OpenAI Presence delivers enterprise AI agents with engineers

OpenAI launched Presence on July 22 as a managed enterprise product offered in limited general availability. Deployments are handled by OpenAI’s Forward Deployed Engineers and a group of selected systems integrators rather than through a self-serve option.

Presence is sold as a project-based service that begins with a defined task, such as resolving a billing dispute, handling an insurance claim or processing an IT service request. Agents are given only the data and system access needed for the assigned task. Customers set the agent’s rules, approval gates and handoff points.

After launch, an internal feedback process called Codex reviews production sessions and escalations and proposes changes. Customers test and approve those changes before any update is rolled out.

OpenAI’s documentation sets out a six-stage implementation process that covers scoping business outcomes, security, privacy and legal review, simulation and acceptance testing, staged rollout and post-launch iteration. The company says simulations and grading tools check whether an agent reached correct outcomes, followed policy, used tools properly and escalated when required. Session records, action histories and structured escalation paths are included to support auditing and human reviewers. Controlled rollouts can be rolled back if needed.

Access to Presence is limited by workflow fit, customer readiness and OpenAI’s available delivery capacity. OpenAI uses Forward Deployed Engineers-staff embedded with customers for configuration and integration work-alongside named systems integrators to install and configure agents. Eligibility and deployment scope are determined per customer and per project.

OpenAI cites performance figures from an internal support channel, 1-888-GPT-0090, reporting that an agent met internal benchmarks, resolved 75% of inbound issues without human assistance and reduced human handoffs by 15 percentage points within ten days through the Codex loop. Those measurements are from OpenAI’s internal grading and have not been independently verified.

Three customers are named in the limited launch. BBVA is listed as a design partner working on voice support for banking in Mexico, SoftBank is testing Japanese-language conversations, and IAG is exploring agent use during high-demand events such as severe weather. OpenAI does not present any of these deployments as operating at large scale.

Pricing is not published and is negotiated for each customer and deployment. OpenAI does not specify a single model for Presence; the product uses OpenAI models with configurations selected for each workflow and subject to change as the workflow evolves. During limited general availability, channel support covers voice or chat with contact-center integration, routing, authentication and handoff design confirmed on a deployment-by-deployment basis. Data handling and architecture are governed by the signed contract and the agreed deployment architecture rather than a single published policy.

Presence is offered separately from ChatGPT Workspace Agents, which remain a self-serve path for teams building in ChatGPT and Slack, and from API access to OpenAI’s frontier models for voice customers. The company now provides similar agent capabilities through multiple delivery paths that differ mainly by who performs integration and operational work.