POC Prison: Why agentic systems never escape the lab and how to fix that in 90 days
summary
Claim: agentic systems fail because enterprises lack ownership, data readiness, and delivery plumbing not because models. Offers a 90-day program (30 days reality mapping, 30 days assign owner and governance, 30 days production under supervision) executed 22 times with one or two use cases reaching prod each time. Builds a paved road: APIs to SAP/Salesforce, service-principal identities, audit logs and kill switches, CI/CD pipelines and dashboards, and a small empowered tiger team that implements governance-as-code so an agent actually runs in production.
AI-written summary · about
chapters (10)
- 0:00Agentic AI and PC prison
- 3:30Introduction
- 7:02Why agentic AI fails
- 16:48AI zombie systems
- 26:4390-day escape plan
- 35:21Success criteria
- 41:50Technical escape plan
- 49:10Homework for production
- 55:09POCs without production path
- 56:33Closing
related talks
18mFrom POC to Profit: Deploying Customer-Facing AI Agents That Actually Work
39mFrom Demo to Production: Why Agentic AI Systems Fail - and How to Fix Them
43mFrom Pilots to Practice: Leading the agentic AI transformation
27mAgentic Systems: Leadership, Risk, and Accountability
37mThe Production AI Playbook: Deploying Agents at Enterprise Scale
33mThe Autonomous Enterprise
46mPanel Discussion: The Agentic Software Factory That Enterprises Actually Need
40mHow To Tame Your Agents?