Published on August 12, 2026
OpenAI Presence is a managed enterprise platform for deploying governed voice and chat agents that can use company systems, follow operational policies, take approved actions and transfer cases to people when human judgment is required.
OpenAI introduced Presence on July 22, 2026, as a platform for organizations that want to move AI agents beyond limited demonstrations and into high-volume, business-critical workflows.
The central challenge Presence addresses is not simply whether an AI model can answer a question. Enterprises need agents that can access the right information, follow company procedures, use approved systems, complete permitted actions and recognize when a case should be transferred to a person.
According to OpenAI’s Presence announcement, each deployment combines model reasoning with policies, permissions, guardrails, evaluations and escalation rules. The platform currently supports voice and chat experiences for customer-facing services and internal operations.
Presence therefore represents a more controlled approach to enterprise AI. Rather than giving a general-purpose assistant broad access and hoping it behaves correctly, organizations define a specific job, the information required to perform it and the boundaries within which the agent can act.
A language model can interpret requests and generate responses, but a production agent needs additional infrastructure.
Consider an AI agent responsible for resolving a billing problem. It may need to identify the customer, retrieve account information, interpret the company’s billing policy, decide which remedy is permitted, update an internal system and explain the result. If the case involves fraud, a policy exception or an unusually large adjustment, it may need to stop and transfer the interaction to a specialist.
OpenAI Presence brings these components together within a managed deployment. A company determines which knowledge sources and systems the agent can access, which actions it can perform independently and which decisions require approval.
The agent is not intended to receive unlimited access to an organization. OpenAI says each deployment begins with a defined workflow, and the agent receives only the knowledge and system permissions required for that job. The exact models, integrations, data arrangements and service commitments are determined separately for each customer.
This job-specific structure may be particularly important in sectors where the same organization handles workflows with very different risk levels. Answering a routine question about opening hours is not equivalent to modifying a financial account, processing an insurance claim or changing an employee’s access permissions.
The most important feature of Presence is its governance layer.
Organizations can encode standard operating procedures, approval requirements and escalation conditions into the agent’s workflow. These policies define not only what the agent should say, but what it is permitted to do.
A customer-support agent, for example, might be authorized to retrieve an order, confirm its delivery status and issue a replacement below a specified value. It might require human approval for a refund above that limit and automatically escalate cases involving suspected fraud or legal threats.
Presence also uses guardrails to intervene when an interaction moves outside the company’s approved boundaries. These controls can restrict inappropriate actions, identify higher-risk situations or prevent the agent from continuing when required information is missing.
Before deployment, simulations and evaluations can test whether the agent reaches the correct outcome, follows policy, uses tools appropriately and escalates when necessary. OpenAI’s Presence product page describes these controls as part of every stage of deployment rather than as protections added after the agent has already been launched.
This distinction matters because a convincing answer is not always a correct operational result. In enterprise workflows, the agent must also use the right system, follow the approved process and leave an auditable record of what occurred.
Presence agents are designed to do more than provide information. With scoped permissions, they can retrieve records, update business systems and complete approved actions.
This moves the interaction from conversational assistance toward resolution. A support agent could potentially verify a caller, check account context and complete an authorized service request. An internal IT agent might guide an employee through troubleshooting, reset access under approved conditions or route a more complex security issue to the appropriate team.
OpenAI currently highlights customer support, inbound lead qualification, insurance claims, human resources, IT help desks and procurement as potential use cases. In each case, the agent combines conversation with a repeatable operational process.
For customer-facing workflows, Presence can maintain shared policies and evaluation standards across voice and chat while allowing the interaction design to change by channel. A voice agent may need to confirm information verbally and handle interruptions, while a chat agent may present links, structured choices or written summaries.
The value for organizations is not simply that one agent can operate in two formats. It is that the same operational rules, permissions and quality standards can remain consistent across them.
Presence is not presented as a platform that removes people from every interaction.
Human escalation is treated as a designed part of the workflow. Organizations establish when an agent should transfer a case, what information should accompany the transfer and which person or team should receive it.
This can be triggered by policy, risk, uncertainty or the need for professional judgment. An insurance agent might handle routine information collection but transfer a disputed claim to a specialist. A human-resources agent could explain a benefits policy while escalating a sensitive employee case. A sales agent might qualify an inbound lead before handing it to the appropriate representative.
The OpenAI Presence help documentation states that escalation paths can include structured context for the receiving team. This is important because a handoff creates little value when the customer or employee must repeat the entire situation to a person.
The broader objective is selective automation: the agent resolves repeatable cases while preserving human involvement where empathy, accountability, negotiation or expert judgment matters most.
Enterprise workflows change after deployment. Products are updated, policies are revised and users discover unexpected ways to describe their problems.
Presence uses production sessions, escalation patterns and quality signals to identify areas where an agent may need improvement. OpenAI says Codex can investigate these signals and propose updates to the agent’s instructions or behavior.
Those changes are not intended to enter production automatically. Teams can test a proposed version against the existing agent, review its performance and approve a controlled rollout. Monitoring and rollback processes can also form part of the deployment.
This improvement loop reflects a practical reality: an enterprise agent is not finished when it launches. It needs to adapt without abandoning the company’s governance requirements.
The system may become more capable through experience, but the organization remains responsible for deciding which changes are acceptable.
OpenAI says Presence powers its English-language AI phone-support channel, where it can answer open-ended questions, verify callers, use account information and perform approved actions.
The company reports that the agent now resolves 75% of inbound issues without human assistance and that its improvement process reduced human handoffs by 15 percentage points over ten days. These are OpenAI’s own operational results and should not be treated as guaranteed outcomes for other organizations.
OpenAI also identifies BBVA, SoftBank and Insurance Australia Group as early Presence collaborators exploring use cases in financial services, Japanese-language support and insurance assistance during high-demand periods.
These examples illustrate the intended market: organizations with large volumes of repeatable interactions, established operational policies and enough technical readiness to connect an agent securely to business systems.
Presence is available through limited general availability for eligible enterprise customers. It is not currently a self-service product that an organization can activate directly through a standard ChatGPT subscription.
Deployments are led by OpenAI Forward Deployed Engineers, selected systems integrators or both. Access depends on the suitability of the workflow, implementation readiness and available deployment capacity.
Pricing, models, integrations, supported channels, data handling and service commitments are defined for each deployment. Organizations must contact their OpenAI account team to discuss access and scope.
This managed model makes Presence less accessible than a conventional software platform, but it also reflects the complexity of deploying agents in workflows involving sensitive information and consequential actions.
OpenAI Presence is designed to close the gap between an AI agent that can demonstrate a task and one that can perform it reliably inside a real organization.
Its defining feature is not voice, chat or access to a powerful model in isolation. It is the combination of company policies, scoped permissions, approved actions, evaluations, guardrails and structured human escalation.
For businesses, this could make AI agents more useful in customer support, claims, sales, HR, IT and procurement. The strongest opportunity lies in repeatable workflows where an agent can resolve routine cases while transferring uncertainty, exceptions and higher-risk decisions to people.
Presence also shows where enterprise AI is heading. The competitive question is becoming less about which model produces the most impressive response and more about which system can act consistently, improve safely and remain accountable once it enters production.
OpenAI — Introducing OpenAI Presence — Published July 22, 2026.
OpenAI — OpenAI Presence — Accessed August 5, 2026.
OpenAI Help Center — OpenAI Presence — Updated July 2026; accessed August 5, 2026.
