Does splitting work with AI a certain way determine whether you actually get better at your job — or just coast?

This explores whether the way a worker splits a task with generative AI (who picks the work, who picks the method) predicts which skills they build, keep, or lose. The clearest answer comes from a field study of 244 BCG consultants. It sorted people into three modes by control, not by tool or task: Centaurs choose the method and hand off pieces, Cyborgs work closely alongside the AI at every step, and Self-Automators hand over both the choice of task and the choice of method Do humans and AI divide labor differently in knowledge work?. Each mode left people with different skills. Centaurs got better at their own field (consulting judgment), Cyborgs gained new skill at working with AI, and Self-Automators gained neither Do different GenAI collaboration modes produce different skill gains?. The surprise is that the most hands-off mode, the one that looks most efficient, was the only one that taught people nothing.

The lesson that carries over: skill growth follows whoever keeps making the decisions. If you keep control of the method, your expertise in the field grows. If you share it moment to moment, you learn the tool. If you give up both, nothing is left for you to learn from. Interviews with 38 Dutch knowledge workers point the same way from another angle. They found four outcomes: developing skills, keeping them, losing them, and finding that existing skills had changed in value. Which one a person got depended on how they used the AI and which of their tasks it took over How does generative AI actually change worker skills?. Read together, the studies line up. Centaurs develop skills, Cyborgs both develop new skills and see old ones change in value, and Self-Automators risk losing skills without noticing.

There is also a motivation cost that skill measures miss. Four experiments with 3,562 people found that after working with GenAI, people felt more in control when they went back to solo work, but they were less motivated and more bored. The AI had taken the interesting parts of the task and left the routine leftovers Does AI collaboration drain motivation when workers return to solo tasks?. That suggests one way the Self-Automator outcome happens: if the AI takes the parts of the work that build skill, you lose the practice and the interest along with them. A related finding is that heavy GenAI users moved toward solo writing and documentation and away from communication and coordination. Some collaborative skills may weaken just because people practice them less Does generative AI shift knowledge workers away from communication?.

Workers seem to sense this. A survey of 1,500 workers found that an equal partnership with AI was the most wanted arrangement in 45% of occupations. That is close to the Centaur and Cyborg middle ground, not full handoff. Yet many startup investments are aimed at automation that workers don't want What collaboration level do workers actually want with AI?. At Procter & Gamble, individuals using AI did as well as two-person teams without it, and their solutions drew more evenly across professional specialties Can generative AI replace the benefits of having a human teammate?. That shows AI can widen what a person reaches for, which may be what Centaur-style upskilling looks like in practice.

Where the collection runs thin: the three-mode outcomes rest mainly on one consulting study, and nothing here tracks skills over months or years. The question to ask about any AI rollout is not whether people are using it. It's which decisions they're still making, because those decisions predict what they will still know how to do.