We have always found ways to feel productive while avoiding the hard work of thinking. AI agents just gave us the best one yet.

There’s a term in behavioral economics that I think helps organize the world in a very interesting way. There are many such terms, but this is one I particularly love: structured procrastination.

The idea is that we want two things at the same time, and they pull in opposite directions. We want to feel that we are making progress. But we also want to procrastinate — not to put in hard effort right now. The hard effort we happily assign to some future version of ourselves; the present version prefers to do things that are easy and fun. Structured procrastination is a kind of compromise we strike between these two desires: activity that delivers the feeling of progress without demanding the effort.

In the old days, we had two main ways of managing this balance.

The first was tidying. We organized our desks. We wrote to-do lists — and if we were really good at this, the first item on the to-do list was “write a to-do list,” so that we could cross something off within seconds. We got the sensation of achievement without investing any real energy.

The second was paperwork. Whenever something difficult and complex was waiting for us, we could escape into it: forms, procedures, all the tasks that don’t require much thought but give us the feeling that we are moving forward. After all, at some point this paperwork has to be done. Why not now?

The third version

I think a third version of structured procrastination — and surely the most interesting one — is emerging now, in the era of AI, and in particular of AI agents. This one is more seductive than its predecessors, because it lets us feel that we are making progress while avoiding the hardest work of all: actual thinking.

Imagine you have a real task in front of you. One path is to think, think, think, plan, plan, plan — and then execute. This is difficult, and not only because thinking is effortful. It’s difficult because as you think and plan, you might discover that you got things wrong. That your idea is not at the quality you expected. That there are parts you can’t yet solve. Thinking is where our ideas get confronted, and confrontation is uncomfortable.

The other path is to take a half-baked thought and send your agent to start working on it.

Here is what makes this genuinely new. Until the era of AI, structured procrastination was a choice between tasks. We could pick something that required deep thinking, or we could pick something shallow — the to-do list, the emails, the bureaucracy. Now the choice lives inside a single task. Let’s say I need to accomplish something for work. Within that very same task, I can decide to spend a lot of mental energy — be deliberate, be thoughtful, make a plan — or I can simply say to my agent: “Hey, I have this half-baked idea. Start working.”

Tokens are the new to-do list

Remember that structured procrastination is a balance: we don’t want to spend the effort of thinking, but we still want to feel that we’re getting somewhere. So where does the feeling of progress come from in this new version?

It comes, in magnificent form, from watching our AI agent burn tokens.

In the past, you would write a to-do list and cross things off, and every crossed-off line gave you a small hit of progress. Now you can watch the agent work. You can see the output streaming, the tasks being spawned, the tokens being consumed — and that gives you exactly the same feeling. Something is happening. Things are moving. I am making progress. Not only that, but it’s burning millions of tokens. What a great feeling of progress. So much hard work is being invested — and there is a counter to prove it.

So what I would like to argue is that we have entered a new era of structured procrastination. It is no longer just about the choice between tasks; we can now procrastinate inside a task. And the feedback we get in the form of tokens consumed is a bottomless supply of exactly the satisfaction that structured procrastination demands from us. The to-do list eventually ran out of items. The token counter never runs out.

What doesn’t change

The uncomfortable part is that the thinking still has to happen. The question is only when. If we do it before we hit “run,” we discover the holes in our idea cheaply, while they are still just thoughts. If we skip it and let the agent run, we discover many of the same holes later — after the tokens have been spent, and after we have grown attached to output that looks finished.

So here is a small test I’m trying on myself. Before I send an agent off to work, I ask: have I done enough of the difficult thinking? If the honest answer is no, then what I’m about to do isn’t delegation. It’s structured procrastination with a very convincing progress bar.

I’d be curious to hear your version. What is your “hey agent, start working” moment — and what did it let you avoid?