upvote
Excellent article, and I fully agree.

I came to the same realization a while ago and started building an agent runtime designed to ensure all (I/O) effects are capability bound and validated by policies, while also allowing the agent to modify itself.

https://github.com/smartcomputer-ai/agent-os/

reply