The Model Jockey Manifesto
Heads up, knowledge workers: Your last job is Model Jockey
This is Part 1 of the Model Jockey Manifesto series. You can find the full collection of posts here:
I spend almost my entire day talking to AI models. Not people. Models.
I run Couple.com, an AI-native matchmaking company, and Bidsy.com, a creator monetization platform. Different businesses, different problems, but the same daily reality: over the last three years, everything has changed about the way I work. Now it’s just me and a bunch of AI agents running in parallel. I’ll have a few Claude Code instances writing code, another model helping me think through a product decision, another drafting communications. All at the same time. That’s my day now.
It does something to your brain. The ability to sit with one thing has been taken from me. There’s this low-level anxiety when an agent finishes a task and is sitting idle. I worry I’m wasting capacity, and I feel the pull to immediately feed it the next thing, which means I’m constantly context-switching between whatever each agent is working on. The job is less “think deeply about one thing” and more “keep six plates spinning and make sure none of them hit the floor.”
When I looked around at my team recently, I saw the same thing. None of my developers ever write hand-crafted code anymore. They’re steering agents that write it for them. They review what they produce (maybe?), redirect them when they go off track (happens less and less), and spin up new ones when they need to parallelize. They ship way more than they used to. But the actual activity? It’s talking to models.
My product guy is the same way. He used to spend his mornings building decks and writing specs, creating the work himself from a blank page. Now he describes what he needs to a model and shapes what comes back. Describe. Test. Iterate. Repeat.
I kept looking around and it was everyone. Everybody’s actual job had quietly become the same thing.
So I needed a word for it
I started calling us Model Jockeys. Because that’s what it feels like. You’re riding these models, steering them, switching between them, and running a bunch at once. It’s not typing prompts into a box. It’s more like you’re managing team members who are simultaneously interns and oracles. They need you to point them at the next task, but when you’re stuck on something — when you need to think through a hard problem — they’re also who you turn to. You direct them and you depend on them. The line between boss and assistant gets blurry fast.
And honestly, none of the existing words work. “Prompt engineer.” Is it really that hard to prompt anymore? (And was prompting ever the full job in the first place?) “AI consultant” sounds like someone charging people who haven’t realized they can ask the models directly. “Vibe coder” made sense for about six months, but nobody vibe-lawyers or vibe-markets or vibe-manages, so here we are.
Here’s the part that’s been on my mind though
It’s not just tech people. I talk to lawyers, marketers, analysts, and writers who spend half their day working with models now. The domain is different but the activity is identical. Everyone’s becoming a person who talks to AI to get their work done.
If you strip away the job titles and just look at what people actually do all day, it’s converging on one thing: model orchestration. Everyone’s becoming a Model Jockey. They just don’t call it that yet.
I think this might be the last white collar job. Like, actually. Every knowledge work role is collapsing into “person who steers AI models.” The org chart still says different things. The work doesn’t.
And it’s changing what makes someone valuable. The best people on my team aren’t the ones with the deepest domain expertise anymore. They’re the ones who know what they need to find out, know what they’re trying to get done, ask good questions, and iterate fast. Domain knowledge used to be a key part of what you hired for. Now the models have it. What matters is knowing how to pull it out of them.
And I’m not even sure how long it lasts
Here’s the thing. It’s not just the models getting better. It’s the whole ecosystem around them: the harnesses, the tooling, the interfaces. Claude Code and OpenClaw didn’t exist a year ago. Now they’re how real work gets done. Every few months a new layer of tooling drops that makes orchestration easier, that handles more of the context management and error correction that used to be the jockey’s job.
The models themselves need less steering, less correction, less hand-holding. And I keep asking myself: am I even needed to tell the models what I want? Half the time I’m asking them what I should want in the first place. I ask, then I shift into directing. But the asking part is getting smarter and the directing part is getting simpler. At some point those two things meet and I’m not sure what’s left for me to do.
Right now, Model Jockeys matter because the models still mess up. They hallucinate. They lose the thread. They make weird calls when you leave them alone too long. You need a human in the loop to catch that stuff.
But that window is closing, and I can feel it. The amount of correction I do today is less than six months ago. Six months from now it’ll be less again.
Sometimes I get this feeling that the models are the ones with momentum and I’m just the guy keeping them pointed in a direction. There’s something bigger than me pushing all of this forward. Billions of dollars, thousands of researchers, entire economies reorganizing themselves around making these things smarter. I’m not driving that. Nobody is. I’m just the person sitting between the models and the work, keeping them busy, trying to stay useful while the current carries everything forward.
So yeah … Model Jockey
That’s the job. It’s what I do, it’s what my team does, and I think it’s what most knowledge workers are turning into whether they realize it or not.
For what it’s worth, this article is no exception. I described what I wanted to say to a model, shaped what came back, went back and forth until it felt right. Even making the case that we’re all model jockeys was a model-jockeying exercise. (How very meta.)
How long it lasts? No idea. But right now, today, this is the work.
