Post

Something we'll have to grapple with as we enter the future of work with AI tools: What to do with the lag time as an LLM processes an output. Or: what do you do when AI is working on something?

People want to pretend like AI is "so fast" at doing everything. But the reality is that it still takes time. Plenty of the complex tasks I give Claude take 10-15 minutes to run and spit out. So what do you do in the meantime with that time?

The challenge is not knowing how long that time will take. And not knowing how deep to get into your own work before it's finished and pulls you back out. Basically, this is another thing people can "multi-task" at, which is really just productivity-killing task switching.

We can imagine treating this like delegating an assignment to a colleague. What did you do while you waited for a co-worker to get back to you? Also, would you let them interrupt you when they finished, and force you to drop everything you're doing?

Keep in mind, this is also a co-worker who has very strict (but vague) limits on how much work they can do in a set period (token caps). So you also have to wait (sometimes hours) for a limit to reset, or pay through the nose for usage overages.