Jon Kershaw - AI Superusers Presentation - 23 Jun 2026

In this talk, Jon Kershaw makes the case that in the agentic era, "intelligence never grants authority": capability is not permission, and an AI agent should have to earn the right to execute. Drawing on a site-reliability background and his recent work at Red Hat, he walks through "Agent Responsibility Engineering," a deliberately deterministic governance layer that sits between an agent's decision and its action. The mechanics are concrete: every agent is issued a scoped, expiring "passport" at creation; policy is enforced as code (OPA/Rego) that deterministically allows, denies, or suspends an action; and every step is written to an immutable, hash-chained ledger, so in regulated industries you can later prove not just who did what, but that it was allowed. Four questions drive the whole system: who is acting, what are they doing, is it allowed, and what proves it. They are the same ones he says he asks his kids. He demos it through "Nigel," a well-meaning compliance officer who clicks the one button to promote a model to production he shouldn't. Nigel isn't malicious, just ignorant, which Kershaw points out is actually worse. His warning lands in a single line: once a prompt reaches a button, hope is no longer a control. Jon Kershaw, Principal Platform Engineer, AI at Red Hat. Connect with Jon on LinkedIn: https://www.linkedin.com/in/jonjonkershaw/ This presentation was delivered at a meeting of AI Innovators of Wichita on the Newman University campus. Join AI Innovators of Wichita on Meetup: https://www.meetup.com/ai-innovators-of-wi... Learn more about Newman University's programs: Newman's new Master of Science in Applied AI, launching Fall 2026: https://newmanu.edu/academics/applied-ai-ms Newman's Master of Science in Data Science: https://newmanu.edu/academics/data-science-ms Newman's new MBA concentration in AI Strategy: https://newmanu.edu/academics/ai-strategy-mba