OpenAI on July 22, 2026 unveiled Presence, an enterprise product for putting production-grade AI agents to work across customer service and internal workflows over voice and chat. The launch marks the ChatGPT maker's clearest shift yet from selling raw model access to selling a fully-managed system for building, governing and continuously improving deployed agents.
A managed agent stack, not just an API
Presence bundles the pieces enterprises typically stitch together themselves: standard operating procedures and policies, tool actions, permission controls, escalation rules, evaluation and simulation, plus a Codex-powered improvement loop that proposes updates as production data comes in. Each deployment starts with a specific job — resolving a billing issue, handling an insurance claim, servicing an IT ticket — and the agent is granted only the knowledge and system access needed to do it.
"The challenge for enterprises is no longer proving that AI agents can work; it's making them reliable enough to do high-value work in production," OpenAI wrote in its launch note. Guardrails intervene when a conversation moves outside company boundaries, and every proposed change can be tested against the version already in production before rollout.
Powering OpenAI's own support line — and BBVA, SoftBank, IAG
OpenAI says Presence already powers its English-language phone support at 1-888-GPT-0090, resolving 75% of inbound issues without a human agent within weeks of launch and cutting human handoffs by 15 percentage points in 10 days after Codex-driven tuning. Design partners include Spanish banking group BBVA, exploring AI-powered voice support in Mexico; Japanese carrier SoftBank, testing natural Japanese-language customer conversations; and Australian insurer IAG, using it for surge support during severe weather events.
Limited GA, delivered by Forward Deployed Engineers
Presence is available today through a limited general availability program. Deployments are led by OpenAI's Forward Deployed Engineers and a small set of global systems integrators — there is no self-serve tier yet. OpenAI framed the product as the culmination of years of internal deployment work and said generalized insights from every customer engagement will flow back into research and product development. The launch dovetails with OpenAI's other July disclosures around its Project Camellia data center campus in Georgia and its widening enterprise footprint.
Reporting based on coverage from OpenAI, VentureBeat and Help Net Security.