EPISODE 2026-08-24

Cloud AI and Open Innovation

AI:AM on agentic cloud infrastructure, why CPUs still matter, Shenzhen open innovation, edge AI hardware, and AI safety risks.

▶ Full show on YouTube

An episode on agentic cloud infrastructure, CPU coordination, Shenzhen hardware, and the safety risks that come with more autonomous AI.

The rundown

4:29Opening26 min
OpeningHow do AI agents produce a podcast? The AI:AM hosts unpack a no-staff production workflow, from dynamic speaker switching and vibe coding to Claude model division of labor, zero-data-retention tradeoffs, and why RAG still matters. They also debate continual learning, model releases, and the gap between raw capability and deployment.
30:08Interview47 min
Mohamed AwadMohamed AwadWhy do AI agents need CPUs? Arm EVP Mohamed Awad explains how always-on agentic workloads change cloud infrastructure—from CPU coordination and model routing to power efficiency and memory bandwidth. He also discusses Arm's Meta partnership, custom CPUs, AI hardware cycles, supply-chain constraints, and why the value of AI is intelligence rather than tokens.
Watch
Open segment on YouTube ↗
Clips
- The CPU Is Not Dead
  0:00 / 0:00
- Why Agentic CPUs Never Sleep
  0:00 / 0:00
- The Next AI Bottleneck Is A Capacitor
  0:00 / 0:00
1:17:27Interview59 min
David LiDavid LiWhy is Shenzhen moving so fast on AI hardware? David Li, founder of Shenzhen Open Innovation Lab, explains China's open innovation ecosystem, flexible factory robots, edge AI machines, small models, AI toys, and how the Chinese AI market differs from the US.
Watch
Open segment on YouTube ↗
Clips
- Robots First Take Dangerous Jobs
  0:00 / 0:00
- AI Will Fill Every Product
  0:00 / 0:00
- The Pick-A-Side Strategy Fails
  0:00 / 0:00
- Rogue-Agent Hype Is Theater
  0:00 / 0:00
- Copying Creates New Products
  0:00 / 0:00
2:16:38Closing38 min
ClosingCan AI agents be trusted? This closing conversation examines rogue-agent incidents, Chinese and American AI security, data-center access, and the case for slowing development responsibly. It also covers agent attribution, prompt injection, runaway tasks, and the risks of training systems to maximize money.

In this episode

Prakash Narayanan and Nathan Labenz open with a look at how their own AI-assisted podcast workflow was built with Claude, then speak with Mohamed Awad of Arm about why CPUs still matter for always-on agentic workloads. Later, David Li of Shenzhen Open Innovation Lab explains Shenzhen’s open innovation pipeline, edge AI hardware, robotics, and how China’s product ecosystem differs from the U.S. The closing conversation turns to rogue agents, prompt injection, attribution, data-center access, and whether the U.S. should slow AI development.

How do AI agents produce a podcast? The AI:AM hosts unpack a no-staff production workflow, from dynamic speaker switching and vibe coding to Claude model division of labor, zero-data-retention tradeoffs, and why RAG still matters. They also debate continual learning, model releases, and the gap between raw capability and deployment.
Why do AI agents need CPUs? Arm EVP Mohamed Awad explains how always-on agentic workloads change cloud infrastructure—from CPU coordination and model routing to power efficiency and memory bandwidth. He also discusses Arm's Meta partnership, custom CPUs, AI hardware cycles, supply-chain constraints, and why the value of AI is intelligence rather than tokens.
Why is Shenzhen moving so fast on AI hardware? David Li, founder of Shenzhen Open Innovation Lab, explains China's open innovation ecosystem, flexible factory robots, edge AI machines, small models, AI toys, and how the Chinese AI market differs from the US.
Can AI agents be trusted? This closing conversation examines rogue-agent incidents, Chinese and American AI security, data-center access, and the case for slowing development responsibly. It also covers agent attribution, prompt injection, runaway tasks, and the risks of training systems to maximize money.