#562 Agentic AI Is Not an Intern: Craig McLuckie on Control, Context, and Enterprise Reality

Agentic AI is moving faster than enterprise readiness.

Boards are pushing adoption. Teams are deploying agents at speed. But security, control, and operational discipline are lagging behind.

In this episode, Mehmet sits down with Craig McLuckie, the co-creator of Kubernetes and founder of Stacklok, to unpack why most agentic AI initiatives break after the demo and what enterprises must do differently to make them durable, secure, and production-ready.

From MCP and context engineering to eval-driven development and why AI agents should never be treated like interns, this conversation goes deep into the realities CTOs, VPs of Engineering, and security leaders are facing right now.

This is not a hype conversation. It’s an operator’s reality check for 2026.

⸻

👤 About the Guest

Craig McLuckie is a foundational figure in modern cloud infrastructure. He is the co-creator of Kubernetes, founder of the Cloud Native Computing Foundation, and former VMware executive behind the Tanzu portfolio.

Today, Craig is the founder and CEO of Stacklok, where he is focused on helping enterprises securely connect agentic AI systems to real-world infrastructure through open, controlled, and auditable platforms.

https://www.linkedin.com/in/craigmcluckie/

⸻

🧠 Key Takeaways

• Why agentic AI represents a true epoch shift, not just another tooling cycle

• The real difference between demos, POCs, and production AI systems

• Why MCP is powerful but dangerous without proper control layers

• How context engineering is becoming more important than writing code

• Why eval-driven development replaces test-driven development in AI systems

• How enterprises should think about permissions, scope, and agent autonomy

• Why most AI failures are workflow problems, not model problems

• What 2026 realistically looks like for agentic AI adoption in the enterprise

⸻

🎯 What You’ll Learn

• How to operationalize agentic AI without exposing your infrastructure

• Why treating AI agents like humans is a security mistake

• How to design guardrails without slowing teams down

• Where CTOs should focus investment to move from hype to ROI

• How leadership metrics and engineering evaluation must evolve in the AI era

⸻

⏱ Episode Highlights & Timestamps

• 00:00 – Introduction and Craig’s journey from Google to Kubernetes

• 03:10 – Why agentic AI feels like a historic inflection point

• 06:05 – MCP explained and where enterprises get it wrong

• 10:45 – The security risks nobody is talking about

• 14:20 – Why AI agents should never be treated like interns

• 18:30 – The danger of permission sprawl and tool pollution

• 23:10 – Why most AI initiatives fail after the demo

• 28:40 – Eval-driven development vs traditional software thinking

• 34:15 – Context engineering as the new leverage point

• 38:50 – How engineering leadership and metrics must change

• 43:30 – What realistic agent adoption looks like in 2026

• 46:20 – Open source, ToolHive, and building durable AI platforms

⸻

🔗 Resources Mentioned

• Stacklok: http://stacklok.com/

• ToolHive (Open Source MCP Platform): https://stacklok.com/toolhive/