Micro VMs Move Into the AI Infrastructure Spotlight
Micro VMs are becoming more important as agentic AI workloads create new demands for speed, scale and security. In this Techstrong TV conversation, Barry Cooks, vice president of compute abstractions at Amazon Web Services, explains why a technology that already powers AWS Lambda is now gaining fresh attention. The shift is tied to AI agents that can launch unpredictable workloads, interact with data and execute code in ways that traditional infrastructure models were not always designed to handle.

The discussion focuses on three practical concerns for AI infrastructure teams. They need low latency when workloads spin up. They need an economic model that avoids paying for idle capacity. They also need stronger trust boundaries when agents run code that developers may not have written themselves.
Security Boundaries Matter for Agentic AI
Cooks describes micro VMs as a way to isolate untrusted execution environments. The goal is to keep an AI agent inside a tightly controlled runtime, rather than letting it run loose beside trusted workloads. That infrastructure-level boundary does not remove the need to review business logic or generated code. It does, however, reduce the risk that agent-driven execution can affect neighboring systems.

This matters because AI agents are designed to accomplish tasks aggressively. Guardrails at the application level still matter, but infrastructure teams also need security by default. Micro VMs provide a smaller attack surface and a stronger execution boundary, which makes them useful for AI workloads that need both agility and containment.
Serverless Economics Fit Unpredictable Workloads
Agentic AI workloads can be hard to size in advance. Cooks notes that serverless models are well suited to this type of uncertainty because organizations can spin up compute on demand and release it when the task is done. That avoids the cost of running persistent virtual machines when workload patterns are unclear.

The conversation also compares micro VMs, containers, Kubernetes and other cloud-native abstractions. Cooks says the right choice depends on the problem being solved. Some teams need simplicity and fast execution. Others need more control, tuning and persistence. The key is matching the abstraction to the workload instead of treating one model as the answer for every use case.
AI Raises the Stakes for Operations
The discussion also explores what happens when AI accelerates software change. Cooks warns that developers still own the code they check in, even when AI helped create it. Teams need testing, rollback and operational safeguards that can keep up with machine-speed changes.

For Techstrong TV viewers, the message is clear. Micro VMs are not new, but the rise of AI agents is making their value easier to see. As organizations adopt agentic AI, infrastructure choices will need to balance cost, speed, security and operational discipline.