AGENT DESIGN · CLIENT AUTOMATIONS
OpenAI launched Presence on July 22, an enterprise agent platform built around simulating a support agent against common, edge-case, and high-risk requests, grading it on policy adherence and escalation, before any of it goes live. This is a small working demo of that shape: the same six customer messages, answered once by a bare agent working straight off an SOP doc, and once by an agent behind a gate that checks policy branches, action authority, and named escalation triggers first.
Same six requests, run before and after a gate
A bare support agent answering straight off an SOP doc, versus the same agent behind a simulate-and-grade gate: policy adherence, approved-action limits, and named escalation triggers checked before a reply goes out. Click each one to run it.
COMMON REQUEST
COMMON REQUEST
EDGE CASE
EDGE CASE
HIGH-RISK
HIGH-RISK
WHY THIS MATTERS FOR CLIENT WORK
I build chatbots and support automations for small PH businesses that will never afford OpenAI's Forward-Deployed-Engineer rollout, but the underlying idea does not need their platform. Write down the edge cases and the high-risk cases a founder actually worries about, run the agent against all of them before launch, and grade it on whether it stayed inside its authority and escalated when it should have, not just whether the reply sounded right. A prompt that says "be careful with refunds" is not a gate. A written test case with a pass condition is.
Built 23 July 2026 · pattern sourced from the OpenAI Presence launch