OpenAI Presence: governed agents for voice and chat work

OpenAI Presence is a deployed enterprise product for putting governed AI agents into customer and internal workflows. Announced on July 22, 2026, Presence combines OpenAI models with company knowledge, policies, permissions, approved actions, simulations, evaluations, guardrails, and human escalation so an agent can operate inside a defined business process rather than merely answer questions.

Definition: OpenAI Presence is an enterprise operating layer for AI agents that need to use company systems, follow policy, take approved actions, and escalate safely.

Example: A billing-support agent can understand a request, verify the customer, look up account context, apply company policy, process an approved action, and hand the interaction to a person when required.

Key takeaway: Presence packages the governance and continuous-improvement work that enterprises otherwise have to assemble around a model, an agent framework, and their own internal systems.

Business impact: Enterprises can evaluate a narrow workflow and move it toward production with OpenAI-led deployment support, but they should expect a high-touch rollout rather than a self-serve software purchase.

What problem is OpenAI Presence designed to solve?

OpenAI Presence is designed for the gap between an AI-agent demonstration and a reliable production workflow. OpenAI describes Presence as a way to deploy agents that answer questions, use company systems, take approved actions, and escalate to people while operating under company-defined policies and guardrails. For operators, the important shift is from asking whether a model can complete one task to defining what an agent may do repeatedly, under which conditions, and with what evidence of performance.

Presence starts with a specific job rather than a blank, general-purpose assistant. OpenAI gives billing issues, insurance claims, and employee IT requests as examples. The agent receives only the knowledge and system access needed for that job, while the customer decides which actions are allowed independently, which require approval, and when a person must take over.

How does Presence govern an agent before launch?

OpenAI Presence gives enterprise teams a pre-launch testing layer for common requests, edge cases, and higher-risk scenarios. The company says simulations and graders check whether an agent reached the intended outcome, followed policy, used tools correctly, and escalated when appropriate; guardrails can intervene when an interaction leaves the organization’s defined boundaries.

This matters because a production agent has more failure modes than a chatbot with no permissions. A useful evaluation has to test not only the answer, but also data access, tool use, approval boundaries, policy compliance, and the decision to involve a human. Presence makes those checks part of the deployment process OpenAI is selling.

What channels and workflows does Presence support today?

OpenAI Presence currently supports real-time voice and chat experiences, including customer support, outbound sales, and high-risk internal workflows. OpenAI’s announcement describes a customer-support example in which an agent understands the request, verifies the customer, retrieves account information, applies company policy, and takes an approved action. More on this: Circles lifts telco ARPU 22% with AI personalization.

The same foundation is intended to support workflows such as billing support, insurance claims, and employee IT service requests. OpenAI does not present Presence as a universal autonomous employee; the product is scoped around a defined job, its required systems, and explicit permissions.

How does Codex fit into the Presence improvement loop?

Codex is the mechanism OpenAI describes for turning production signals into proposed agent updates. Presence can surface sessions, escalations, and quality signals that indicate where an agent is performing well or needs attention; Codex, using the Presence plugin, investigates those signals and suggests changes.

The proposed change is not supposed to ship automatically. Teams can test it against the version already in production and then approve a controlled rollout. That approval loop is a meaningful distinction from an agent that changes its own instructions or tool behavior without an auditable review step.

What results does OpenAI report from its own support channel?

OpenAI says Presence powers its English-language phone-support channel at 1-888-GPT-0090, where the system handles open-ended requests, verifies callers, uses account context, and takes approved actions. OpenAI reports that the channel resolves 75% of inbound issues without human assistance and that its Codex-powered improvement loop reduced human handoffs by 15 percentage points in 10 days.

Those figures are company-reported, not independent benchmarks. They are useful as an indication of the workflow OpenAI is targeting, but buyers should ask for the measurement definitions, evaluation set, escalation policy, and service-level commitments before comparing the figures with their own support operation.

Why is Presence a high-touch enterprise product?

OpenAI Presence is available to eligible enterprise customers through a limited general availability program, with deployments led by OpenAI Forward Deployed Engineers and select global systems integrators. OpenAI says the product is not yet available as a self-serve offering. Background: Hugging Face CEO calls for ‘radical transparency’ after ‘unprecedented’ OpenAI hack. More on this: OpenAI Takes ChatGPT Ads Global, Expanding Across Europe and Asia. More on this: OpenAI's First Hardware Device Is Reportedly a Screenless AI Companion.

That delivery model reflects the integration work required to connect company knowledge and systems, define permissions, establish policies, test the agent, and put an approval process around changes. It also means the buying decision is not only about model usage: customers need to assess implementation effort, ownership, security review, operating responsibilities, and the cost of ongoing support.

What remains unclear about OpenAI Presence?

OpenAI has not yet published broad pricing, detailed interoperability requirements, contractual service levels, or a self-serve setup path. The initial availability is limited and scoped to individual customer use cases, so an enterprise cannot yet estimate total cost from a public product plan alone.

That uncertainty is important because Presence is positioned as a production operating layer, not just an API endpoint. Buyers should evaluate how the product integrates with existing identity, CRM, ticketing, contact-center, policy, logging, and evaluation systems—and clarify which controls remain with the customer, OpenAI, or a systems-integrator partner.

What OpenAI Presence signals for enterprise AI

OpenAI Presence shows OpenAI moving beyond model access toward managed deployment of agents inside business operations. The product’s core promise is not simply that an agent can reason; it is that the agent can act within a constrained workflow, prove that it followed the rules, and improve through a reviewed change process.

For enterprises, the practical question is whether a high-touch OpenAI deployment can reduce the time and risk of assembling those pieces internally. Presence is now available for selected enterprise deployments, but public pricing and broader technical details will determine whether it becomes a broadly adopted platform or remains a closely managed offering for a smaller set of customers.