deep Office, Teams, and Windows integration | Embedded within Salesforce CRM ecosystem |
| Multi-Model | Yes: supports OpenAI, Google, Microsoft, Anthropic, and custom agents | Anthropic Claude models only | Yes: OpenAI models and Claude (via Frontier program) integrated into Copilot | Primarily Salesforce Einstein models; OpenAI partnership for select capabilities |
| Agent Identity/IAM | Full agent IAM with scoped permissions and audit trails | User-level permissions through Claude account | Integrated with Microsoft Entra ID and existing Microsoft 365 admin controls | Salesforce-native permissions and sharing model |
| Data Integration | Connects to CRM, data warehouses, ticketing tools, and internal applications via semantic layer | Local file access on macOS; plugin-based integrations | Deep integration with Microsoft 365 data (SharePoint, Outlook, Teams, OneDrive) | Native access to Salesforce CRM data, pipelines, case history, and business rules |
| Availability | Limited enterprise customers; broader rollout planned for later 2026 | Research preview available to all paid Claude users ($20/month Pro plan) | Microsoft 365 E7 "Frontier Suite" launching May 1, 2026 at $99/user/month | Generally available; 29,000+ deals closed; $800M ARR |
| Pricing | Custom enterprise contracts (not publicly disclosed) | Available with Claude Pro ($20/month) or Team/Enterprise plans | $99/user/month (E7 Frontier Suite bundles E5 + Copilot + Agent 365) | Agentic Enterprise License Agreement (fixed-price model); custom enterprise pricing |
| Target User | Large enterprises deploying autonomous agents across multiple systems | Individual professionals and small teams seeking an AI coworker | Organizations already in the Microsoft 365 ecosystem | Organizations heavily invested in Salesforce CRM |
| Support Model | Forward Deployed Engineers + Frontier Alliances consulting partners | Standard Anthropic support | Microsoft enterprise support and partner ecosystem | Salesforce consulting partner ecosystem |
| Key Differentiator | Vendor-agnostic orchestration across all enterprise systems and agent providers | Deep computer use capabilities; local file access; individual empowerment | Tight integration with the world's most widely deployed productivity suite | Native CRM data access; no integration project required for sales/service workflows |
The four platforms represent distinct architectural philosophies for enterprise AI:
OpenAI Frontier takes the overlay approach: it positions itself as a neutral coordination layer that sits on top of all existing enterprise systems. The advantage is flexibility and vendor neutrality. The risk is that the overlay model requires significant integration work to connect to each enterprise's unique technology stack.
Anthropic Claude Cowork takes the personal empowerment approach: rather than orchestrating agents across an enterprise, it gives individual users a capable AI colleague that can access local files, use desktop applications, and complete tasks through a plugin ecosystem. It is designed for individual productivity rather than enterprise-scale agent orchestration.
Microsoft Copilot with Agent 365 takes the embedded ecosystem approach: agents are deeply woven into the Microsoft 365 applications that hundreds of millions of workers already use daily. The advantage is that there is no integration gap for organizations already in the Microsoft ecosystem. The limitation is that the value proposition weakens for workflows outside of Microsoft's application suite.
Salesforce Agentforce takes the system of record approach: agents operate directly within the CRM where customer data, pipeline information, and case history already reside. For sales and customer service workflows, this provides an immediate advantage because no data integration is required. The limitation is that Agentforce is strongest within Salesforce's own ecosystem.
Notably, these platforms are not entirely zero-sum competitors. Salesforce and OpenAI announced a partnership in October 2025 that brings OpenAI's models into Agentforce and integrates Agentforce capabilities into ChatGPT. Similarly, Microsoft's Copilot now includes Claude models alongside OpenAI models, suggesting that the enterprise AI market may evolve toward interoperability rather than winner-take-all dynamics.
As of March 2026, OpenAI has not publicly disclosed Frontier pricing. The platform operates on a custom enterprise sales model. At the press event for the February 5 launch, OpenAI's Chief Revenue Officer declined to discuss pricing specifics.
Based on reporting from multiple outlets, pricing is likely determined by a combination of factors.
| Pricing Factor | Description |
|---|---|
| Number of Agents | The total number of AI agents deployed within the organization |
| Data Volume | The amount of enterprise data connected through the semantic layer |
| API Usage | Volume of API calls and computational workload generated by agents |
| Deployment Environment | Whether agents run on OpenAI-hosted infrastructure, enterprise cloud, or on-premises |
| FDE Support Level | The extent of Forward Deployed Engineer engagement and duration |
| Consulting Partnership | Whether the deployment involves a Frontier Alliance partner |
This approach is consistent with enterprise software sales norms for platforms of this complexity. The lack of published pricing means that smaller organizations and startups may find it difficult to evaluate Frontier against alternatives with transparent pricing, such as Claude Cowork (available starting at $20/month) or Microsoft's E7 Frontier Suite ($99/user/month).
Frontier is built on the same security and compliance foundation that supports OpenAI's existing business products, which serve millions of users. The platform meets the following certification standards:
| Standard | Scope |
|---|---|
| SOC 2 Type II | Security, availability, processing integrity, confidentiality, and privacy controls |
| ISO/IEC 27001 | Information security management systems |
| ISO/IEC 27017 | Cloud security controls |
| ISO/IEC 27018 | Protection of personally identifiable information in public clouds |
| ISO/IEC 27701 | Privacy information management |
| CSA STAR | Cloud security assurance |
The agent identity system provides granular control: administrators define exactly what data each agent can access, what actions it can take, and what systems it can interact with. Agent actions are logged with the same granularity as human user actions, enabling compliance with GDPR, HIPAA (for healthcare use cases), and industry-specific regulations.
The March 2026 acquisition of Promptfoo further strengthens the security posture by adding automated red-teaming and vulnerability assessment capabilities designed specifically for agentic AI workflows.
Frontier's launch reflects a broader shift in the enterprise AI market from "which model is smartest" to "which platform handles my company's data, agents, and workflows best." This shift has significant implications for the traditional software-as-a-service (SaaS) industry.
Sam Altman's comment that AI would be "quite bad for some software companies" points to a future where AI agents could potentially automate workflows that today require dedicated SaaS applications. If an agent can navigate a CRM, update records, generate reports, and respond to customer inquiries autonomously, the value proposition of the underlying CRM software shifts from "tool employees use" to "data layer agents operate on."
Fortune reported that Frontier "could reshape enterprise software" by positioning AI agents as an intermediary layer between users and their business applications. Rather than employees interacting directly with Salesforce, Workday, or ServiceNow, agents would handle routine operations within those systems, with employees managing and supervising the agents.
This vision is not without skepticism. Critics have noted that enterprise system integration is notoriously difficult, and that connecting an AI platform to the dozens or hundreds of applications that a large organization operates is a challenge that many previous "integration platform" providers have attempted without transformative success. The Frontier Alliances with major consulting firms can be read as an acknowledgment that this integration work requires significant human expertise that cannot be automated away.
| Date | Event |
|---|---|
| January 2026 | OpenAI CFO Sarah Friar announces 2026 as the year of "practical adoption" for enterprise AI |
| February 5, 2026 | OpenAI Frontier launches with early adopters including Uber, State Farm, Intuit, Thermo Fisher Scientific, HP, and Oracle |
| February 23, 2026 | Frontier Alliances announced: multiyear partnerships with Accenture, BCG, Capgemini, and McKinsey |
| March 9, 2026 | OpenAI announces acquisition of Promptfoo for AI agent security integration into Frontier |
Altman has articulated a vision in which AI agents become a standard part of the enterprise workforce. In this vision, companies do not merely adopt AI tools; they restructure their operations around teams of human employees managing teams of AI agents. The "AI coworker" framing is deliberate: it implies that agents should have the same organizational infrastructure as human workers, including onboarding, permissions, performance reviews, and career development (in the form of evaluation and optimization loops).
Altman has drawn an explicit parallel between the current moment in AI and earlier platform shifts. Just as the internet transformed how companies operated in the late 1990s and cloud computing reshaped IT infrastructure in the 2010s, Altman sees AI agents as the next platform shift that will fundamentally change how work gets done.
The Frontier platform is designed to be the infrastructure for this shift. By providing the governance, security, and orchestration capabilities that enterprises require, Frontier aims to lower the barrier between experimental AI usage and production-scale deployment. Whether this vision is realized will depend on how effectively Frontier can deliver on its promise of seamless integration across the diverse, complex, and often messy technology environments that characterize large organizations.