Training a company's AI will become as much a part of the job as the work itself. When Meta began recording employees’ keystrokes and mouse clicks this spring to train its AI agents, it told staff they could help simply by doing their daily work. Other employers are likely to follow, requiring workers to train agents that can do their work when they’re off the clock or no longer with the company.
That leaves experts with an uncomfortable choice. If you were hired for your skill, keeping your value means either adapting faster than ever to stay ahead of a model that's always looking over your shoulder, or strategically keeping some of what you know out of its view. Whatever you can do that the model can’t is what makes you worth hiring and keeping.
There's an old distinction between a broker, who profits by keeping information private, and an expert, who profits from knowledge that took years to build. The broker could write it down and won't; the expert couldn't write it down if you asked. The line between them used to be obvious, but AI is wearing it thin.
When an agent can capture a person's judgment and their decision-making, experts will have to decide what to share with an organization and what to keep to themselves. Some may work as contractors, in tools they control, and hand over finished results instead of a record of every decision. They may train their own agents but decline to train a company’s, ensuring their expertise stays portable and exclusive. Others may demand a greater share of ownership in products or companies whose systems they train to replicate their expertise.
To an employer, though, an expert who trains a personal agent but not the company's looks a lot like the broker: someone charging a premium for knowledge they could share, but won’t. Companies may make training its AI a condition of employment; experts may demand more control or a greater stake in what they help build. Who sets the terms depends on who can afford to walk away.
