OpenAI on Wednesday introduced Presence, a "battle-tested" enterprise product that packages its models with guardrails, policies and Codex-driven evaluations so large organizations can deploy trusted voice and chat agents in production. The company said the platform already runs its own English-language phone support at 1-888-GPT-0090 and resolves roughly 75% of inbound issues without a human handoff.
A shift from raw models to managed agents
Presence marks OpenAI's clearest move yet from selling raw model access to running a full agent lifecycle. Each deployment is scoped to a specific job — resolving a billing dispute, processing an insurance claim, servicing an internal IT ticket — with only the knowledge and system access required for that task. Companies define policies, approvals and escalation rules; OpenAI's launch team wires in the systems, tests the agent through simulations and graders, and moves it into production.
Guardrails, Codex updates and continuous evaluation
After launch, production sessions and escalations feed a Codex-powered improvement loop that proposes updates human teams can test and approve before rollout. OpenAI said this loop reduced human handoffs on its own support line by 15 percentage points in 10 days. Guardrails intervene when interactions drift outside the enterprise's boundaries, and the same policies and evaluations can be reused as customers expand from one workflow or channel to another.
BBVA, SoftBank and IAG as design partners
Three global enterprises are already building on Presence as design partners: BBVA is exploring AI-powered voice support for everyday banking needs in Mexico, SoftBank is testing natural Japanese-language customer conversations, and Insurance Australia Group (IAG) is targeting timely support during high-demand events such as severe weather. Presence is available today through a limited general-availability program run by OpenAI's Forward Deployed Engineers and select systems integrators; a self-serve offering has not yet launched, though voice customers retain access to frontier models via the OpenAI API.
The launch lands the same week OpenAI unveiled its Project Camellia data center campus in Georgia and continues its enterprise push after the recent ReliaQuest Daybreak cyber partnership and the Google Gemini 3.6 Flash rollout.
Reporting based on coverage from OpenAI, Bloomberg, VentureBeat and Help Net Security.