OpenAI Presence: AI Agents Resolve 75% of Customer Calls

30-Second Summary

OpenAI launched Presence on 22 July 2026, a platform designed to deploy AI agents across customer and internal business workflows through voice and chat. The system brings together policy controls, guardrails, approved action sets, simulation tools, evaluation frameworks, and a Codex-powered improvement loop in a single deployment platform. OpenAI has confirmed Presence already handles its own English-language phone support line and resolves 75% of inbound calls without any human involvement, making it one of the most concrete AI agent performance disclosures to date.

At a Glance

• Topic: Agent Systems
• Company: OpenAI
• Date: 22 July 2026
• Announcement: OpenAI Presence, an enterprise AI agent deployment platform for voice and chat workflows
• What Changed: Businesses can now deploy AI agents that handle customer and internal service requests with built-in guardrails, approved actions, and structured escalation to humans
• Why It Matters: A 75% call resolution rate without human intervention signals that AI agents are commercially viable in routine customer-facing roles
• Who Should Care: Business operators managing customer support, sales development, HR, IT services, or procurement functions

Key Facts

• Company: OpenAI
• Launch Date: 22 July 2026
• What Changed: A production-ready AI agent platform for enterprise voice and chat deployments
• Who It Affects: Businesses running customer communications, sales development, IT helpdesk, HR services, or procurement workflows
• Primary Source: OpenAI (openai.com/index/introducing-openai-presence)

What Happened

On 22 July 2026, OpenAI launched Presence, a deployment platform that lets businesses run AI agents across customer support and internal service workflows. The platform is designed as an end-to-end deployment environment rather than a standalone model or simple chatbot builder.

Presence brings together the components required to run agents in production: policies and standard operating procedures, guardrails, approved action sets, simulation tools for pre-launch testing, evaluation frameworks for ongoing performance review, and a Codex-powered process for continuous improvement. Agents deployed through Presence can verify customer identity, access live account data, apply company-specific policies, and execute approved actions including billing resolution and refunds. When a request falls outside defined guardrails or agent capability, the system escalates to a human.

OpenAI confirmed that Presence already handles its own English-language phone support line and resolves 75% of inbound calls without human intervention. The company is using this internal deployment as a proof of concept before broader rollout. Target use cases named by OpenAI include customer support, outbound sales development, procurement, IT services, and HR functions.

The platform is currently rolling out through a limited general availability programme for enterprise customers, with OpenAI Forward Deployed Engineers and select integration partners involved in each implementation. OpenAI has not disclosed pricing, geographic availability, or the expected timeline for broader access.

Why It Matters

• A 75% call resolution rate without human intervention is a commercially significant benchmark. It shifts the question from "can AI handle customer calls?" to "how do we set it up correctly?"
• The guardrail and escalation architecture directly addresses the primary reason businesses hesitate to deploy AI in customer-facing roles: loss of control when situations go outside script.
• The Codex-powered improvement loop means deployed agents improve from live interactions over time rather than requiring manual retraining cycles.
• A single platform spanning customer support, sales development, HR, and IT reduces the cost and complexity of deploying multiple point solutions.
• OpenAI deploying Presence on its own support line before releasing it to customers is an unusually strong proof of confidence in the product.
• The simultaneous announcement that ChatGPT Work has reached 10 million users signals that AI agent adoption across businesses of all sizes is accelerating.

The David and Goliath View

For years, large organisations held a structural advantage in customer service: the budget to staff call centres, train agents consistently, and absorb the overhead of managing high volumes of routine requests. Presence changes the geometry. A business with 20 people can now deploy the same class of AI agent as a company with 2,000, including the policy controls, guardrails, and human escalation paths that make that deployment trustworthy rather than reckless.

The current limitation is access. Presence is rolling out through enterprise contracts with on-site OpenAI engineers, which puts it out of reach for most small and medium businesses in the short term. That will change. The pattern in AI deployment over the past three years has been consistent: enterprise-grade capability becomes mid-market accessible within 12 to 18 months of initial launch. The operators who use that window to document their workflows, draft their policies, and map their approved action sets will deploy faster and more safely than those who wait and build from scratch.

The practical action right now is not to wait for Presence. ChatGPT Work, available at approximately $25 per user per month and now used by 10 million users, handles multi-step tasks across connected tools including Slack, Google Drive, and Microsoft Teams. Operators who start building agent workflows with accessible tools today will have the institutional knowledge to scale to a platform like Presence when access broadens.

Where This Fits in the AI Stack

AI Growth Engine: AI agents that handle inbound customer queries and outbound sales development tasks free your team to focus on relationship-building and complex deals rather than routine communications.

Employee Amplification Systems: Internal agent deployments for IT helpdesk, HR queries, and procurement workflows extend your team's effective capacity without adding headcount, particularly useful in teams where people currently manage high volumes of repetitive internal requests.

Secure AI Brain: The policy, guardrail, and approved action architecture in Presence mirrors the governance layer that a well-constructed AI brain requires. Documented standard operating procedures and defined action boundaries are not just a Presence requirement, they are the foundation of any controlled, auditable AI deployment across your business.

Questions Operators Are Asking

Can a business with 30 employees use OpenAI Presence right now? Not directly. Presence is currently available only through a limited general availability programme for enterprise customers, with OpenAI engineers involved in each setup. Smaller businesses should evaluate ChatGPT Work and similar agent platforms available at small-business pricing as near-term options and monitor Presence for access expansion.

What makes Presence different from a standard chatbot? Standard chatbots follow scripted decision trees. Presence agents can verify identity, access live account data, apply company-specific policies, execute approved actions such as billing adjustments and refunds, and escalate to humans when a situation requires it. The Codex-powered improvement loop also means performance improves with live usage without requiring manual updates.

What does a 75% call resolution rate mean for staffing decisions? It means that for every 100 routine customer calls, 75 can be completed without a human agent. The remaining 25 are escalated to people. For a business spending significant team time on customer communications, that reduction could allow redeployment of that capacity into higher-value work or growth activity.

What groundwork do businesses need before deploying an AI agent platform? Any production-grade agent deployment requires documented policies and standard operating procedures, a defined set of approved actions the agent can take, access controls, and an escalation path. Businesses that have already mapped and documented their customer service and internal service workflows will deploy faster and with fewer problems than those starting from scratch.

Is Presence available outside the United States? OpenAI has not confirmed geographic availability for Presence. The current limited programme focuses on English-language customer support, which suggests initial deployment is primarily US-based. International availability timelines have not been disclosed.

Citable Summary

What happened: OpenAI launched Presence on 22 July 2026, a production AI agent platform for voice and chat workflows that already resolves 75% of the company's own inbound customer calls without human intervention.

Why it matters: The combination of policy guardrails, approved actions, human escalation paths, and a Codex-powered improvement loop makes AI agents in customer-facing and internal service roles commercially viable at scale.

David and Goliath view: The governance and escalation architecture in Presence signals where enterprise AI agent deployment is heading. Business operators who document their workflows and policies now will be positioned to deploy when access to platforms like this broadens beyond enterprise contracts.

Offer relevance:

• AI Growth Engine: AI-handled customer support and sales development queries free the team for higher-value relationship work
• Employee Amplification Systems: Internal agent deployment for IT, HR, and procurement extends team capacity without adding headcount
• Secure AI Brain: Policy and guardrail architecture is the operational foundation for controlled, auditable AI deployment