Forget who's winning the harness wars. Pick one and start.

Hermes had a loud week and the timelines lit up. For a CEO, the winner of that fight matters far less than the fact that you still have not picked a side.

This week my X feed filled up with two words: "HERMES WON." Nous Research had just shipped a stronger memory system for its Hermes agent, operators started posting their Hermes setups for killing AI slop (the generic, soulless output you get from AI by default), and the whole harness crowd had a moment.

Here is the thing. I run OpenCLAW, not Hermes, and I did not feel a flicker of worry watching it. The question the timeline was fighting about is not the question that matters for your business.

The honest answer

The model is rented. The harness is the moat.

A harness is the layer that turns a chat model into something that runs parts of your company: memory, connectors, skills, and routines wired around the model so it can act, not just answer. The model underneath is close to a commodity, swappable in an afternoon. The harness, and everything you build on top of it, is the part that compounds.

So when the timeline crowns a winner this week, it is arguing about the packaging of the replaceable layer. The part that actually creates leverage, the workflows and context you build, is portable across all of them. Whoever is ahead today, the move is the same: get on a harness and start building. The building is where your year goes, not the choosing.

Concretely: the workflows I lean on most, a Monday scan that surfaces where every deal stands, my meetings flowing into a folder I can ask questions of, team and investor updates that draft themselves, are written as portable skills. If I woke up tomorrow convinced Hermes was the better bet, I would move them over in a weekend and lose nothing. That is the real test of whether you have built leverage or just chased a tool: can it come with you?

What's actually happening

Strip out the scoreboard noise and three real things happened. Hermes got a stronger memory system, which matters because memory is one of the handful of things a harness has to nail. Operators started using their harnesses specifically to strip slop out of their output, which is a sign these tools are maturing from toys into things people run real work through. And people are openly comparing what their harness does for them day to day, which is the conversation you want to be inside, not watching from the sidelines.

The signal under all of it: the action moved. A year ago the argument was which model is smartest. Now the operators getting real leverage have stopped arguing about the model and started arguing about the harness and what they have built on it. That shift is the whole story. The model race is loud. The harness is where your leverage now lives.

So which one should you run?

This is the one question worth a few minutes, and I wrote the full answer separately so this post does not turn into a spec sheet. The short version: it comes down to whether you want it living on your laptop or your phone, how much off-the-shelf ecosystem you need, and whether a cost edge on heavy usage matters to you. OpenCLAW vs Hermes walks the actual decision in about five minutes.

Do not let the comparison become a stall. Both are open-source, both solve the same problem, and the skills and workflows you build move with you if you ever switch. The cost of picking the "wrong" one is low. The cost of not picking is another quarter spent at a tenth of the leverage you could have.

Want to maximize your AI leverage? Upgrade to Pro.

What to do next

Pick a harness this weekend and wire up one workflow on it. Do not wait for the timeline to crown a winner. The harness you start on this month beats the perfect one you choose in six. Tell me which one you picked and what you ran first. I am genuinely curious what earns its keep in your opening week.

Related

• OpenCLAW vs Hermes: which harness should you run?
• What is a harness?
• What is OpenCLAW?
• What is Hermes?
• What is an AI agent?