OpenAI Presence: Enterprise Voice & Chat Agents Resolve 75% of Contacts

What happened

OpenAI has launched Presence, an enterprise platform that deploys voice and chat agents capable of handling customer and employee support requests end-to-end. The agents can answer queries, operate IT systems, and escalate or transfer cases to human staff based on rules set by the deploying organisation — giving enterprises granular control over where automation ends and human judgement begins.

OpenAI is already running Presence on its own English-language phone support line, where the system verifies callers, retrieves account data, and executes approved actions autonomously. The company reports that 75 per cent of inbound contacts are resolved without any human involvement. A companion service, Codex, can monitor live agents and recommend process improvements; in OpenAI's internal testing, Codex-driven suggestions cut human handoffs by 15 percentage points within ten days.

Presence also ships with simulation and evaluation tooling, allowing enterprises to stress-test agent behaviour before going live — a notable concession to the operational risk concerns that have slowed enterprise AI adoption in regulated industries.

Why it matters

Presence is a direct challenge to the architecture of frontline service operations. For CX leaders, the meaningful shift is not simply that AI can answer questions — it is that a commercially available platform now handles resolution, not just deflection. The distinction matters behaviourally: customers who reach a channel expecting a human and receive a capable, autonomous agent will form expectations that carry across every other brand they interact with, compressing the industry's tolerance for slow or inconsistent service.

The configurable escalation model — where operators define which actions require human sign-off — is also a service-design signal worth watching. It mirrors the principle of appropriate autonomy from behavioral economics: systems that give customers a credible path to a human when stakes feel high tend to generate more trust than those that appear to trap users in automation. How enterprises calibrate that boundary will likely determine whether Presence raises or erodes customer satisfaction scores over time.

By the numbers

• 75% of inbound support contacts resolved without human assistance on OpenAI's own Presence deployment.
• 15 percentage points reduction in human handoffs achieved within ten days using Codex-generated process recommendations.

The Renascence take

Most commentary on Presence will fixate on job displacement. That is the wrong frame for operators who actually want to win on experience. The more consequential design question is where, precisely, the human handoff threshold is set — and whether that threshold is calibrated around customer anxiety or purely around cost reduction.

A 75% autonomous resolution rate is impressive as an efficiency metric, but it is meaningless — or actively damaging — if the 25% of cases that do reach a human arrive frustrated, unheard, and already primed to churn. The behavioral risk in platforms like Presence is not automation itself; it is the temptation to push the handoff threshold too far in pursuit of the headline number. Customer-obsessed operators should instrument the emotional state of customers at the moment of escalation, not just the volume of escalations. That single design decision — when the agent yields — will define whether Presence becomes a loyalty asset or a loyalty liability.