Field Notes

Agentic AI is moving fast, in IT and in the business. The question I keep coming back to is whether you can actually run it on real infrastructure. The governance for that does not exist yet, or not completely, and honestly we do not understand these systems well enough yet either. These are the pieces where I work that problem out in the open, against a live stack, usually starting from something that broke. The lab is where the argument gets tested. This is the argument.

  1. I told the agent not to do the one thing that would have fixed my problem

    Where agents belong and where they don't, why hallucinations aren't the real failure mode, and an ESXi 9 certificate mismatch that cost five bring-ups.

  2. Agents are like genies in a bottle, be careful what you wish for

    why you can't glue deterministic systems on non deterministic agents

  3. Shared Memory Across Agents Is an Identity Problem: The Story of the Badge and the Filing Room

    What a journey to shared memory taught me about agent identity.

  4. I Put Claude Code in Charge of Hermes, and Then I Couldn't Leave the Room

    I made Claude Code be orchestrator above Hermes and it was exhausting.

  5. Nemo vs. Hermes: The Track to Get My Agent to Stop Lying to Me

    Moving from NemoClaw to Hermes, the fabrication problem that came with it, and the track to get grounding right, from SOUL.md to the harness itself.