Up to $200,000. That was the number in the email Adam Liska got one morning in early August, from an AI data company that wanted to license Airspeed’s codebase to train the next generation of models. The exact wording was “up to $200k+, depending on volume and complexity.”

Liska is co-founder and CEO of Airspeed, the AI revenue execution platform formerly known as Glyphic, which raised a $20M Series A in June 2026. He passed.

The refusal is not the notable part. The reasoning is.

Your operating history is the asset. The repo is just the receipt.

Liska draws a line that most people never draw. Here is how he put it in his LinkedIn post: “Code tells you what a company intended its systems to do. The operating trace tells you what actually happened.”

Read that again with your own company in mind. Everything in your repository is the output of a long argument nobody wrote down. Someone proposed an approach. Someone else killed it in review. A customer complaint forced a rewrite three sprints later. A stage moved in the CRM because of a specific call, someone acted on it, and the deal either closed or died. The repo holds the final answer. It holds none of the working.

Liska makes the point sharper by naming what public repositories are missing. GitHub is full of implementation. What it almost never contains is a complete trajectory with the outcome attached at the end. A stage change in a CRM, the conversation that triggered it, the action somebody took next, and whether the deal closed are worth far more together than any one of them alone. That chain is what a model needs to learn judgment rather than syntax, and it only exists inside operating companies.

That distinction is the thesis I have been writing and selling for a while now, stated by someone outside our company, with a price tag attached to it for the first time.

At IQ Source we call that record the Process Reality Map, and it is the central deliverable of AI Maestro. Two months of consulting, education and training whose entire job is to document how the operation actually runs today. Not the version in the process manual. The version where the finance lead approves things over WhatsApp on Fridays and one senior engineer is the only person who knows why the retry logic exists. That map gets scored into an AI Opportunity Score, and the program ends at a Go or No-Go gate before anyone builds a thing.

It is unglamorous work. It is also the only work that turns a history scattered across email threads, tickets, calls and commits into something you can read, govern and reuse.

And notice the direction of the two options in front of Liska. A lab pays you once for a photograph of that record. A system you control keeps learning from the record every quarter, while the record keeps growing.

This is a priced market now, not a thought experiment

What makes the email more than an anecdote is that buying operating history has moved fully into the open.

Turing’s Project Lazarus pitches it with no ambiguity at all: “Your operating history is a training asset.” One payment of up to $1M, no retainer, no recurring commitment. The categories they want listed on the page are software and engineering records, finance and accounting, support tickets, sales and CRM pipeline, legal and contracts, operations and logistics. That is not a code purchase. That is a purchase of how your company thinks.

Mercor runs the same play at the other end, licensing enterprise workflow data to frontier labs on the stated premise that no enterprise-grade model gets built from public data alone.

The reason this demand appeared right now is technical. Previous-generation models trained well on text: docs, forums, repositories. Models that execute multi-step work need something else, full action sequences with an outcome attached, so they can be trained against a real reward signal. Forbes reported in April 2026 that Anthropic has considered spending as much as $1 billion on reinforcement learning environments. That is the gap the $200,000 email is trying to fill.

That same Forbes piece covers the grim end of the market: dead startups selling their old Slack and email archives. SimpleClosure closed nearly 100 of those deals in a year, typically $10,000 to $100,000 per company, over $1 million recovered on behalf of founders whose businesses no longer exist.

And at the expensive end, the loudest signal of all: SpaceX agreed to acquire Anysphere, the company behind Cursor, for $60 billion in stock. That deal is about far more than data, and Liska says so himself. But it points at the same scarcity: a system used daily by real experts accumulates a private record of how the work gets done, and that record cannot be downloaded from anywhere.

All of it is one movement. The model layer got cheap and interchangeable, which is the argument I made when agent runtime became a commodity and the moat moved to the workflow. What is getting priced now is the layer underneath the workflow: the trace of how your company executes.

Thirty-six years of watching the code get cheaper

I started programming in 1990, at fifteen, on a Commodore 64 and a Texas Instruments machine. Since then I have watched the cost of producing code collapse repeatedly. Commercial libraries. Open source. The browser eating the thick client. Cloud removing the server room. Now models that write whole functions in one pass.

In none of those cycles was typing the bottleneck. The bottleneck was always knowing which thing was worth building, at what tradeoff, and what would break in the operation once you shipped it. That part never came out of a library, and it does not come out of a model today either. I made the same case when building got cheap and deciding what to build did not.

Which is why a $200,000 bid on a repository reads cheap to me from the buyer’s side, not generous from the seller’s. Nobody is paying for lines. They are paying for the chance to reverse-engineer, from a finished artifact, a decision process that nobody ever wrote down.

Three things the $200,000 email leaves out

First, Liska is selling something too. Airspeed’s product is precisely the layer he describes, agents acting across calls, emails, tickets and CRM. His commercial interest lining up with his argument does not make the argument wrong, but you should read it knowing that. I reached the same conclusion from a different direction, and I also have something to sell.

Second, selling that history carries a cost most companies never price. The Forbes reporting quotes Marc Rotenberg of the Center for AI and Digital Policy: “I think the privacy issues here are quite substantial.” The internal messages being sold were written by employees who never agreed to it, anonymization fails more often than vendors admit, and buried in those threads are trade secrets, customer data and identifiable people. A $200,000 license can end up costing considerably more than $200,000.

Third, and this is the one I run into constantly: most companies could not sell their operating history even if they wanted to, because it does not exist in legible form. It lives in three people’s heads, in email threads nobody will ever open again, and in a project board last updated eight months ago. A process that only exists in the memory of whoever performs it is not an asset. It is turnover risk wearing a disguise.

That is where our work starts. At IQ Source we do not open with which AI tool you want installed. We open by building the record you should already have and do not: what your operation does today, in what order, who decides what, where work stops, and what happens when someone skips the process entirely. That is the Process Reality Map. With it on the table, the AI Opportunity Score stops being an opinion and the decision to automate stops being a bet.

Liska is right about the hard part. A competitor can copy your playbook and cannot copy the history that shaped it. What his post does not address is the step before that, which is the one almost everybody is missing. Before you can decide whether to sell your operating history, you have to be able to see it.

Map your company’s operating history before someone offers to buy it