All posts
Infrastructure
6 min read
Why every agent needs its own machine
Containers were designed for stateless services. Agents are the opposite: they accumulate context, files and credentials over days. Here’s why we built Kernal around persistent microVMs.
Stateless was a choice, not a law
The last decade of infrastructure optimised for one thing: replaceability. A container that dies is replaced by an identical one and nobody notices. That property is wonderful for web servers and terrible for agents, which spend most of their time building up state that is expensive to recreate.
An agent that has cloned a repo, installed dependencies, logged into three dashboards and half‑finished a migration is not replaceable. Killing it costs you the work, the credentials and the trust of the person who asked for it.
What a machine gives you
A persistent machine keeps the working directory, the browser profile and the shell history exactly where the agent left them. It can sleep for a week and wake in forty milliseconds with everything intact. That single property removes most of the plumbing teams currently write around their agents.
