OpenAI launches Presence to stop AI agents from going rogue in production

Getting an AI agent to work in a demo is easy. Getting it to not embarrass you in front of actual customers is a whole different beast. OpenAI’s new offering, Presence, is its latest attempt to bridge that chasm for enterprise clients.

The company already had Workspace Agents — those customizable GPTs you can spin up for internal tasks. But Presence is designed for the sharp end of the stick: customer service and production workflows where mistakes have immediate, public consequences. It’s a recognition that the gap between a cool prototype and a reliable business tool remains stubbornly wide. When the out-of-the-box stuff isn’t enough, OpenAI dispatches its Forward Deployed Engineers to effectively become your AI SWAT team, wiring the agent into your existing systems, setting guardrails, and managing the whole mess through to launch.

But here’s what’s conspicuously absent: any hard detail on compliance. The specter of the EU AI Act looms large, and all OpenAI is offering right now is a vague mention of “trust mechanisms.” No legal frameworks, no data residency guarantees, no audit trails spelled out. For any enterprise with a serious legal department, that’s less a footnote and more a flashing red light.

I’ve watched enough enterprise AI rollouts to know that the engineering is only half the battle. The real killer is governance. Presence might solve the technical plumbing, but if OpenAI doesn’t get specific about how this squares with regulators — particularly in Europe — it’s selling a car without seatbelts. The Forward Deployed Engineers can’t code their way out of a GDPR fine.