On 22 July OpenAI introduced Presence, a product for putting AI agents on the phone lines and chat windows of large companies. The agents answer questions, use company systems, take actions the company has approved and hand the conversation to a person when needed. Two months on it is still in limited release, has no public price, and its most striking results come from OpenAI's own support line.

What OpenAI is selling

OpenAI calls Presence "a battle-tested product that helps enterprises deploy trusted AI agents" in its launch announcement. Each deployment starts with one specific job, such as resolving a billing issue, supporting an insurance claim or handling an employee's IT request. The agent receives only the knowledge and system access that job requires, and the company sets the policies: what the agent may do, when it needs approval and when a person should take over.

At launch the product covered real-time voice and chat; OpenAI had not confirmed email support, VentureBeat reported. An OpenAI spokesperson told the publication that "Presence uses OpenAI models for the core agent", while customers can connect third party models and services for guardrails, tools and other parts of their workflow.

Tested before launch, changed only with approval

Before an agent reaches customers, teams can test it against common requests, edge cases and higher risk scenarios, with graders checking whether it reached the right outcome, followed policy, used tools correctly and escalated when it should have. In live conversations, guardrails can step in when an interaction leaves the company's boundaries.

After launch, production sessions and escalations feed an improvement loop: Codex, OpenAI's coding agent, investigates them and proposes changes, and staff test each proposal against the version in production before approving a controlled rollout. VentureBeat called it "a formal mechanism for updating behavior without allowing an automated system to rewrite itself unchecked."

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, announcing Presence, 22 July 2026

Sold like a consulting project

Presence cannot be bought from a pricing page. It is offered to eligible enterprise customers through a limited general availability programme, deployments are led by OpenAI's Forward Deployed Engineers and selected global systems integrators, and it "is not yet available as a self-serve product", OpenAI says. The page still said so on 30 September. OpenAI has not disclosed pricing, contract terms or the cost of the engineering work that comes with a deployment; VentureBeat said it had asked about pricing twice and had no answer when it published.

The race is about implementation as much as models. In May OpenAI launched the OpenAI Deployment Company, a consulting firm backed by Bain & Company, and a week before Presence, Anthropic launched Ode, a consulting unit built around forward deployed engineers, VentureBeat noted. Ready made support agent platforms such as Sierra, Salesforce's Agentforce and Decagon were already on the market, Digital Applied wrote. On 24 September Google made Gemini 3.8 Live generally available on its enterprise agent platform, citing "improvements in voice quality, model reliability, and agent orchestration" in its release notes.

The numbers, and who stands behind them

OpenAI's main evidence is its own English language phone support line, which runs on Presence. The company says the agent verifies callers, uses account context and takes approved actions, that within weeks it met or exceeded the benchmarks OpenAI uses to grade human frontline support, and that it now resolves 75 percent of inbound issues without human help. OpenAI also says the improvement loop cut handoffs to people by 15 percentage points in 10 days.

75%
Share of inbound issues that OpenAI says its own support line resolves without a human. Company reported, not independently verified (22 July 2026).

These are vendor claims. "Those figures are company-reported and have not been independently verified," VentureBeat wrote. Digital Applied, an agency that builds agent systems for clients, called the 75 percent "a first-party ceiling, not a customer baseline", because OpenAI runs the line on its own products with full access to its own systems. The developer studio Codercops advised reading it as "OpenAI's own number for OpenAI's own use case".

The named customers are earlier in the process. According to OpenAI, BBVA is exploring AI voice support for everyday banking in Mexico, SoftBank is testing natural Japanese language customer conversations, and the Australian insurer IAG is exploring support during high demand events such as severe weather. "At BBVA, we are working closely with OpenAI to explore how trusted customer agents can help shape the future of financial services," said Daniel Ordaz, head of AI transformation at BBVA Mexico. The announcement does not describe any of the three as running Presence in full production.

What changes for support teams

The design moves work to the people who run support. Someone has to decide in advance which actions an agent may take alone, such as a refund, which need approval, and when a caller goes to a person: in Digital Applied's words, "human escalation as a designed path rather than a failure mode". Giving an agent only the access one job needs is the least privilege rule security teams already apply to staff, and the timing underlined it: Presence launched a day after OpenAI disclosed that AI agents running on its models had broken out of their test environment during an internal test, as we reported.

In the European Union there is also a legal duty. Since 2 August 2026, Article 50 of the EU AI Act requires AI systems that talk directly with people to be designed so that people know they are dealing with an AI, unless that is obvious, and the notice must come "at the latest at the time of the first interaction". For a phone agent, that points to the start of the call. Breaches can cost up to EUR 15 million or 3 percent of worldwide annual turnover, whichever is higher, under Article 99. The Commission's guidelines of 20 July explain what counts as a directly interactive system; our explainer covers the rest.

What is still unknown

The open questions are the ones buyers ask first: price, service levels, compliance certifications, and how the 75 percent figure holds up on a customer's systems rather than OpenAI's own. Until then, VentureBeat concluded, customers "still lack much of the information needed to assess total cost and operational risk." OpenAI says it will keep offering its models to voice customers through its API alongside Presence.