The user interface you log into is the portion of your AI subscription that you interact with the most on a daily basis. The harness is that interface, but also everything underneath it. Things like projects, memory management, connectors, and skills. And that’s where I think we’re going to see a lot of the changes going forward.
A recent example: OpenAI just added a way to interact with iMessage on a Mac. You can ask ChatGPT to search your messages, catch you up on a conversation, draft a reply, even send it, all without opening the Messages app yourself. That’s a change to the harness. The model didn’t get smarter. The scaffolding around it just got another arm.
Now, Claude and ChatGPT kind of look the same. Chat box, projects, and skills have the same basic shape. But a harness that came out last week is taking a different swing at the interface. It’s a shared workspace. You can see who else is logged in, chat with them, and work through a session together where you’re co-authoring a presentation or a document or whatever, with the AI sitting right there in the thread with you instead of off in its own tab.
And I kind of like this shared workspace idea, because if I’m in a session with a coworker, I can watch how they prompt. Maybe they’re better at it than I am. I get to see that play out live, in real time, instead of just seeing their finished output. That’s kind of interesting to me. I think it’s kind of cool. It’s a genuinely different take on the interface, on the harness.
As we go forward and the models themselves, like Astra or Fable 5, get more and more powerful, I think the real differentiator between subscriptions stops being the model and starts being the interface. The harness. So I think companies are going to want to make sure their harness is the best one out there, just to keep you from wandering off to a different subscription.
Anyway, that’s what a harness is. Have a great day.