The AI Waiting Room: What Are We Actually Doing While the Robots Think?

Right now, as you read this, somewhere in the world a person is staring at a spinning cursor waiting for an AI to finish thinking. They have been waiting for eleven seconds. It feels like eleven minutes. They have already opened two other AI tools in separate tabs. They are considering making their fourth coffee of the morning. They will not make the coffee. They will just sit there, refreshing the mental equivalent of F5, watching that little animated circle go around like it owes them money.

That person is me. That person is probably you. That person is the defining professional experience of 2025, and nobody is talking about it.

We talk endlessly about what AI is doing to our work. We do not talk nearly enough about what we are doing while AI is doing our work. Which is, mostly, frantically context-switching to another AI tool, opening Instagram, staring at a wall, or conducting a deeply personal audit of whether we actually enjoy our careers.

The Seven-Tab Life

Here is my honest daily reality: I am running seven AI tools regularly. Claude for thinking and writing. ChatGPT because sometimes I want a second opinion and also because old habits. Perplexity when I need to search and do not want to wade through SEO slop. Midjourney when something needs to look good. A couple of others that I will not name here because this is not a sponsored post and also because I switch them in and out depending on the week.

The practical result of running seven AI tools is that I am never actually waiting. I am just waiting in rotation. I send a prompt to Tool A, immediately open Tool B, send something there, check back on Tool A, open Tool C, forget what I asked Tool B, go back to Tool B, and so on, in an infinite loop that somehow produces useful output and also a mild background anxiety that I cannot quite shake.

This is the modern knowledge worker's version of the old spinning plates act. Except the plates are intelligent and sometimes the plates talk back and occasionally a plate tells you it cannot help with that request.

We are all plate spinners now. And between spins, we are all doing something. The question is what.

What People Actually Do While AI Thinks (A Completely Unscientific Survey of My Own Observations)

I have started paying attention to this. Here is what I have catalogued, in rough order of how often I have caught people — including myself — doing it.

Opening another AI tool. This is number one by a wide margin. The wait time of one tool has effectively become the productive time of another. We have accidentally created a new form of parallel processing where the human is the CPU scheduling jobs across multiple inference engines. Nobody designed this. It just happened.

Making coffee. The classic. The eternal. There is only so many times you can make a coffee in a day before you have to have a real conversation with yourself about what you are avoiding. I am told the average knowledge worker makes 2.3 coffees per AI response. I made that statistic up but it feels accurate.

Checking email, then immediately closing email. Not to reply. Not to process. Just to look at it and confirm it still exists and that the world still has problems that require your attention later.

Staring out the window thinking profound thoughts. This is the one people will not admit to but it happens constantly. AI-induced micro-meditations. Someone should study this. There is probably a wellness product in it.

Re-reading the prompt and quietly judging yourself. Was that a good prompt? That was not a good prompt. You asked a five-hundred-dollar question in a five-dollar way and now you are going to get a five-dollar answer and it will be your own fault. You will still use it.

Going on Instagram. No comment. We all know.

Having a full philosophical crisis about the nature of work and value. If this machine can do in eight seconds what took me three hours, what exactly am I being paid for? What is the value I am adding? Is it the prompting? Is it the judgment? Am I the prompt engineer or the prompt? Heavy questions for a Tuesday morning. Too heavy for before the third coffee.

The Caribbean Angle: We Have Always Been Good at Waiting

I am going to make a regional observation here that my Caribbean readers will recognize immediately and that my non-Caribbean readers might find instructive.

The Caribbean has always had waiting as a feature, not a bug. We wait at the bank. We wait at the warden office. We wait for the contractor to show up on the day he said, which turns out to mean sometime in the next four days. We have developed an entire cultural relationship with waiting that is, frankly, more sophisticated than anywhere else in the hemisphere. We do not just endure waiting. We have turned it into a social activity. You talk to the person next to you. You get information. You learn things. You do a little business right there in the queue.

The AI waiting room is, structurally, the same thing. And Caribbean operators might actually have a slight cultural advantage here, because we have never confused being busy with being productive. We know the difference between the time you spend doing and the time you spend thinking about what to do next.

The person who uses the AI wait time well is the person who already knows what question they are asking next. The person who uses it to go on Instagram is the person who, when the AI finishes, has to read the output, figure out where they left off, and then wonder why the whole thing took three times longer than expected.

The Real Question: Are We About to Wait More or Less?

Here is where it gets interesting. Models are getting faster. Inference costs are dropping. Response times that used to be thirty seconds are now three seconds. You would think this means less waiting.

I do not think that is what happens. I think what happens is Jevons Paradox for AI: as the cost and time of a resource drops, consumption of that resource increases proportionally or more. When AI responses get faster, we do not use AI less frequently. We use it more. We ask more complex questions. We run longer chains of prompts. We build workflows that would have seemed absurd when we were waiting ninety seconds per response but that make perfect sense when each step takes four seconds.

The wait time does not go to zero. It gets redistributed. Instead of one thirty-second wait, you have twelve three-second waits. Total wall clock time: roughly the same. Cognitive experience: completely different. You feel like you are moving fast. You are basically playing a video game at this point, tapping through prompts and responses at a pace that would have been unrecognizable two years ago.

And then there is the other direction. Agentic AI — systems that do not just answer questions but actually go and do things across multiple steps, multiple tools, multiple decisions — these take longer per task, not shorter. You fire off an agent to research, synthesize, draft, and format something that would have taken you four hours. The agent takes twenty minutes. You wait twenty minutes. What are you doing for twenty minutes?

The answer had better not be Instagram.

The Opportunity Nobody Is Building Yet

Here is my genuine thesis: the AI wait time is an underexplored surface area, and someone is going to build something useful in it.

Think about what you actually need when you are waiting on an AI response. You need to stay in context — to not lose the thread of what you were working on. You need your next prompt ready. You need the relevant materials nearby. You need to be in a mental state where you can evaluate and redirect the output intelligently when it comes.

What most people do instead is break context. They open something else. By the time the AI responds, they are three cognitive steps away from where they were, and now they have to pay a re-entry cost to get back into the problem.

The product opportunity here is anything that keeps you in-context during the wait. A notes surface that auto-populates with context from your current prompt. A sidebar that suggests your likely next steps based on what you just asked. A timer that shows you not just that the AI is thinking, but what it is likely doing and how close it is to a meaningful checkpoint. Something, anything, that converts the dead space into forward motion rather than drift.

There is also a wellness angle here that I find genuinely interesting. The AI wait time might be the last natural break in the knowledge worker's day. The moment before the response arrives is the moment before you have to think again. Someone will figure out how to make that moment intentional and valuable instead of accidentally anxious. Micro-recovery between cognitive bursts. Structured micro-reflection. A fifteen-second breathing prompt between prompts.

I am serious about this. Do not laugh. The Pomodoro technique turned rest into a productivity system and people paid for it. The AI wait time is two to thirty seconds of enforced pause, happening dozens of times per day, for hundreds of millions of people. That is a lot of surface area for someone to do something useful with.

What I Actually Recommend Doing While You Wait

Since you are going to wait anyway, here is what I think is actually worth doing in those gaps.

Write your next prompt before the current one finishes. This is the highest-leverage thing you can do. If you already know what you are going to ask based on the likely range of responses, you cut your total session time significantly and maintain much better context throughout.

Note the assumption you are testing. Before the answer arrives, write one sentence about what you expect it to say and why. This sounds unnecessary. It is not. It forces you to actually have a hypothesis rather than just consuming output passively, and it sharpens how you evaluate and use the response.

Do not context-switch to anything that requires reading. Instagram, email, news — all require you to read and process new information, which competes directly with what you were just doing. If you need to break context, break it with something physical. Stand up. Look out the window. Do ten seconds of something that does not involve a screen.

And if you are going to make coffee — and you are going to make coffee — at least make a good one. We are in the Caribbean. The coffee is excellent. You might as well.

---

The waiting is not going away. The models are getting faster and we are getting more ambitious with what we ask them to do, and those two forces are going to roughly cancel each other out for the foreseeable future. We will always be in the waiting room. The only question is whether we are going to use the time well or spend it watching a little circle go around.

The waiting room has always been where the real work happened. Ask anyone who has ever sat in a line in Port of Spain long enough to think clearly about their life.

*Miles Abraham builds businesses in markets people underestimate. He has been doing it in the Caribbean for 25 years.*