OpenAI Launches Presence, an Enterprise Product for Production AI Agents

OpenAI says its own English-language phone support channel, run by its new Presence product, now resolves 75% of inbound issues without human assistance — and that within weeks it met or exceeded the benchmarks the company uses to grade frontline human support quality. The company announced Presence today as a deployed enterprise product aimed squarely at a problem that has moved past proof of concept: making AI agents reliable enough to handle high-value work in production.

Presence supports real-time voice and chat experiences such as customer support, outbound sales, and high-risk internal workflows. Each deployment starts with a specific job — resolving billing issues, supporting insurance claims, or handling employee IT service requests — and the agent receives only the knowledge and system access that job requires. The customer sets the policies: what the agent can do, when it needs approval, and when a person should take over.

The launch matters because enterprise adoption of AI agents has hit a reliability ceiling. As OpenAI frames it in its announcement, "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." Agent behavior must also adapt as products, policies, and user behavior change. That, the company argues, requires more than a model — it requires "the systems, evaluations, and deployment expertise to improve agents as those conditions change without giving up control."

How a deployment works

OpenAI positions Presence not as a self-serve API product but as a jointly executed deployment. OpenAI works alongside each customer to identify a high-value workflow, connect the necessary knowledge and systems, establish permissions and policies, test the agent, and bring it into production. After launch, production sessions and escalations reveal gaps in the agent's behavior. Codex, OpenAI's coding agent, then proposes updates that internal teams can test and approve, helping the agent adapt as customer behavior shifts.

As a deployment expands, OpenAI and select systems integrators can continue supporting it. The product was developed in tight collaboration with OpenAI's Research team, and the company says generalized insights from every deployment feed back into ongoing research and product development, improving the product for all customers over time.

A typical interaction, as OpenAI describes it, might involve resolving a billing issue: understanding the request, verifying the customer, looking up account information, applying company policy, and taking an approved action. Companies decide what stays consistent across deployments — policies, evaluations, escalation rules — and what changes per workflow or channel, which the company says lets teams expand to new use cases without starting over.

Presence bundles the components OpenAI says teams need to run agents in production: policies and standard operating procedures, guardrails, approved actions, simulations, evaluation tools, and a Codex-powered improvement process.

The 1-888-GPT-0090 benchmark

The most concrete evidence OpenAI offers for Presence comes from its own operations. The product powers OpenAI's English-language phone support channel at 1-888-GPT-0090, handling open-ended requests, verifying callers, using account context, and taking approved actions. According to the company, the agent met or exceeded OpenAI's frontline human-support quality benchmarks within weeks, and now resolves 75% of inbound issues without human assistance.

The improvement loop delivered measurable gains on a short timeline. Working with OpenAI's launch team, the Codex-powered improvement process reduced human handoffs by 15 percentage points in just 10 days. That figure is the clearest signal in the announcement of what OpenAI is actually selling: not a static model, but a system that gets measurably better after launch.

Several large enterprises are building on the same foundation. BBVA is exploring AI-powered voice support for everyday banking needs in Mexico. SoftBank is testing natural Japanese-language customer conversations. IAG, the insurance group, is exploring timely support during high-demand events such as severe weather. All three are described as explorations or tests rather than full production deployments.

Trust before, during, and after launch

Presence's evaluation machinery operates in three phases, according to the announcement. Before a deployment reaches users, teams can test it against common requests, edge cases, and higher-risk scenarios. Simulations and graders check whether the agent reached the right outcome, followed policy, used tools correctly, and escalated when appropriate. Guardrails can intervene when an interaction moves outside the company's boundaries.

That work continues after launch. Production sessions, escalations, and quality signals show teams where the agent is working well and where it needs attention as policies, products, and user behavior change. Codex, using the Presence plugin, investigates those signals and suggests updates. Teams can test each proposed change against the version in production, then approve a controlled rollout.

OpenAI emphasizes control stays with the business throughout. "Presence is also designed to learn with you—improving as your business, customers, and employees evolve while the business stays in control," the company says. The product is built on OpenAI's research base and is positioned to advance as the underlying models improve. When a use case goes beyond what the product supports today, OpenAI Forward Deployed Engineers (FDEs) and partners can work with the customer to bring it into production.

Availability and what it signals

OpenAI Presence is available to eligible enterprise customers as a deployed product through a limited general availability program. Deployments are led by OpenAI Forward Deployed Engineers and select global systems integrators. Presence is not yet available as a self-serve product. OpenAI says it will continue supporting voice customers with access to its frontier models through the OpenAI API, and directs interested organizations to contact their OpenAI account team.

The delivery model is itself part of the story. By selling a hands-on, FDE-led deployment rather than an API endpoint, OpenAI is competing on operational outcomes — resolution rates, handoff reduction, policy compliance — rather than raw model capability. That puts it in the same territory as enterprise AI deployment firms and systems integrators, while the research feedback loop from every deployment gives the company compounding data advantages across its customer base.

For enterprises weighing agent deployments, the 75% autonomous resolution rate and the 15-point handoff reduction achieved in 10 days at 1-888-GPT-0090 now serve as OpenAI's reference case. Whether Presence can replicate those numbers beyond OpenAI's own support channel — and in BBVA's Mexican banking operations, SoftBank's Japanese-language conversations, and IAG's weather-driven demand spikes — will determine whether the product's "battle-tested" framing holds at scale.