When employers adopt AI chatbots, many workers take on new tasks within two years, yet pay and hours stay flat.

This explores what new work appears when an employer adopts AI chatbots, as opposed to which old work disappears. The clearest evidence in the collection comes from Danish administrative records. Employers adopted chatbots widely, and within two years of ChatGPT's launch many workers had taken on new AI-related tasks. Yet earnings and hours stayed flat within a 2% margin Does AI chatbot adoption change worker pay and hours?. So the first thing AI changes is what the job is made of, before it shows up in the paycheck. One caveat: the collection records that new tasks appeared but doesn't list them in detail. The rest of this answer works out what those tasks probably are from neighboring studies.

The first place to look is where the new work lands. Workers who actually hand tasks off to AI through set workflows cluster in information-heavy jobs. That pattern follows what the technology can do, not how many people chat with it casually, and it doesn't match older predictions that automation would hit routine jobs first Where have workers actually delegated tasks to AI?. A large chunk of the new work is setting up those handoffs: deciding which parts of a job go to the AI, and building the steps around it. That kind of work seems to grow out of information-heavy jobs rather than routine ones.

The second new task is supervision, and who has to do it depends on experience. In a study of 5,172 customer support agents, AI assistance raised issues resolved per hour by 15% on average. The least experienced agents gained the most. The most experienced agents saw small declines in quality Does AI assistance help less experienced workers most?. For a novice, the chatbot works like a coach. For an expert, it can be a distraction they have to check against what they already know. Checking matters because chatbots sound authoritative whether or not they're right. Trust tends to attach to how expert the answer sounds, not to whether it's accurate Does chatbot language style actually shape how much we trust it?. Someone has to read critically for that gap, and the job lands on the employee.

The third new task covers what the AI still can't do. In a simulated company, leading AI agents finished only about 30% of tasks on their own. They failed mostly at social interaction, navigating professional software, and domain knowledge Why do AI agents fail at workplace social interaction?. Those gaps become human work: chasing colleagues for missing information, doing the steps inside clunky software, and filling in context. Context is a recurring weakness. Assistants have no built-in way to track what they don't yet know about the user, and giving them an explicit list of unknowns cut sycophancy and harmful advice by 50–75% Do language models know what they don't know about users?. So briefing the AI, meaning saying what it's missing before it guesses, is real and skilled work.

The surprising part is that the new tasks may depend more on how the AI system is built than on how capable the model is. A chatbot gives answers that vanish when the conversation ends. A 'colleague' system keeps state between sessions, stores reusable procedures, and closes out tasks What makes an AI system feel like a colleague rather than a chatbot?. With chatbots, people end up re-explaining context, copying outputs into other tools, and remembering what worked last time. With persistent systems, the work shifts toward maintaining the shared workspace and curating the procedures the AI reuses. That suggests the new tasks the Danish workers picked up are a moving target. They describe the current tooling as much as AI's lasting effect on work.