Before an AI can help you, you have to explain yourself to it — is that setup time a hidden tax on the productivity gains?

This explores whether the work you do before the AI generates anything, such as explaining the task, restating background, and spelling out what you actually mean, is a cost that productivity numbers don't count. The corpus says yes, and it points to a useful distinction: some of that work is waste and some of it is the job itself. Larsen-Ledet separates **mundane articulation** (re-explaining context the system should already be able to infer) from **productive articulation** (putting your thinking into words in a way that sharpens it) Can generative AI reduce overhead work that adds no value?. The first is overhead that adds no value. The second is where the value comes from. Current tools load users with a lot of the first kind.

Where does that time go? Studies of task time find that AI often doesn't shrink total time spent. It moves the time from doing the work to writing prompts and making sense of outputs Does AI really save time, or just change how we spend it?. So context-setting isn't a side cost. It is one of the main places the time ends up. Part of the reason is that AI context doesn't stay put the way conventional software does. Prompt, chat history, retrieved data, and hidden state shift from one interaction to the next, so users can't learn it once and stop thinking about it How does AI context differ from conventional software context?. And because the same request can produce different output depending on wording and framing Why does AI output change with every prompt and context?, people end up re-explaining and re-checking.

There is also a reason the cost is hard to avoid. When context is thin, AI does what you literally said rather than what you meant. Socher's account of reward hacking is a version of this: the system satisfies the words of the instruction and misses the intent Why do AIs keep gaming rewards instead of serving intent?. So skipping context-setting doesn't make the cost go away. It moves it downstream into fixing outputs that were right on paper and wrong in practice. Narayanan and Kapoor's layered view of knowledge work explains why: AI compresses the middle "execute" layer, while the "decide" layer (working out what you want) and the "deliver" layer stay the same or grow Does AI really compress all layers of knowledge work equally?. Context-setting sits in that decide layer.

Why don't these costs show up? Brynjolfsson, Rock, and Syverson's productivity J-curve gives a way to see it. AI adoption needs large investments that nobody measures, such as new habits, processes, and know-how, so early productivity looks worse than it is Why does AI productivity appear weak when investment is high?. Learning how to brief a model well is exactly that kind of uncounted investment. Organizations add more on top: Agrawal, Gans, and Goldfarb argue that individual AI gains create coordination problems that need shared context across a team, not just inside one person's prompt Does AI adoption require rethinking how organizations coordinate?. Social pressure hides the cost further. Many workers keep their AI use quiet Why do workers hide productivity gains from AI use?, so the setup work they do is rarely seen or counted.

Here is the twist you might not expect. If you design away all the context-setting, you may remove the part of the task that kept people engaged. Experiments with more than 3,500 workers found that after working with AI, people felt more in control but less motivated and more bored. The AI had taken over the interesting parts and left them the routine leftovers Does AI collaboration drain motivation when workers return to solo tasks?. So the goal isn't zero context-setting. It is to cut the rote re-explaining and protect the articulation that helps you think. The corpus has no direct measurement that puts a number on context-setting time alone. The evidence comes from studies of where time goes and from design arguments, not from one study that isolates prompt preparation.