Working with AI makes your output better right away, but does it leave you more skilled once the tool is off?
Does AI assistance improve worker learning on the job?
This explores whether working alongside AI makes people better at their jobs over time, so they keep the skill once the AI is gone, or whether it only makes their output better while the tool is in use.
This explores whether AI assistance builds lasting skill in workers or only lifts their output while the tool is switched on. On the evidence here, the two come apart. AI reliably improves what workers produce in the moment, but that gain does not reliably turn into learning. In one study, workers who used generative AI did much better on content tasks. When they later did similar tasks alone, they performed no better than before Does AI assistance help workers learn lasting skills?. One useful image for this is an exoskeleton: the AI lets you lift heavier things, but your muscles don't grow, and when you take it off you're back where you started Does AI assistance build lasting skills or temporary abilities?.
Why doesn't the learning happen? The clearest answer in the collection is about mistakes. People learning without AI ran into more errors and fixed them on their own, and they kept more of the skill afterward. People using AI handed the debugging to the AI. Even the AI users who did the most debugging with the AI's help scored lowest on later skill tests Does AI assistance remove a core learning channel through error work?. Getting stuck and working your way out seems to be where much of the learning happens, and that is the part AI is best at removing. A related finding shows that AI's productivity gains come mostly when people apply skills they already have. When workers used AI to learn something new, the gains disappeared and learning suffered When does AI actually boost worker productivity?.
This creates a tension that's easy to miss. In a study of more than 5,000 customer support agents, AI lifted productivity by about 15% on average. The biggest gains went to the least experienced agents. The most experienced agents got slightly faster but saw small drops in quality Does AI assistance help less experienced workers most?. That study measured performance, not learning. Read alongside the error-work finding, though, it raises an uncomfortable question. Newcomers benefit most on paper, but they are also the workers who most need the struggle that builds expertise. Gains that 'level the playing field' could also be skipping the steps that would have made newcomers experts. Over time, this connects to a broader warning: using AI to assist people rather than replace them is not automatically safe. Heavy reliance can slowly wear down workers' skills and their ability to judge whether the AI is right Does AI augmentation protect workers from skill erosion?.
Two social findings make this harder to fix. Recruiters give candidates who list AI skills noticeably more interview invitations, whether or not those skills are verified Do AI skills help candidates get more job interviews?. So the job market rewards looking fluent with AI, not what the worker can do on their own. Meanwhile, people who use AI expect colleagues to see them as less competent, so they tend to hide it Do people fear judgment when they use AI at work?. That keeps AI use out of view, which is where managers would need to see it to notice whether anyone is learning. There is also a quieter cost: even correct AI suggestions can break a person's concentration in the middle of reasoning Does AI assistance always help reasoning or does it carry hidden costs?. Sustained focus is part of how hard skills form.
The collection leaves one gap. It documents the problem much better than the fixes. It shows that learning depends on how AI is used, especially whether workers still meet and solve problems themselves. But it has little direct evidence on workplace designs that keep the productivity boost and still let people learn. The practical takeaway is that 'did output go up?' and 'did the worker get better?' are separate questions, and most workplace AI gains reported so far answer only the first.
Sources 9 notes
Wu et al. found that workers using generative AI performed substantially better on content tasks, but when performing similar tasks independently afterward, their performance showed no improvement. The capability did not transfer across contexts.
Research shows AI assistance creates temporary capability extensions—workers produce skilled-looking output while AI is present but revert to baseline performance when access is removed. This differs fundamentally from true skill, which persists independently.
Research shows learners without AI encountered more errors and resolved them independently, resulting in higher skill retention. AI-assisted learners delegated debugging to AI, bypassing the cognitive work that produces learning—even those who debugged most with AI scored lowest on skill assessments.
Studies showing AI productivity gains measured tasks within workers' existing domains. When workers used AI to learn new skills, productivity gains disappeared and learning suffered, suggesting prior findings do not generalize to skill acquisition.
A study of 5,172 support agents at a Fortune 500 firm found a 15% average productivity gain from AI assistance, with gains concentrated among less experienced workers who improved both speed and quality. The most experienced agents saw small speed gains but slight quality declines.
Show all 9 sources
Research mapping 8,356 workplace AI risk scenarios found that augmentation mode does not inherently prevent harm. Overreliance on AI agents can gradually erode worker skills and their capacity to provide meaningful oversight, undermining augmentation's core safety justification.
A conjoint experiment with 1,725 recruiters found AI skills significantly increased interview invitations across occupations, though certificates added only moderate gains over self-declaration, suggesting recruiters reward AI proficiency without verifying actual competence.
Across four experiments with 4,439 participants, people using AI expected others to judge them as less competent and diligent, and reported lower willingness to disclose AI use to managers and colleagues. The gap suggests a social cost that users foresee and act on.
Well-intentioned AI suggestions can damage reasoning performance by severing cognitive immersion, forcing users to rebuild focus before continuing. Evaluation must measure flow preservation across entire tasks, not just local suggestion accuracy.
Papers this line draws on 8
The research behind the notes this line reads — ranked by how closely each paper relates.
- How AI Impacts Skill Formation
- Toward Measuring AI's Effects on Skill Formation: The Stock-Formation Gap
- AI Skills Improve Job Prospects: Causal Evidence from a Hiring Experiment
- What 81,000 people told us about the economics of AI
- Does generative AI narrow education-based productivity gaps? Evidence from a randomized experiment
- Generative AI at Work
- Evidence of a social evaluation penalty for using AI
- AI Assistance Reduces Persistence and Hurts Independent Performance