https://tube.pompat.us/videos/watch/86993a4a-c991-40f1-ada4-dd42ecfe6535 Magnus Hedemark of Groktopus argues that safe AI adoption depends on clear ownership, authentication vs. authorization, and a model he calls “earned autonomy,” where humans grant and revoke AI autonomy based on trust and guardrails. He explains how AI mishaps like deleting production databases often come from overbroad permissions, especially when agents inherit a human owner’s credentials. Instead, he recommends creating dedicated AI credentials with narrowly scoped permissions tied to a defined purpose statement—e.g., allowing an agent to read and inspect a production database to surface issues, but not write, delete, or repair anything—plus limited notification abilities (Slack, PagerDuty). He emphasizes that earned autonomy starts with human governance conversations and introduces his AI Governance agent skill designed to guide teams in building frameworks, meetings, artifacts, cadence, and decision boundaries. 00:00 AI & Earned Autonomy 00:06 Introduction 01:38 Authentication & Authorization 03:05 Earned Autonomy 06:25 Purpose & Guardrails 08:06 AI Governance Tool Mon, 21 Sep 2026 21:04:55 GMT https://validator.w3.org/feed/docs/rss2.html PeerTube - https://tube.pompat.us https://tube.pompat.us/lazy-static/avatars/db1923f1-57e5-4d93-a997-3eb296f149c9.png https://tube.pompat.us/videos/watch/86993a4a-c991-40f1-ada4-dd42ecfe6535 All rights reserved, unless otherwise specified in the terms specified at https://tube.pompat.us/about and potential licenses granted by each content's rightholder.