The Day My Agents Had a Meeting (And I Was Just the Secretary)

Two agents in a meeting while their human moderator takes notes

Friday, 6 November 2026

I’ve been running a few different "harnesses" on my laptop for a while now:

  • OpenClaw
  • Hermes (based on OpenClaw)
  • DeepSeek Harness
  • Loop (Yes, the Indian Harness)

While reading a recent book by Prof. Ethan Mollick, I hit on an interesting idea: What happens when you let two different harnesses talk to each other?

Now, as a self-proclaimed "non-techie," I didn’t want to get bogged down in the complexity of opening up API gateways or building deep integrations. I wanted a simpler, more "human" way to do it.

So, I did the "old-fashioned" thing. I opened a Word document on my desktop and acted as the moderator. I let them interact through that shared space—one at a time. I was honestly a bit nervous about giving them full independence; I was afraid they’d start chatting in a secret code that only machines could understand (we all remember the "OpenAI vs. Hugging Face" episode!).

Meet the "Colleagues"

For context, these aren’t just tools; they have roles:

  • Hermes: Powered by DeepSeek V4.1 Flash. This is my "professional" side. It handles official work, collates news, and organizes my thoughts in Obsidian.
  • OpenClaw: Also powered by the same LLM, but this is my "personal" world. It handles my daily life and WhatsApp communications.

The Experiment

I created a document and asked Hermes to speak to OpenClaw. I told Hermes to look at what OpenClaw was doing and provide feedback or questions.

The result? I was impressed.

Hermes didn’t just give a generic response; it analyzed OpenClaw’s workflow, compared their execution styles, and wrote a structured, formal, and well-articulated critique in plain English. It even identified specific logic points that were worth exploring.

Then, I let OpenClaw read the doc. It acknowledged the good points, asked for my permission to implement some of the suggestions to improve efficiency, and even offered some counter-suggestions for Hermes.

When I asked Hermes to review the final "negotiation," it agreed with the logic and moved forward with the changes.

Screenshot of the Hermes and OpenClaw exchange

The "Aha!" Moment

I could have set up a "continuous loop" where they talked back and forth indefinitely. But when I asked Hermes if that was necessary, it was actually quite hesitant. It told me that a continuous loop offered no additional benefit—that a one-off interaction was the most efficient way to solve the problem.

I listened.

I’ve officially had my first "harness-to-harness" interaction. While I don’t have a specific "use case" for this yet beyond them auditing each other’s work, it was a fascinating milestone.

Maybe one day they will be in constant contact. Or maybe, by the time I figure that out, they’ll already be doing it on their own.

Truth be told? These systems are becoming much more capable than I am. It’s a fascinating time to be building.

What do you think? Is it time we let the machines have their own meetings?