People everywhere are handing a little of their thinking to AI. Researchers call this cognitive offloading. The 2026 research shows the question is how people offload, as much as how much.
Two kinds of offloading
In July 2026, Zhu and colleagues published a study of 589 students and early-career knowledge workers in Frontiers in Psychology. They separated dependent offloading from autonomous offloading.
Dependent offloading means letting AI do your core thinking. Autonomous offloading means using AI as a scaffold while you stay in charge of the reasoning.
Dependent offloading was linked to poorer judgement, creativity and deep processing over time. Autonomous offloading was linked to better outcomes.
Both felt equally helpful in the moment. That makes harmful habits hard to spot from experience alone. Checking your own thinking weakened part of the slide into dependence.
Confidence changes how hard people think
In 2025, Microsoft Research and Carnegie Mellon surveyed 319 knowledge workers about 936 real uses of AI. The more people trusted the AI, the less critical thinking they reported.
The more they trusted their own judgement, the more critical thinking they applied. Trust in the tool and trust in yourself pull in opposite directions.
Build this capability in your organisation
Growth Performance designs bespoke programmes that turn insight into measurable behaviour change.
Deskilling is a design problem
Ferdman argues in AI & Society that AI deskilling is a structural problem. Skills grow through long practice and learning from others. AI can quietly remove both.
She calls these "capacity-hostile environments". The answer lies in how work is designed, as well as in individual effort.
Microsoft's 2026 Work Trend Index shows what people need now. Quality control and critical thinking were the top skills AI users said they needed.
From producing to checking
AI changes the nature of thinking work. People shift from writing answers to checking them, and from finding information to joining it up.
These are real skills. Most organisations have not yet helped people build them, or given them time to use them.
What this means for leaders
Protecting judgement is part of Adaptation. People must keep directing and checking the AI, as well as using it. Three practices help.
Think first, then prompt. Ask people to write a short view before they open the AI. This keeps their reasoning independent.
Make checking visible. Ask what was checked and how. Thank people who challenge a confident output.
Keep some work unaided. Choose a few tasks where people practise without AI, so core skills stay sharp.
Try this
Take one recent AI-assisted decision to your next team meeting. Ask what would have to be true for it to be wrong, and whether anyone checked.
Make that question routine. Then use the Checking Capacity Calculator to see whether your team has the hours to check properly.