What Is OpenAI Presence? The Enterprise Agent Platform Resolving 75% of OpenAI's Support Calls

OpenAI announced OpenAI Presence on July 22, 2026 — a product for putting AI agents to work in production business workflows¹. Beyond answering questions and resolving issues, Presence agents can use company systems, take approved actions, and escalate to people when needed, all governed by company-defined policies and guardrails¹.

OpenAI frames the product bluntly: “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.”¹ Presence is aimed at the gap between agent demos and dependable operations — the stage after the proof of concept.

Start with One Job, Let Codex Keep Improving It

Each Presence deployment starts with a specific job, such as resolving billing issues, supporting insurance claims, or handling employee IT service requests¹. The agent receives only the knowledge and system access required for that job, and the company sets the policies: what the agent can do, when it needs approval, and when a person should take over¹.

The product bundles the components needed to run agents in production: policies and standard operating procedures, guardrails, approved actions, simulations, evaluation tools, and a Codex-powered improvement process¹. Before launch, simulations and graders check whether the agent reached the right outcome, followed policy, used tools correctly, and escalated when appropriate. After launch, production sessions and escalations reveal gaps; Codex — OpenAI’s coding agent — investigates quality signals and proposes updates, which teams can test against the production version before approving a controlled rollout¹.

Today Presence supports real-time voice and chat experiences, with customer support, outbound sales, and high-risk internal workflows cited as target uses¹.

Proven on OpenAI’s Own Phone Line: 75% Resolved Without Humans

OpenAI’s proof point is its own English-language phone support channel at 1-888-GPT-0090, which Presence powers — handling open-ended requests, verifying callers, using account context, and taking approved actions. It now resolves 75% of inbound issues without human assistance¹. Within weeks of going live, it met or exceeded the benchmarks OpenAI uses to grade frontline human-support quality, and its Codex-powered improvement loop reduced human handoffs by 15 percentage points in just 10 days¹.

Among external adopters, BBVA is exploring AI-powered voice support for everyday banking needs in Mexico and serves as a design partner for the voice experience¹. SoftBank is testing natural Japanese-language customer conversations, and Australian insurer IAG (Insurance Australia Group) is exploring timely support during high-demand events such as severe weather¹³. Japanese-language customer service appearing among the first use cases is worth noting for readers in Japan.

The Backstory: Production Operations Became the Industry’s Focus

Through 2026, the industry’s attention has shifted from making agents work to keeping them safe and reliable in operation. On the protocol side, the MCP specification’s 2026-07-28 revision expanded how approvals and safety are handled between agents and tools. On the workspace side, products like Block’s Buzz are built around teams of humans and agents. OpenAI itself has been shoring up the model layer with releases like GPT-5.6’s general availability.

Presence productizes the operations layer rather than the model: policy configuration, guardrails, simulation, and a continuous improvement loop — essentially packaging what an AI agent needs before it can be trusted with production work.

Enterprise-Only for Now, but a Useful Checklist

Availability is the catch. Presence is offered to eligible enterprise customers through a limited general availability program, with deployments led by OpenAI’s Forward Deployed Engineers (FDEs) and select global systems integrators. It is not yet available as a self-serve product¹, and pricing has not been disclosed². Smaller organizations cannot simply sign up.

Even so, the component list this launch lays out — job-scoped access, company-controlled approval policies, pre-launch simulation testing, and a post-launch improvement loop — reads as a vendor-neutral checklist for anyone planning an agent deployment. Companies weighing customer-support automation can also calibrate expectations against OpenAI’s own trajectory: reaching 75% autonomous resolution took weeks of tuning plus a continuous improvement loop, not a turnkey launch¹.