Uploaded June 2026 | Updated September 2026, 2 weeks ago
🎙️ Future of Data and AI Podcast: Highlight with Mark Cavage (COO, Docker)
Containers vs Sandboxes — which one does AI actually need?
Mark's answer: both. But for very different reasons.
Containers are not going anywhere. 92% of the world runs them in production and that number is only going up. They are the standard unit for packaging, shipping, and sharing software — and that will be true whether you are building for humans or agents.
But agents have a problem containers were never designed to solve. Containers are built around immutability. Agents want to mutate everything about their environment all the time. The moment you run an agent, it wants to install packages, call APIs, rewrite files. That breaks the entire social contract containers were built on.
That is where sandboxing comes in. Not as a replacement — but as a different primitive for the runtime execution layer. The part that thinks. The part that acts. The part that needs to be able to mutate freely without taking everything else down with it.
Containers for packaging. Sandboxes for execution. That is how Mark sees Docker evolving for the agentic era.
Mark breaks this down with host Raja Iqbal on the Future of Data and AI podcast.
🎧 Watch the full episode: youtu.be/eiKm99VIcSs?si=86I3RlDWlEzv05B0
🔗 View all podcasts: datasciencedojo.com/podcast
#AgenticAI #Docker #Containers #AIInfrastructure #FutureofDataAndAI
🎙️ Future of Data and AI Podcast: Highlight with Mark Cavage (COO, Docker)
Containers vs Sandboxes — which one does AI actually need?
Mark's answer: both. But for very different reasons.
Containers are not going anywhere. 92% of the world runs them in production and that number is only going up. They are the standard unit for packaging, shipping, and sharing software — and that will be true whether you are building for humans or agents.
But agents have a problem containers were never designed to solve. Containers are built around immutability. Agents want to mutate everything about their environment all the time. The moment you run an agent, it wants to install packages, call APIs, rewrite files. That breaks the entire social contract containers were built on.
That is where sandboxing comes in. Not as a replacement — but as a different primitive for the runtime execution layer. The part that thinks. The part that acts. The part that needs to be able to mutate freely without taking everything else down with it.
Containers for packaging. Sandboxes for execution. That is how Mark sees Docker evolving for the agentic era.
Mark breaks this down with host Raja Iqbal on the Future of Data and AI podcast.
🎧 Watch the full episode: youtu.be/eiKm99VIcSs?si=86I3RlDWlEzv05B0
🔗 View all podcasts: datasciencedojo.com/podcast
#AgenticAI #Docker #Containers #AIInfrastructure #FutureofDataAndAI
