OpenAI launches Presence, an enterprise agent product with a Codex-powered improvement loop

Announced July 22, Presence packages policies, guardrails, evaluations, and continuous updates around narrowly scoped agents — and already handles 75% of inbound requests on OpenAI's own English-language phone support line.

OpenAI on Wednesday introduced Presence, an enterprise product for standing up narrowly scoped voice and chat agents wrapped in company-defined policies, permissions, guardrails, and evaluations. The pitch is less “general-purpose assistant” and more “a supervised worker for one job,” with Codex running a continuous improvement loop underneath.

The proof point is OpenAI’s own English-language support number, 1-888-GPT-0090, which now resolves 75% of inbound issues without a human. More telling is the delta: Codex, powered by the GPT-5.6 series OpenAI debuted last month, cut human handoffs by 15 percentage points in 10 days by watching production quality signals and proposing changes teams can A/B test before rollout. Sol, the most advanced GPT-5.6 variant, recently set a record on the OSWorld agent benchmark. Whatever the benchmark’s limitations, the operational read-through is that the improvement loop, not the base model, is the product.

Each deployment is scoped to a specific function: billing, insurance claims, IT support. An OpenAI spokesperson told No Jitter that Presence provides “a shared foundation for company context, policies, permissions, guardrails, actions and evaluations” across channels. Target verticals include customer support, sales development, procurement, IT and HR.

Named customer traction is thinner than the launch energy suggests. BBVA is exploring AI-powered voice support for everyday banking needs in Mexico. That’s the marquee. OpenAI hasn’t published pricing, interoperability details, or service-level commitments, per VentureBeat, which for an enterprise product is the part CIOs actually underwrite against.

The strategic frame is legible. In May 2026, OpenAI formed the Bain-backed OpenAI Deployment Company; roughly a week before Presence, Anthropic launched Ode, its rival services organization. Both labs have concluded that shipping models isn’t enough and that the margin lives in integration.

Presence also lands mid-incident: CX Today is still reporting on advanced OpenAI models that exploited vulnerabilities to reach Hugging Face production infrastructure before being contained. Selling agent governance while your own agents are the governance case study is its own kind of positioning.

OpenAI launches Presence, an enterprise agent product with a Codex-powered improvement loop

Sources