OpenAI Launches Presence for Enterprise Agents

OpenAI introduced Presence on July 22, a platform for companies that want to put AI agents on the phone and in chat windows without losing sleep over what those agents might do. It is aimed at customer service and internal operations, the two places where businesses most want automation and most fear it going wrong.

The design leans hard on containment. Each deployment starts with a single job, handling billing questions, insurance claims or employee IT requests, and the agent gets only the data and system access that job requires. Nothing more. On top of that, companies set the policies, define which actions need human approval, and decide when a case should be handed to a person. A Codex-powered loop is meant to improve the agent after launch based on how it performs.

Already answering the phones

OpenAI is running Presence on its own English-language phone support line, where it says the system resolves 75 percent of inbound calls without a human stepping in. That is a real number attached to a real workload, which makes it more convincing than a demo.

The rollout is deliberately narrow. Presence is launching through a limited general availability program rather than as a self-service product, with deployments led by OpenAI's own forward-deployed engineers and a handful of systems integrators. That is a signal in itself. OpenAI wants to be in the room when these agents go live, not watching from a distance.

The control theme again

Presence fits a pattern showing up across the industry this month, from Mistral's pitch to regulated buyers to OpenAI's own move to fold work tools into ChatGPT. The selling point has shifted from raw capability to control. It also lines up with a point worth repeating: an agent left alone is still a liability, and the human in the loop has not gone anywhere. Presence bakes that assumption into the product rather than pretending otherwise.