OpenAI Presence: Enterprise AI Agent Deployment Framework

OpenAI has introduced OpenAI Presence, a product designed to help enterprises deploy reliable AI agents capable of resolving issues, using company systems, and executing approved actions. The system addresses the challenge of moving AI agents from experimental proofs-of-concept to high-value production environments where reliability, policy adherence, and controlled adaptation are required.

Production-Ready Agent Capabilities

OpenAI Presence enables the deployment of agents for specific high-value workflows, such as resolving billing issues, supporting insurance claims, or managing employee IT service requests. These agents are deployed across both voice and chat channels, supporting use cases including customer support, outbound sales, and high-risk internal workflows.

To ensure reliability, Presence utilizes a combination of the following components:

• Policies and SOPs: Defined rules governing what the agent can do and when it requires human approval.
• Guardrails: Systems that intervene when interactions move outside established company boundaries.
• Approved Actions: Specific, authorized system interactions the agent can perform.
• Escalation Rules: Clear triggers for when an agent must hand off a session to a human operator.
• Knowledge and System Access: Agents are restricted to only the knowledge and system access required for their specific job.

The Codex-Powered Improvement Loop

A central feature of OpenAI Presence is its ability to adapt to changing user behavior, products, and policies without sacrificing control. The system uses a Codex-powered improvement process to maintain performance after launch:

1. Signal Detection: Production sessions, escalations, and quality signals are monitored to identify gaps in agent performance.
2. Proposed Updates: Codex investigates these signals and proposes specific updates to the agent's configuration.
3. Testing and Approval: Teams can test proposed changes against the current production version before approving a controlled rollout.

Performance Benchmarks and Enterprise Adoption

OpenAI uses Presence to power its own English-language phone support (1-888-GPT-0090). According to OpenAI, this deployment has achieved the following results:

• Resolution Rate: The agent resolves 75% of inbound issues without human assistance.
• Quality: Within weeks, the agent met or exceeded benchmarks used for frontline human-support quality.
• Efficiency: The Codex-powered improvement loop reduced human handoffs by 15 percentage points within 10 days.

Other enterprises currently utilizing the foundation of Presence include BBVA (exploring voice support for banking in Mexico), SoftBank (testing Japanese-language customer conversations), and IAG (exploring support during high-demand weather events).

Deployment and Availability

OpenAI Presence is not a self-serve product. It is available to eligible enterprise customers through a limited general availability program. Deployments are managed by OpenAI Forward Deployed Engineers (FDEs) and select global systems integrators who work with customers to identify workflows, connect systems, and establish permissions.

OpenAI continues to support voice customers who prefer to access frontier models directly via the OpenAI API.