OpenAI launches Presence, managed enterprise AI agents

OpenAI announced Presence on July 22 as a managed enterprise AI agent product offered in limited general availability. Deployments are led by OpenAI Forward Deployed Engineers and a set of selected systems integrators, and the product is sold as a project rather than a self-serve service.

Each engagement starts with a single, defined task such as resolving a billing dispute, handling an insurance claim or clearing an employee IT request. The agent receives only the data and system access required for that task. Customers set rules for agent behavior, including when the agent must seek sign-off and when a human should take over. After launch, OpenAI’s Codex system reviews production sessions and escalations and proposes changes that customers test and approve before wider rollout.

OpenAI’s documentation outlines a six-stage deployment process that runs from scoping business outcomes through security, privacy and legal review, simulation and acceptance testing, staged rollout and post-launch iteration. The documentation states that an agent does not become production-ready simply by ingesting documents, and it emphasizes integration, permissions and change management.

Presence includes simulation and grading tools to verify outcomes and policy compliance, guardrails to limit agent actions, session records and action histories for audits, escalation paths that provide structured context for staff, and controlled rollouts with rollback capability.

Access is limited by workflow fit, implementation readiness and available delivery capacity. OpenAI evaluates customers on those criteria. Delivery capacity reflects a consulting constraint: Forward Deployed Engineers work embedded with customer systems, require clearances and perform time-intensive integration work, so deployments do not scale in the same way as API-based software.

Because OpenAI combines model provision with hands-on implementation, contracts document the company’s role as both vendor and integrator and specify accountability for production behaviour and policy application.

OpenAI describes Presence as built from prior agent deployments and cites internal testing on its English-language phone support line, 1-888-GPT-0090. The company reports the agent met or exceeded internal benchmarks for frontline support, resolving 75% of inbound issues without human assistance and reducing human handoffs by 15 percentage points in ten days through the Codex improvement loop. Those figures originate from OpenAI and have not been independently verified.

Three named early partners are BBVA, which is testing voice support for everyday banking in Mexico; SoftBank, which is testing Japanese-language conversations; and IAG, which is exploring agent support for high-demand events such as severe weather. OpenAI identifies some participants as design partners that help shape and refine experiences; the company does not present any of these customers as running Presence at scale.

OpenAI has not published pricing. Implementation scope and cost are set case by case. Presence uses OpenAI models, but specific model names and fixed configurations are not published; configurations are selected for each workflow and may change over time.

During the limited general availability period, channel support covers voice and chat with contact-centre integration, routing, authentication and handoff design confirmed per deployment. Data handling and system architecture follow the signed contract and the deployment design rather than a single public policy.

Presence is offered alongside ChatGPT Workspace Agents for self-serve teams and API access for voice customers. The three delivery routes provide similar technical capabilities but differ in who performs the implementation. OpenAI states that delivery capacity is being rationed, and buyers choose the delivery path that matches their readiness and operational needs.