This is a Talmyn Debate submission — a published argument on a genuinely contested question, presented in the author’s own words. Publication doesn’t mean Talmyn endorses the position; it means the argument was strong enough to publish.
My position: if a company trains an internal AI system on an employee’s work product — their code, their writing, their sales calls, their design files — with the explicit goal of eventually reducing how many people are needed to do that job, that’s a materially different use of the employee’s labor than the job they were hired and paid to do, and current compensation structures don’t account for it.
The argument
When someone is hired to write code, the assumption embedded in their salary is that they’re producing working software. It is not, typically, an assumption that they’re also producing training data for a system explicitly designed to eventually need fewer people like them. Those are two different economic activities, even if they happen to look identical from the outside — the same commit, the same pull request. One is doing the job. The other is helping build the job’s replacement. An employee who understood they were doing the second thing might reasonably want different terms.
The usual counterargument is that this is just normal technological progress — every tool that makes workers more productive has always, eventually, reduced headcount for a given unit of output, and nobody pays a machinist extra because a CNC lathe was designed by studying how machinists work. That’s a real point, and it’s mostly right for tools that make an existing worker faster. It’s weaker for a system whose specific training data is that worker’s own individual output, collected without a separate conversation about what it would be used for. The difference isn’t the automation — it’s the lack of disclosure and negotiation before the data collection started.
A reasonable middle position, and the one I’d actually argue for, isn’t that companies can never do this. It’s that using an employee’s specific work product to train a system aimed at reducing their own role should require the same kind of upfront disclosure as any other material change to a job’s terms — not necessarily extra pay in every case, but at minimum, employees should know it’s happening before it happens, not find out when the tool ships.
