OpenAI Launches Frontier: An Enterprise Platform for Managing AI Agent Workforces

OpenAI launched Frontier, an enterprise platform designed for building and managing AI agent workforces at scale. The platform introduces two capabilities that have been conspicuously absent from the agent ecosystem: Business Context — a system for connecting agents to an organisation's CRM, databases, and internal knowledge — and enterprise-grade identity and access management (IAM) for agent identities. First customers include Uber, Intuit, and State Farm.

This is not another agent framework. It is a management layer for organisations that want to deploy dozens or hundreds of specialised agents across their operations.

What problem does Frontier solve that existing tools don't?

Current agent frameworks — LangGraph, CrewAI, Google's ADK — focus on building individual agents or small teams of agents. They handle the mechanics of agent construction: tool use, memory, reasoning loops, orchestration. What they do not handle is the organisational layer: who is this agent, what is it allowed to access, how does it connect to our business data, and how do we manage a fleet of them?

Frontier addresses the gap between building an agent and operating an agent workforce. Consider an insurance company like State Farm deploying agents for claims processing, customer service, underwriting analysis, and fraud detection. Each agent needs different data access, different permissions, and different oversight levels. Managing this with individual agent frameworks creates an operational nightmare.

How does Business Context work?

Business Context is Frontier's answer to the knowledge grounding problem. Rather than each agent maintaining its own RAG pipeline or being manually loaded with context, Business Context provides a centralised layer where organisational data — CRM records, product catalogues, internal documentation, customer histories — is made available to agents based on their role and permissions.

The practical benefit is consistency. When five different agents need to know the current pricing structure, they all pull from the same source of truth rather than each maintaining a potentially stale copy. When a policy changes, it propagates through Business Context rather than requiring updates to every agent's individual knowledge base.

For practitioners, this is the piece that transforms agents from clever demos to production systems. The gap between a working prototype and a deployed agent is almost entirely about data access, freshness, and permission management. Business Context handles all three.

Why does agent IAM matter?

Giving AI agents identities within an enterprise IAM system is a deceptively significant move. When agents have identities — roles, permissions, access scopes, audit trails — they become manageable entities within existing security infrastructure rather than ungoverned processes running with whoever launched them's credentials.

This means agents can be subject to the same access controls as human employees. A claims processing agent gets access to claims data but not financial trading systems. A customer service agent can read customer records but cannot modify billing. Permissions are defined, enforced, and auditable.

The alternative — agents running with service accounts or shared credentials — is what most current deployments use, and it is a security liability that compliance teams rightly flag. Enterprise IAM for agents is not a feature; it is a prerequisite for deployment in regulated industries.

What do the first customers tell us?

Uber operates at a scale where agent workforces make immediate sense. Driver support, rider issue resolution, marketplace optimisation, fraud detection — each is a domain where specialised agents can handle routine cases while escalating complex ones to human operators. The volume justifies the infrastructure investment.

Intuit serves small businesses through TurboTax, QuickBooks, and Mailchimp. Agent workforces could handle tax guidance, bookkeeping assistance, and marketing campaign optimisation — high-value advisory services delivered at software scale.

State Farm represents the insurance vertical, where claims processing, policy management, and customer service are high-volume, rule-governed processes. Insurance is one of the industries where agent deployment economics are most compelling: the tasks are well-defined, the data is structured, and the cost of human processing is high.

The customer list tells us Frontier is targeting industries with high transaction volumes and well-defined workflows — exactly the domains where agent economics work best.

What should teams evaluating agent platforms consider?

Frontier creates a new layer in the agent stack between the model and the application. Teams need to evaluate whether that layer belongs to OpenAI or to their own infrastructure.

Lock-in risk is real. Business Context connects deeply to your organisational data. Migrating away from Frontier after deployment means rebuilding every data connection and permission mapping. The deeper the integration, the higher the switching cost.

Build vs. buy becomes urgent. Frontier bundles capabilities — agent management, data integration, IAM, monitoring — that teams would otherwise build separately. For organisations deploying fewer than ten agents, building these layers is manageable. For organisations deploying fifty or more, the build cost may exceed the platform cost.

Model coupling. Frontier is model-agnostic in principle but optimised for OpenAI models in practice. Teams that want to run Claude, Gemini, or open-source models as agent backends should verify that model flexibility is real, not theoretical, before committing.