OpenAI Presence turns agents into a managed operations layer

OpenAI introduced Presence as a product for deploying agents into customer support, internal IT, sales and other operational workflows. The company says its own English language phone support at 1-888-GPT-0090 now resolves 75 % of inbound issues through Presence without human assistance.

Presence is built around permissions, not nicer chat

Presence combines model reasoning with company policies, guardrails, simulations, evals, approved actions and human escalation. The agent does not get broad access to everything. Each deployment starts with a specific job, such as billing, insurance claims or employee IT requests.

OpenAI also describes a post launch improvement loop. Production sessions and escalations reveal gaps, Codex uses the Presence plugin to propose changes, and teams can test those changes against the production version before approving a controlled rollout.

Enterprises are buying the machinery around the model

For large companies, this matters more than the model answering a question. The value is that an agent can use internal systems, follow policies and hand off to a person at the right moment.

OpenAI is moving its enterprise offer closer to a forward deployed model. Presence deployments are led by Forward Deployed Engineers and selected systems integrators. That is slower than a self serve API, but for risky workflows it is rational: someone has to sign off on permissions, tests and escalation rules.

Limited availability cuts against the scaling story

Presence is available only to eligible enterprise customers through a limited general availability program. OpenAI explicitly says it is not yet a self serve product.

That matters. A 75 % resolution rate inside OpenAI phone support is a strong signal. It does not mean a bank, insurer or airline gets the same result without expensive integration and hard measurement.

Live processes outside OpenAI will decide the story

The references to watch are deployments beyond OpenAI’s own support operation. BBVA is exploring voice support in Mexico, SoftBank is testing natural Japanese conversations and IAG is exploring support during demand spikes such as severe weather.

The real test arrives when customers publish their own escalation rates, error rates and costs after several months in production. Until then Presence is a credible operating frame, but still a frame.

Lilith's verdict

Presence seats OpenAI at the phone line where accounts, refunds and escalations are decided. If enterprises sign the access rules, the agent stops being a demo and starts taking a shift.