OpenAI Launches Presence, an Enterprise Service to Push AI Agents Into Production

OpenAI has introduced Presence, a new enterprise offering designed to help businesses deploy AI agents in actual production environments rather than internal testing or prototyping.

The move addresses a persistent gap in enterprise AI adoption: many companies can build agent demos, but few can reliably run them in customer-facing or operationally critical settings. According to OpenAI, Presence is built specifically for that transition.

What Presence Adds

OpenAI already offers Workspace Agents — customizable GPTs — but the company says those are primarily suited for internal use cases. Presence is positioned as a step beyond that, targeting production deployments in two areas: customer service and internal business workflows.

When a company's requirements exceed what Presence handles out of the box, OpenAI deploys what it calls Forward Deployed Engineers. According to OpenAI, these engineers work directly with customers to:

• Select appropriate workflows for automation
• Connect the agent to existing enterprise systems
• Establish operational guidelines
• Manage testing through to production launch

This white-glove, services-heavy model mirrors an approach other AI vendors have used to bridge the gap between general-purpose models and the specific, often messy requirements of enterprise IT environments.

Limited Availability, Undisclosed Terms

Presence is not an open, self-serve product. OpenAI has made it available only to qualifying enterprise customers, and the company has not disclosed pricing for the service or for any underlying model usage tied to it.

OpenAI also has not clarified how Presence addresses regulatory requirements such as the EU AI Act. The company references unspecified "trust mechanisms" in its agent framework but has not published legal or compliance documentation detailing how those mechanisms satisfy specific regulatory obligations in the EU or elsewhere.

No technical specifications — such as underlying model architecture, context window limits, or benchmark performance — have been disclosed for Presence, as it functions as a deployment and integration service layered on top of OpenAI's existing model infrastructure rather than a standalone model release.

What This Means

Presence signals that OpenAI is shifting some of its enterprise strategy from selling raw model access toward selling outcomes — packaged deployment services backed by human engineers, not just API endpoints. This mirrors a broader trend among frontier labs: as base model capabilities converge, differentiation increasingly comes from deployment tooling, integration support, and reliability guarantees rather than raw benchmark scores.

The reliance on Forward Deployed Engineers also suggests OpenAI is acknowledging that off-the-shelf agents still can't handle the last mile of enterprise deployment — legacy system integration, workflow-specific guardrails, and production-grade testing — without significant human involvement. That's a tacit admission that "AI agents" remain far from plug-and-play for many businesses.

The unresolved compliance question is the bigger flag. Enterprises in regulated industries, particularly in the EU, will need concrete answers on AI Act conformity before they can commit budget to a system with unclear regulatory footing. Until OpenAI publishes specifics, Presence will likely see slower adoption among risk-averse, heavily regulated customers than among domestic U.S. enterprises with fewer compliance hurdles.