Hands-On Mercor Playbook

What four months and $37,000 taught me about thriving on AI's strangest new platform

A few weeks ago, I published a breakdown of how I made $37,000 on Mercor over four months. The response blew up. My inbox flooded. People wanted more—not just the “what,” but the “how.” How do you actually survive and thrive on a platform where projects vanish overnight, the rules change mid-game, and your “employer” is technically a language model that hasn’t learned to think yet?

So here it is. The playbook I wish someone had handed me on day one.

But first, a mindset shift. Because if you walk into Mercor thinking like a traditional freelancer—or worse, like a salaried employee—you’re going to have a terrible time.

The AI-Era Uber

Let me give you the mental model that changed everything for me.

Mercor is Uber for AI training. Structurally.

Think about what Uber did. It sat between people who needed rides and people who could drive. It didn’t employ the drivers. It didn’t own the cars. It built the marketplace, handled the matching, and took a cut. The drivers showed up, completed trips, got rated, and either kept getting rides or didn’t.

Mercor does the same thing, except the “riders” are frontier AI labs—OpenAI, Google DeepMind, Anthropic, Meta, and others building the most powerful models on the planet—and the “drivers” are domain experts like you and me. The labs need human-generated data to train and evaluate their models. They need people who can spot where GPT-5 or Gemini or Claude falls apart, write better responses, and judge which outputs are actually correct. Mercor finds those people, organizes them into project teams, and manages the whole operation.

Once you internalize this, a lot of things that frustrate people suddenly make sense.

Projects ending abruptly? That’s normal. The AI lab finished that phase of training, or their evaluation goals shifted, or the model got good enough at that particular task that they no longer need human data for it. It’s not personal. It’s not a performance issue. It’s a ride ending because the passenger reached their destination.

I see people in community groups losing their minds every time a project gets terminated. “I just got off-boarded!” “They cut us without warning!” I understand the anxiety. Losing income is stressful. But panicking about project termination on Mercor is like an Uber driver panicking every time a passenger gets out of the car. That’s… how it works.

The better analogy might be commercial fishing. You go where the fish are. Some seasons are abundant. Some aren’t. You maintain your gear, keep your skills sharp, and stay ready for the next run. The fisherman who prospers isn’t the one who catches one big haul and retires. It’s the one who shows up consistently, reads the water, and adapts.

So rule number one of the Mercor playbook: get comfortable with impermanence. Don’t over-index on any single project. Don’t mentally spend money you haven’t earned yet. And don’t mistake a project ending for a career ending.

What You're Actually Getting Paid For

Here’s something that trips up a lot of new Mercor contributors, especially people coming from traditional tech jobs: you’re not getting paid to be smart. You’re getting paid to find where the model is stupid.

That distinction matters enormously. The AI labs aren’t hiring domain experts because they want you to demonstrate your knowledge. They’re hiring you because their model needs to get better, and the only way it gets better is if humans identify the gaps and fill them with high-quality data.

Your job, at its core, is quality control and gap detection. You’re looking at model outputs and asking: Where did this go wrong? What did it miss? What does it think it knows but actually doesn’t? And then you’re writing responses that show the model—or, more precisely, the training pipeline—what a correct, complete, expert-level answer actually looks like.

This requires a specific kind of thinking. You need to resist the urge to just “write a good answer.” Instead, you need to interrogate the model’s attempt first. Find the failure modes. Spot the hallucinations, the subtle logical errors, the places where the model sounds confident but is actually making things up. Then, and only then, craft the expert response that addresses those exact weaknesses.

The people who do this well become invaluable. The people who just bang out competent answers without really engaging with why the model failed get average ratings and wonder why they’re not getting staffed on the good projects.

The SPL: Your Most Important Relationship

Every Mercor project has an SPL—a Special Project Lead. If Mercor is Uber, the SPL is the dispatcher who also happens to be your local manager, client liaison, and quality gatekeeper rolled into one.

The SPL sits between you and the AI lab client. They translate the client’s requirements into actionable guidelines for the writing team. They communicate quality expectations. They handle escalations. They’re the reason you don’t have to deal directly with a Google DeepMind product manager who has very specific ideas about what “expert-level reasoning” looks like in organic chemistry responses.

Here’s what I learned about working with SPLs: treat this relationship like you’d treat a relationship with the best manager you’ve ever had. Communicate proactively. Ask smart questions early. Don’t wait until you’ve submitted 50 tasks to find out you’ve been interpreting the guidelines wrong. If something in the project playbook is ambiguous—and something always is—raise it immediately.

The SPLs I’ve worked with have been sharp, responsive, and genuinely invested in their contributors’ success. They want you to do well because your quality scores reflect on their project. This alignment of incentives is a gift. Use it.

But here’s the flip side: SPLs are also the people who notice when you’re not meeting requirements. They track your quality scores, your throughput, and your adherence to guidelines. If you’re consistently missing the mark, they’re the ones who’ll have to bench you. So the relationship cuts both ways. Earn their trust early, maintain it consistently, and you’ll find yourself getting first pick of tasks and early access to new projects.

Respect the AHT (Or Pay the Price)

Let’s talk about the metric that governs your life on Mercor: AHT, or Average Handling Time.