OpenAI launches Presence for agents
A new platform pairs voice and chat agents with company policies, permissions, and escalation rules; OpenAI says it runs most of its own support calls.
OpenAI introduced Presence on July 22, an enterprise product for deploying agents that answer questions, use company systems, take approved actions, and escalate to a human when they hit a wall. Rather than shipping a new model, Presence packages a “shared foundation” of company context, policies, permissions, guardrails, and evaluations so an agent behaves consistently whether a customer reaches it by voice or by chat. OpenAI is targeting the product at customer support, sales development, procurement, IT, and HR, calling those out as its highest-value enterprise use cases.
The company is using Presence on itself as the proof point: its own English-language phone support line now runs on the platform and, OpenAI says, resolves 75% of inbound calls without a human stepping in. Presence is available now to eligible enterprise customers through what OpenAI describes as a limited general-availability program, rather than a broad public launch.
The pitch is aimed at a specific failure mode operators keep running into: agents that sound competent in a demo but have no consistent policy for when to escalate, what they’re allowed to touch, or how to prove what they did afterward. That’s the same gap covered in putting an AI agent on your phone lines and in choosing a desktop AI agent: the model is rarely the bottleneck, the surrounding guardrail and audit layer is. If you’ve already built that layer yourself on top of the Assistants or Realtime APIs, Presence is worth comparing against your own scaffolding on cost and auditability before you switch, not a decision to make on the strength of an 75%-resolution number.