---
title: "Comment by Karthik Karunanithi on \"Harness Engineering: The Part of Agentic AI Nobody Writes About\" | Hashnode"
description: "The autonomy boundary is the part people underestimate most. Too timid and it's useless, too loose and it's taking actions nobody signed off on. And that's not"
url: "https://hashnode.com/posts/harness-engineering-the-part-of-agentic-ai-nobody-writes-about/6a34dbbedfbba3127b16707e/comment/6a3a150eb781fd151c0e0e38"
---

[Discussion](https://hashnode.com/posts/harness-engineering-the-part-of-agentic-ai-nobody-writes-about/6a34dbbedfbba3127b16707e)

Comment [Harness Engineering: The Part of Agentic AI Nobody Writes About](https://hashnode.com/posts/harness-engineering-the-part-of-agentic-ai-nobody-writes-about/6a34dbbedfbba3127b16707e)

[Karthik Karunanithi](https://hashnode.com/@karthik006)

[Jun 23](https://hashnode.com/posts/harness-engineering-the-part-of-agentic-ai-nobody-writes-about/6a34dbbedfbba3127b16707e/comment/6a3a150eb781fd151c0e0e38)

The autonomy boundary is the part people underestimate most. Too timid and it's useless, too loose and it's taking actions nobody signed off on. And that's not a model setting, it's a judgment call someone has to deliberately encode.

The "approval gates that exist on paper but never halt execution" one hits home. Seen that exact thing, looks governed, isn't.

Thank you and Appreciate the thoughtful read. This is the conversation I wanted the post to start.
