AI Can Already Act on Its Own. The Harder Question Is Who Answers for It

AI is moving beyond recommendation and into execution, bringing new questions around oversight, escalation and accountability. In this discussion, leaders from ParagonCorp, Kawan Lama Group and DANA Indonesia explore where AI can act autonomously, where human judgement still matters, how enterprises are designing escalation into workflows, and what it takes to make AI enterprise-ready at scale. 00:00 Intro 00:24 When an AI agent makes the wrong call 01:30 Meet the panel 02:36 Where AI moved from advising to executing 03:04 Paragon: AI as co-pilot, humans as the pilot 04:44 Kawan Lama: governing adoption after the fact 05:59 DANA: human in the loop vs human on the loop 08:35 Drawing the line in supply chain and R&D 11:44 Why the data foundation has to come first 14:22 AI-generated code shipping into production 17:36 Act Two: escalation, intervention, kill switches 18:23 What a kill switch actually looks like 19:35 Mistakes that don't announce themselves 22:58 Can a model learn expert intuition? 25:18 Guardrails without killing the speed advantage 27:57 When the AI got it wrong 28:05 Vibe coding: three weeks vs three months 30:26 Token economics and right-sizing your models 32:18 If you can't model it, don't outsource it 33:38 Act Three: who owns the outcome 35:40 A single named owner and the RACI model 37:21 Closing question to the panel