OpenAI Presence: AI now resolves 75% of calls without a human

On July 22, 2026, OpenAI unveiled Presence, a platform built for large organizations to deploy and manage voice and chat AI agents in production, complete with company-defined policies, guardrails and escalation to a human when needed. The headline isn’t the product itself, though — it’s the number attached to it: on OpenAI’s own English-language phone support line, the agent already resolves 75% of inbound calls without any human involved. For any business that currently pays a team to staff phone or chat support, that figure is a clear signal of where the market is heading, even though access to Presence itself is, for now, far from reach for most small and mid-sized companies.

What Presence actually is

Presence isn’t a standalone chatbot or an API you have to wire up yourself — it’s an end-to-end managed product. It bundles the policies and standard operating procedures that define what the agent can and can’t do, safety guardrails, a catalog of approved actions the agent can execute against a company’s systems, simulation environments to test the agent before it goes live, continuous evaluation tooling, and a Codex-powered improvement loop that refines behavior from real usage. Per OpenAI’s own announcement, every deployment starts with one specific job — resolving billing issues, handling an insurance claim, or fielding an internal IT service request — rather than a vague promise of “an agent for everything.”

Why the 75% matters more than the product name

The real story isn’t that OpenAI shipped another product — it’s that a voice AI, running in actual production rather than a demo, resolves 3 out of 4 calls without a person, and that a Codex-driven improvement cycle cut human handoffs by 15 percentage points in just 10 days, according to Help Net Security. It’s production-grade confirmation — not a lab result — of a trend we’ve already been tracking on this blog, from AI-made hotel and restaurant bookings to the rise of agentic commerce: voice and chat customer service has stopped being an experiment and become a production capability that measurably cuts operating cost.

The fine print every SME should read

Here’s the catch that matters for a mid-sized company: Presence launches in limited general availability, not as a self-service product. Per the same announcement, every deployment requires OpenAI’s own Forward Deployed Engineers or select global systems integrators — the consulting arm OpenAI built after acquiring the firm Tomoro and absorbing roughly 150 engineers into its new OpenAI Deployment Company, valued at $10 billion with backing from TPG, Goldman Sachs, SoftBank, Capgemini and McKinsey. In practice, that means Presence today is built for large accounts with budget and an integration team of their own — not for a 20- or 200-person company that wants to automate its support line next month.

What this means if you’re not a multinational

A player like OpenAI proving in production that a voice agent can resolve 75% of calls is good news for any sector that leans on phone or chat support — retail, hospitality, private healthcare, professional services. But it also exposes the gap: while the “turnkey” offering from the big labs rolls out to the biggest accounts first, an SME that wants that same capability today has two paths. Wait for access to open up — with no confirmed date — or build its own custom voice agent, scoped to the integrations its business actually needs, without depending on a limited-access program. That’s exactly the approach behind Elop, our own AI voice agent platform: it answers and makes real phone calls, executes actions against a business’s own systems, and leaves every conversation transcribed, analyzed and scored — no waiting list required.

How to get started without waiting for Presence

1. Pick one repetitive, measurable task — confirming appointments, resolving billing questions, qualifying an inbound call — instead of trying to “automate all customer service” at once.
2. Define what the agent can decide on its own and what it can’t, the same way Presence does with its policies: which actions it executes solo, and when it hands off to a person.
3. Measure the outcome from day one — calls resolved without intervention, average handling time, satisfaction — so you know whether the project pays off.
4. Bring in applied AI consulting and implementation for the part that’s usually harder than the conversational agent itself: integrating with your existing systems — CRM, bookings, billing.

Bottom line

OpenAI Presence confirms, with production numbers — 75% of calls resolved, a 15-point drop in human handoffs within 10 days — that voice and chat agents already work at real scale, not just in pilots. Access being limited today to large accounts with Forward Deployed Engineers doesn’t change the conclusion for an SME: the technology to automate a meaningful share of customer service already exists, and it can already be built to measure, without waiting your turn.

Want to find out if your business could automate part of its customer service with your own voice or chat agent? Get in touch: the first consultation is free and we reply within 24 hours.