A new Nature study confirms what we suspect: AI makes our work better but leaves us feeling worse.
Researchers tested 3,562 people across four experiments. Those who worked with ChatGPT produced better results: longer texts, sharper analysis, more engaging posts. The performance boost was real.
But when these same people moved to work alone, their motivation dropped. They felt bored. The work seemed dull compared to working with AI.
The Trade-Off
Here's what happened. Participants using ChatGPT consistently outperformed those working alone. The AI-assisted work was more comprehensive and polished.
Then came the second task, completed without AI. The results were mixed. In most studies, performance gains didn't carry over—people weren't better at solo work afterward. In one study, people did produce more creative ideas after AI collaboration, but no more ideas overall.
More importantly, something had shifted. People felt less motivated and more bored after switching from AI-assisted to solo work. But they also felt more in control, suggesting AI had initially reduced their sense of autonomy.
Researchers called this a "psychological deprivation effect." People had tasted AI collaboration. Returning to solo work felt like a step backward.
Why Order Matters
The order of collaboration makes a difference.
Skills don't transfer. AI collaboration improved immediate output but rarely made people better at subsequent solo work. One exception: people showed more creativity (though not more ideas) in brainstorming after AI collaboration.
Control follows a pattern. When people moved from AI-assisted to solo work, they felt more autonomous. When they went from solo work to AI collaboration, their sense of control dropped.
Motivation consistently declines. People felt less motivated across all conditions. The decline was steepest for those moving from AI collaboration back to solo work.
Transitions disrupt. The fourth study tested all combinations: solo-solo, AI-AI, solo-AI, and AI-solo. Transitions in either direction were more disruptive than staying in one mode.
What Job Are You Really Doing?
Different goals require different approaches.
If your job is "produce better content today," AI collaboration works. Output quality improved in every study.
If your job is "develop capable workers," AI collaboration mostly fails. People didn't become more skilled at independent work, and their motivation declined.
If your job is "maintain long-term team performance," consider both the performance benefits and the psychological costs of different sequences.
Sustained AI collaboration maintains stable experiences. Transitions create more disruption.
Four Questions Before You Start
What outcome do you want? Better output today, skilled workers tomorrow, or sustained engagement? Your answer should shape how you use AI.
How will you sequence the work? Moving from AI-assisted to independent work affects people differently than the reverse. Plan these transitions.
How will you preserve control? AI collaboration reduces people's sense of autonomy. How will you maintain human agency?
How will you sustain motivation? Motivation declines across all conditions but drops most sharply when people transition from AI collaboration. How will you keep people engaged?
Beyond the Quick Win
The study's most important finding may be about control. People working with AI felt less autonomous, even when their performance improved.
This doesn't mean avoiding AI. It means preserving human agency within AI collaboration. The best implementations will enhance human capability, maintain control, and create intentional transitions between different work modes.
The goal isn't just better work today. It's systems that make people and organizations more capable over time while keeping workers engaged and in control.
That starts with knowing what outcome you actually want.