kumayama

Author here. This is a manifesto we published after working on an alternative cloud architecture for a few years.

The core idea: AI-generated code will only be useful if we can run it safely and deterministically. Today’s infra wasn’t built for this. So we built something new – and wrote this to explain why.

It’s not product marketing – just a proposal for how the future of AI-native systems might need to look.

Curious to hear your thoughts – especially from those who’ve worked with AI agents, cloud security, or autonomous deployment flows.

Just checked out the site, this feels surprisingly refreshing.

Most platforms just wrap Kubernetes or slap on another abstraction layer. Heim looks like it’s actually trying to remove layers instead of adding them.

A few quick questions though: – Can you run long-lived or stateful services (e.g. websockets, streams)? – How do you handle secrets or database connections? – Any way to run Heim locally for testing?

Definitely curious to try it out.