Handing a whole task to AI doesn't just save time — it changes which human skills actually pay off.

This looks at what happens to the value of different worker skills once people start handing whole tasks to AI: which skills become more important, which fade, and which ones only seem to grow. The short answer from the corpus is that delegation doesn't just remove tasks. It makes judgment about the AI's work more valuable, and it can wear away the hands-on practice that judgment depends on.

Start with where delegation is actually happening. It clusters in information-heavy jobs and follows what the technology can do, not the old idea that routine work gets automated first. In those jobs the wage patterns even flip at the advanced-degree level Where have workers actually delegated tasks to AI?. When AI touches only a few tasks in a job, workers can move their time to the tasks it doesn't touch, and job losses are mostly offset Does concentrated AI exposure enable workers to adapt and reallocate?. So the first skill that shifts in value is knowing your own job well enough to see which parts AI can take and where your time should go instead. The economic incentives work against this, though. Acemoglu, Autor and Johnson argue that firms earn more from AI that replaces expertise than from AI that creates new work for people, so the tools that would make worker skills more valuable get less investment Why do firms build automating AI instead of pro-worker AI?.

The less obvious finding is that AI rewards skill you already have and doesn't help you build new skill. Productivity gains show up when people use AI inside a field they already know. When they use it to learn something new, the gains disappear and their learning suffers When does AI actually boost worker productivity?. Anthropic's own engineers describe the other side of this. They report about 50% productivity gains, yet most can fully hand off only 0–20% of their work. They also worry that letting Claude do routine coding wears down the hands-on practice they need to catch its mistakes Does AI assistance erode the skills needed to oversee it?. The skill that matters most, checking the AI's work, is kept up by the same routine practice that delegation takes away.

The effects on skills aren't all one direction. Interviews with Dutch knowledge workers found four outcomes rather than a simple upskilling-or-deskilling split. Some skills develop, some are maintained, some erode, and some change in value, depending on how each person uses the tool How does generative AI actually change worker skills?. Self-reports are hard to trust here. Anthropic found that its heaviest delegators are the most optimistic about their careers and say their skills are gaining value, but that is a correlation within its own user base, not an independent test of their skills Does delegating work to AI actually damage worker skills?. Separate work describes an 'LLM Fallacy': people credit AI output to their own ability, even when the output is accurate and even when they are checking it How does AI-assisted work reshape how people see their own abilities?. So feeling more skilled and being more skilled can drift apart.

One more skill emerges: deciding where to keep a human approval step. A small study found that people stopped trusting an AI agent when its actions were irreversible and visible to others, like sending an email. High stakes alone didn't have that effect if the result could be corrected What makes people distrust AI agents they delegate to?. That points to a practical skill: sorting your tasks by whether they can be undone, not by how important they seem. A caveat: most of this evidence is surveys, interviews and small studies. The corpus doesn't yet have long-term measurements of how workers' skills actually change.