OpenAI launches Presence to run enterprise voice and chat agents

Getting an AI to answer a prompt in a sandbox is trivial. Trusting an AI to answer a corporate phone line, verify identity, and issue refunds without going completely rogue is an entirely different nightmare.

OpenAI just launched OpenAI Presence, a dedicated enterprise platform designed to run production-ready AI agents across voice and chat.

Guardrails first, reasoning second

Instead of just handing companies an API key, Presence is built as an end-to-end framework. It combines model reasoning with strict enterprise guardrails, custom escalation triggers, and fine-grained permissions.

Each agent is scoped to a specific task, such as resolving a billing dispute or diagnosing an IT ticket. The agent only gets access to the internal databases required for that exact workflow.

Before going live, deployments face simulated edge cases to check whether the agent stays within company policy. Once in production, Codex analyzes failed calls and escalations, automatically proposing policy adjustments that human supervisors can test and approve.

Tested on its own support line

OpenAI battle-tested the platform by replacing frontline staff on its own English phone support channel at 1-888-GPT-0090. The system verifies callers, looks up backend accounts, and takes approved account actions on the spot.

According to OpenAI, the voice agent now resolves 75% of inbound calls with zero human intervention. Its Codex-powered feedback loop helped slash human handoffs by 15 percentage points in just 10 days.

External enterprise partners are already onboard. BBVA is testing it for banking voice support in Mexico, SoftBank is running trials for conversational Japanese, and insurer IAG is using it to absorb spike volumes during extreme weather.

There is no self-serve sign-up button yet. Presence is currently in limited general availability, deployed exclusively through OpenAI Forward Deployed Engineers (FDEs) and certified systems integrators.

Why it matters

OpenAI is taking a page straight out of the Palantir playbook. By pairing models with in-house Forward Deployed Engineers, they are admitting that raw frontier intelligence cannot survive inside corporate legacy stacks on its own. Enterprise value lives in the messy plumbing of policies, permissions, and audit logs.

For enterprise operators, hitting a 75% autonomous resolution rate on live voice calls is the holy grail of contact center economics. If these numbers hold up in complex banking environments, voice AI transitions from a pilot curiosity into a mandatory balance-sheet move.

However, this move places OpenAI on a direct collision course with its own developer ecosystem. Workflow platforms, contact center providers, and software vendors that built their roadmaps around wrapping OpenAI APIs now find their supplier selling a complete, turnkey replacement directly to the Fortune 500.