OpenAI introduced Presence on July 22 as an enterprise product for deploying voice and chat agents into customer service, sales and internal workflows.
The offering is narrower than the language of a general, always-present assistant might suggest. Each deployment begins with a defined job, such as resolving a billing problem, supporting an insurance claim or handling an employee IT request. The agent receives the knowledge and system access required for that job, while the customer sets policies for what it can do, when approval is required and when a person must take over.
OpenAI says Presence combines model reasoning with guardrails, evaluations and deployment support. Once an agent is live, production sessions and escalations reveal gaps. Codex can propose changes, but a team tests and approves them before the operating behavior is updated.
The product is the control system
Presence reflects a maturing enterprise AI market. The model is only one component. A production agent also needs identity, permissions, workflow connections, quality measurement, policy enforcement and a reliable human handoff. Buyers that focus on the model name risk underestimating the larger system they are commissioning.
The narrow-job approach is important. An agent that can do everything is difficult to govern; an agent responsible for one outcome can be evaluated against a defined policy and a known set of exceptions. That does not eliminate risk, but it makes ownership more concrete.
Companies considering the product should ask who controls the deployment after launch. Which changes require approval? How are failures sampled? Can the customer reproduce an action from the logs? What happens if an upstream policy or product changes before the agent’s instructions do?
They should also price the human system around the agent. Escalations, quality review and policy maintenance do not disappear; they become a new operating function whose workload needs to be measured from the beginning.
That workload belongs in the business case, not in a footnote after launch.
The name Presence points toward AI that remains embedded in an organization’s operations rather than waiting inside a blank chat box. The practical version of that future is less magical and more useful: systems that stay available for a specific job, learn from governed feedback and know when to stop.
If OpenAI can make that lifecycle repeatable, Presence may be more consequential than another standalone agent demo. It makes the improvement loop—the difficult work after launch—part of the product being sold.
Sources for editorial review
Drafting note: This draft was prepared with AI assistance from the linked source material and requires author review, independent fact-checking and final editorial approval before publication.