Surge AI has offered as much as $1,000 an hour for the judgment of startup founders and venture capitalists. Parmy Olson reported that in Bloomberg Opinion back in February, alongside her estimate that roughly 100,000 professionals now do this work.

That market got a new entrant this month.

Forbes reported on September 1 that AfterQuery reached a $3.2 billion valuation, five months after raising at $300 million. Worth holding that number loosely, since the round has not closed and no lead investor has been named. Founded eighteen months ago by two people in their early twenties. Already profitable.

The product is the reasoning, not the answer

AfterQuery pays credentialed professionals to sit down and write out how they think through a real problem. Every step. Then internal researchers validate the trace, and the company sells it as a dataset or a reinforcement learning environment to labs building models.

A lawyer with twenty years of practice carries that judgment around in their head. Sitting there, it trains nothing.

AfterQuery charges for getting it out.

A cook and a recipe are two different assets. The cook solves tonight’s dinner. The recipe lets a stranger solve it next month in another city. The labs are buying recipes, and they are overpaying because those recipes mostly do not exist in written form anywhere.

Two claims from the same six months

In January, Ethan Forrest, Anthropic’s head of product legal, spoke at Stanford Law School. He said AI now performs most tasks assigned to first- and second-year associates better than most junior lawyers do, and that the work junior attorneys do under senior direction is now a machine’s job.

Over the same stretch, the market that buys reasoning from actual practicing professionals went from roughly nothing to tens of billions.

Mercor is in talks at $20 billion and pays out over $1.5 million a day to more than 30,000 experts. Surge AI was seeking $25 billion. Meta bought 49 percent of Scale AI for $14.3 billion.

Both things hold. Models are absorbing junior work, and senior judgment has never been more expensive.

Follow the split, not the headline number

Mercor’s widely quoted $2 billion annualized figure is gross billings, counted before contractors get paid. Experts keep somewhere between 60 and 70 percent of it. Handshake shows the same shape, roughly $1 billion gross against about $300 million net. (Those percentages come from The Information via Sacra and I could not read the original, so treat them as secondhand.)

The platform is charging for access and quality control. The value sits with whoever did the reasoning, and the market is splitting the money in exactly that proportion.

That is a pricing lesson that travels well outside this sector.

The inventory you have never taken

Your company holds the same raw material AfterQuery packages.

How does your best salesperson decide which prospect gets Tuesday? Why does your operations lead turn down an order that looks fine on paper? What does your controller check before approving an exception? None of them have it written down. When they resign, it leaves with them and nobody logs the departure.

We covered the intellectual property version of this in the model trained on your judgment is your database. This post sits one step earlier. Before you ask who will own the model, check whether the trace it would train on exists at all.

And no, I am not suggesting you go sell your data. That is a different conversation with different risks, and we had it in what cannot be trained. The point here is smaller and more annoying. The asset a lab pays $250 an hour for is sitting inside your company, uninventoried, with nobody assigned to it.

In AI Maestro, that map is the first deliverable, before anyone touches a tool. Where judgment lives in your processes, who holds it, and what happens on the Tuesday that person does not show up.

Pick one process to start. The one that hurts most when the person who knows it is out.

Let’s map where judgment lives in your operation