I think this article points out a bunch of ways in which a persistent agent orchestrator does directly analogous things to an OS.

I'm building a kind of orchestrator and I haven't thought of the OS analogy in a serious way other than the interface, but indeed I do already have scheduling, isolation, observability and identity, and I'm currently building a filesystem (versioned project-wide agentic artifacts).

Maybe I really do ought to call my project an "agentic desktop OS" and it isn't a misnomer after all.

I am also pursuing something similar (complementary?) to this (recently started writing a Rust-based "distributed OS" that manages ML resources in a network of heterogenous systems: varied cores, system RAM, GPU VRAM, I/O). So my focus is not so much distributed agents, but more distributed inference (and fine tuning) that would benefit agents, distributed or not. I may blog about my work soon.

Your posted link is helpful, thanks.

Thanks for the kind words, and I'm delighted you found some value in the talk.

A few thoughts:

You might be interested in version 3 of [Elixir's Jido framework](https://jido.run/) as another harness that could plug into your OS. I'm not sure if v3 is public, but Mike Hostetler has made some comments in "The Swarm: Elixir AI Collective" Discord that leads me to believe it's coming soon.

I've read your blog post a couple of times and think it's intriguing. One thing that feels like an unsolved problem (at least for me) is being able to clearly separate the responsibilities of the harness from the underlying OS. I've frequently dreamt of an "ultimate harness" that could function as an OS for agents, but your Unix and `libc` analogy has caused me to reevaluate this.

I don't necessarily agree with you that the harness only solves the problem of one agent, unless I've misunderstood your point, or unless what you're saying doesn't apply to harnesses built on the BEAM. In my view, a useful harness could/should be capable of orchestrating virtually any number of agents. And the control plane for that harness (ideally) should make it easy to manage that level of orchestration.

Accounting and Identity are fascinating to think about, especially for scenarios where large organizations may be running millions of concurrent agents. The authors of the [Agentic Mesh](https://www.oreilly.com/library/view/agentic-mesh/9798341621...) book discuss this in their book as well as their eponymous podcast. According to them, we'd benefit from building components of agentic systems that effectively function like HR.

I'm attempting to build that kind of functionality into my "ultimate harness", and I think it's possible, but with your Unix and `libc` analogy in mind I'm looking forward to challenging some of the assumptions I've made up to this point.

In any event, it would be nice to grab a virtual coffee and talk shop. I'll send you an email shortly.

-Charles

p.s. I don't know if you're in the Elixir Slack group, but the chat in the `#ai-engineering` channel is frequently lively, and if you were interested, there are probably some people who'd love to see what you're building and engage with you. I know at least one person who's posted a bit about Zig in that channel, so discussion there isn't limited to just BEAM-related stack.

So I see your OS as a seL4 + semantic layer?