---
title: "Comment by Kartik N V J K on \"A Crew of AI Agents Runs My Homelab\" | Hashnode"
description: "The line that the harness is where the real engineering lives, not the model choice, is something I keep arguing in reviews. Per-hour action caps plus an orches"
url: "https://hashnode.com/posts/a-crew-of-ai-agents-runs-my-homelab/6ac2ceba38aa7c69feb13799/comment/6ac40d80f45d343ee565e39c"
---

[Discussion](https://hashnode.com/posts/a-crew-of-ai-agents-runs-my-homelab/6ac2ceba38aa7c69feb13799)

Comment [A Crew of AI Agents Runs My Homelab](https://hashnode.com/posts/a-crew-of-ai-agents-runs-my-homelab/6ac2ceba38aa7c69feb13799)

[Kartik N V J K](https://hashnode.com/@kartik-nvj)

[Oct 5](https://hashnode.com/posts/a-crew-of-ai-agents-runs-my-homelab/6ac2ceba38aa7c69feb13799/comment/6ac40d80f45d343ee565e39c)

The line that the harness is where the real engineering lives, not the model choice, is something I keep arguing in reviews. Per-hour action caps plus an orchestrator routing to read-only versus action specialists is a far saner default than one all-powerful agent with a good model behind it. The hard part is verifying the "careful colleague that says when it doesn't know", how do you test that EDITH abstains instead of guessing?
