OpenAI Presence packages guardrails for enterprise AI agents

The interesting move in OpenAI's Presence launch isn't the agent tech, it's who OpenAI is willing to send along with it. As Help Net Security reported, Presence bundles policies, guardrails, approved actions, simulations, evaluation tools and a Codex-powered improvement loop into one platform for voice and chat agents that touch real company systems. Crucially it isn't self-serve. Enterprises engage through OpenAI account teams and its Forward Deployed Engineers, The Register noted, a consulting arm reinforced by the May acquisition of Tomoro.

The numbers OpenAI is putting behind it are the eye-catch. The company says Presence already runs its own English-language phone support and resolves 75% of inbound calls without human intervention, and that the Codex review loop cut handoffs by 15% over a 10-day period. Take those as reported, not settled, since they are OpenAI's numbers on OpenAI's own workflow.

The named early explorers give a better read on the target market. BBVA is trying it on routine banking calls in Mexico. SoftBank is testing Japanese-language customer conversations, with Tadahisa Murakami, its VP of data and digital transformation, saying frontline teams rate the agent's Japanese "highly for their natural and accurate quality." Australian insurer IAG is exploring support during high-demand periods such as severe weather. That is regulated finance, non-English voice, and a compliance-heavy insurer, the exact accounts where a stack of guardrail options is not enough on its own and where the differentiator is the humans that come with the software.

The honest caveat is that pricing is not public. The Register says broader pricing is "to come as availability expands" and current deployments are "scoped individually based on each customer's use case," which is another way of saying enterprise consulting rates. What the reporting also doesn't give you is any independent benchmark of Presence's guardrail or evaluation tooling against rival agent platforms, so the moat here is process and people, not model.

If the model layer really is commoditizing, whoever owns the deployment layer (the graders, the sim harness, the change-review loop) collects most of the enterprise budget. Presence is OpenAI's bet that it can be that layer for its biggest customers before Salesforce, ServiceNow or a big-four consultancy locks them in first.