Do people settle into one steady way of working with AI, or switch roles — director, partner, hands-off boss — depending on the task?

This explores whether a worker's way of collaborating with AI stays fixed or keeps shifting. The collection doesn't have a long-term study that follows the same people across months, so it can't settle the question directly. What it does show is that the 'mode' is mostly set by who controls a given task, not by the person or the tool. Since control changes from one task to the next, there's good reason to expect movement rather than a single stable style.

The clearest evidence comes from a study of 244 BCG consultants. It found three distinct patterns: Cyborgs, who weave AI into every step; Centaurs, who split the work cleanly between themselves and the AI; and Self-Automators, who hand whole tasks over Do humans and AI divide labor differently in knowledge work?. The key finding is what decides the mode: who picks the task and who picks the method. Neither the tool nor the task type settles it alone. These are positions on a control map rather than personality types, so the same consultant could be a Centaur on a strategy memo and a Self-Automator on a data summary. Survey data points the same way. Workers want an equal partnership in 45% of occupations, but the level they want varies task by task across 844 tasks What collaboration level do workers actually want with AI?. The tasks they have actually delegated cluster in information-heavy work, where AI capability is strongest Where have workers actually delegated tasks to AI?. The overall picture is a patchwork: delegate here, partner there, do it yourself somewhere else.

One point you might not expect is that switching between modes has hidden costs. Across four experiments with 3,562 people, those who went back to working solo after collaborating with AI felt more in control. They also felt less motivated and more bored, because the AI had taken the engaging parts of the task and left them the dull remainder Does AI collaboration drain motivation when workers return to solo tasks?. Handoffs in the other direction are tricky too. In a field experiment on Taobao, human agents stepping in after an AI failure fixed technical problems but not emotional ones, and only when they stepped in early When human workers take over from AI, does it actually fix the problem?. Switching roles is a skill with timing, not a free toggle. Something also gets lost at each switch: workers protect their voice and their authorship, while signs of effort and uncertainty quietly disappear into the AI-assisted output Which workplace cues survive AI mediation and which disappear?.

At the organizational level, the framing is moving toward permanent fluidity. In one survey, 76% of executives saw agentic AI as a coworker rather than a tool, even though it both replaces and supports human labor. That double role breaks the usual either-or way of thinking about tools and workers Does agentic AI fit the tool-or-coworker framework?. McKinsey's model puts humans 'above the loop' as strategic overseers Can AI agents truly operate above the loop at enterprise scale?. Real capability keeps pulling people back into the loop, though: agents finish only about 30% of realistic workplace tasks on their own Why do AI agents fail at workplace social interaction?, which is why researchers argue for collaboration before full autonomy Should AI systems stay collaborative rather than fully autonomous?. The practical takeaway is that workers will probably keep moving between overseer, partner, and fixer as long as AI is reliable on some tasks and unreliable on others. The more useful question is how to make those switches less costly.