OpenAI introduced Presence on July 22, an enterprise platform for deploying AI agents that handle customer and internal support requests over voice and chat.
What Presence Adds
Unlike a basic chatbot integration, Presence gives companies control over what an agent can access, which internal systems it can act on, and when it must hand a conversation to a human. Teams write policies and standard operating procedures the agent must follow, and set guardrails around which actions it is authorized to take before anything goes live.
Before deployment, companies can run simulations against common requests, edge cases, and high-risk scenarios, and use built-in evaluation tools to check how an AI agent handles policy compliance and tool use. Once live, a Codex-powered review process analyzes production conversations and proposes behavioral fixes, which staff must approve before they take effect.
75% of Calls, No Human Needed
OpenAI said Presence already runs its own English-language phone support line, where it resolves about 75% of inbound calls without an agent stepping in. Over a 10-day trial, the Codex-driven improvement loop cut handoffs to human staff by 15%, the company said.
Sushil Kumar, CEO of enterprise testing firm Cyara, said governance for deployed agents “has to move at machine-speed, with automated validation, guardrails, and real-time testing” — the kind of oversight Presence is built to provide.
A Limited, White-Glove Rollout
Presence isn’t self-service. OpenAI is rolling it out through a “limited general availability” program, pairing customers with its own forward-deployed engineers and a small number of systems-integrator partners to handle onboarding.
The launch adds to OpenAI’s growing set of tools for businesses, following products like ChatGPT Work and Codex, and lands as rival labs race to sell governed agents into enterprise workflows.