OpenAI Presence brings managed voice and chat agents to enterprises

OpenAI is moving another step past model access with Presence, a limited general availability product for enterprises that want governed AI agents in production voice and chat workflows. The company says Presence can answer questions, resolve issues, use approved company systems, and escalate to people when the workflow requires human control.

This is not a self-serve API launch. OpenAI says eligible enterprise customers need to work through account teams, Forward Deployed Engineers, and selected systems integrators. That detail matters: Presence is a managed deployment product, not a dashboard every developer can switch on today.

OpenAI Presence packages the agent system, not just the model

The pitch is that companies do not only need a stronger model. They need permission design, test coverage, workflow ownership, guardrails, evaluation, production monitoring, and a route for improving the agent after it starts handling real users.

OpenAI describes Presence as a product for specific jobs such as billing resolution, insurance-claim support, employee IT service requests, customer support, and outbound sales. Each deployment is supposed to receive only the knowledge and system access needed for that job. The customer sets the policy boundaries: what the agent can do, when approval is required, and when a human takes over.

Business Insider frames the launch as part of OpenAI's shift into corporate software, where customers buy complete systems rather than raw model access. That is the practical change for buyers: Presence competes less with an API call and more with enterprise support platforms, contact-center automation, service-desk tooling, and systems-integrator projects.

Voice and chat are the first supported channels

OpenAI says Presence is available today for real-time voice and chat agents. The examples are concrete: a support agent can understand a billing request, verify a customer, look up account information, apply company policy, and take an approved action.

The product also includes simulations, graders, guardrails, standard operating procedures, approved actions, and escalation rules. Those pieces are the difference between a demo agent and an operational deployment. They give teams a way to test whether the agent reached the right outcome, followed policy, used tools correctly, and escalated when it should have.

Presence also links back into Codex. OpenAI says Codex can investigate production sessions and escalations, then propose updates that teams test and approve before controlled rollout. In other words, Codex is being positioned as part of the agent-maintenance loop, not only a coding assistant.

Early customers show the intended market

OpenAI named BBVA, SoftBank, and IAG as design partners or early enterprise users exploring Presence-style deployments. BBVA is looking at voice support for everyday banking needs in Mexico. SoftBank is testing natural Japanese-language customer conversations. IAG is exploring support during high-demand moments such as severe weather and insurance claims.

OpenAI also says Presence powers its own English-language phone support line, where it resolves 75% of inbound issues without human assistance and reduced human handoffs by 15 percentage points in 10 days through its Codex-powered improvement loop. Those are company-reported figures, so buyers should ask how the same metrics are defined in their own environment before comparing them with internal support benchmarks.

Availability is limited, and that limits the near-term impact

Presence is available through a limited general availability program for eligible enterprise customers. It is not listed as self-serve, and OpenAI says deployments are led by Forward Deployed Engineers and selected global systems integrators.

That access model makes Presence more relevant to large organizations with high-volume support, regulated workflows, or complex internal systems than to small teams looking for a quick agent builder. It also suggests implementation cost, governance work, and integration design will be part of the buying decision.

For teams already comparing AI agent platforms, the verification checklist is straightforward: ask which systems the agent can access, how approvals are enforced, what logs are retained, which failure cases trigger human escalation, how simulations are built, and who owns policy updates after launch.

Sources and methodology

This article is based on OpenAI's July 22, 2026 Presence announcement and Business Insider's independent coverage of the launch. LinkLoot treated the launch as a major enterprise-agent product update because it changes OpenAI's packaging from model/API access toward managed production agent deployments. The main caveat is access: Presence is limited general availability for eligible enterprise customers, not a public self-serve product.