OpenAI launches Presence to run enterprise voice and chat agents
OpenAI's new platform packages enterprise voice and chat agents with guardrails and Codex-driven tuning — it already resolves 75% of OpenAI's own support calls.
OpenAI on July 22 introduced Presence, an enterprise platform for deploying and operating AI agents across voice and chat channels, marking the company's most direct move yet from selling model access to selling finished business applications.
Scoped agents with an improvement loop
Rather than offering a general-purpose chatbot, Presence starts each deployment from a narrowly defined job — handling billing disputes, processing insurance claims, or fielding internal IT and HR requests — and grants the agent only the data and system access that task requires. The platform wraps OpenAI's models in company-specific policies, guardrails, pre-deployment testing, and escalation paths to human staff.
A notable design choice is the built-in improvement cycle: OpenAI's Codex coding system analyzes performance signals from live conversations and proposes configuration updates, which human teams review, test, and approve before they reach production. OpenAI says Presence already runs its own English-language phone support line, resolving about 75% of inbound calls without human intervention.
Early customers span three continents
Launch partners include BBVA Mexico, which is using Presence for customer interactions; SoftBank Corp., which has deployed Japanese-language voice agents; and Retail Insurance Australia, part of insurer IAG. For now, Presence is limited to eligible enterprise customers rather than self-serve signup.
Why it matters
Presence signals that frontier labs intend to capture value further up the stack, competing directly with contact-center vendors, Salesforce's Agentforce, and startups like Sierra rather than merely supplying them with models. For OpenAI — whose infrastructure commitments reportedly now total hundreds of billions of dollars — packaged enterprise agents offer higher-margin revenue than raw token sales. The Codex-powered self-tuning loop is also a preview of how agent products will be maintained: not by prompt engineers editing configurations by hand, but by AI systems proposing their own updates under human sign-off. How well that loop holds up in regulated industries like banking and insurance will be an early test of whether autonomous customer-facing agents can be trusted at scale.