The Five Thousand Minutes Problem.
The short version: a case for AI etiquette. When something took two minutes to generate and takes thirty to read, the sender has spent the reader’s time, not their own — and the backlash lands on everyone doing serious AI-assisted work. Two norms to start with: spend the time you expect your reader to spend, and show how you used AI. And when you review AI-built work, improve the factory, not the car.
01Anecdotal hook
I spend a lot of time on stages and panels these days.
People often ask how I got so comfortable up there, and assume I love the spotlight. I’m not an extrovert. It comes down to practice.
The spark came from a book: Scott Berkun’s Confessions of a Public Speaker. One idea in it stayed with me. When a hundred people listen to you for an hour, that’s a hundred hours of their lives. Do the same sum for ten minutes in front of five hundred people and you are asking them for five thousand minutes. You had better be confident in yourself, in your ability and in what you will bring. If you don’t think it’s worth putting serious preparation in to earn those five thousand minutes, don’t do it.
That doesn’t mean five thousand minutes of preparation for every talk. It’s a value equation: if people will spend time on what you give them, you should put time into it yourself. That balance is exactly what has gone wrong in much of AI-generated content, and it’s the source of most of the frustration.
It’s not only about time, though. It’s also about intent. A painter can cover a canvas in a single black square, or a musician can write a very simple song, and the work can carry more artistic value than a beautiful picture that took ten times as long. The value lies in the person and the intent behind it. Out of everything they could have made, they chose this one thing and took a risk on it. That choice is what makes it real.
The same holds for AI. The value is not that a model produced something. It is that out of everything you could have spent your time on and everything you could have said, you worked with AI and chose to deliver this specific thing. The reader should be able to see that choice.
02Conceptual swing
Two kinds of AI.
AI has split us into two groups, and the split runs through both the people who create content and the people who read it.
On one side are the early adopters who have built real systems. They have frameworks, guardrails, curated context, data governance and testing. Their output is shaped by actual thinking, real calculations and a deliberate strategic framework. They tailor a message to its audience because the system underneath was designed to do that.
On the other side is everything else. Our organisations, our media and even our music are filling up with low-substance content. It wastes people’s time and annoys readers. It also does damage beyond itself, because it creates a problem for everyone producing high-quality AI-assisted strategy, analysis, music, video and art. When most of what arrives is noise, the signal gets ignored along with it.
What I value, and what I don’t.
What I value most in my colleagues is their connection to customers and the market, and how sharp they are about both. AI and information systems still capture that poorly. A recording of a meeting is not the meeting. The texture of the room, the hesitation before an answer and the thing a client almost said are all in the people who were there.
I also value the people who see patterns, connect dots, run the numerical work and the lift studies, and shape strategy from it. That work is changing. Increasingly, I want to reward the people who build the right bridges at an architectural level, between systems, datasets, trends and market events. I value someone who has set up AI properly far more than someone who has drawn a nice framework on a slide. A strategist who gives AI the right guardrails and context, picks the right themes at the right moment, and helps others move with those insights can produce and land strategy at a pace we have never seen before.
What I do not value is something else. It’s the person with little substance, little context and little time spent with customers, who uses AI to answer their own questions and worries, generates a document and sends it to me as something I now need to respond to.
The faux pas.
There is a real risk that people decide AI-generated work is not worth their time, and to be fair, that reaction has grounds. Most of us have received something that took the sender two or three minutes to produce and will take us thirty to work through.
Worse is the feeling of being tricked: receiving something that was never meant to be real, presented as if it were. Examples include life advice on raising your children, getting healthier or balancing your life, produced after ten seconds of prompting and handed over as a five-minute read. There are the fabricated “real life” viral videos. And in our own world, there’s the person who did not do the homework, did not have the customer meeting and did not gather client insight, yet uses AI to build something large around a moment that was in fact very small.
These are poor uses of AI. They are disruptive, and they are a breach of etiquette.
The spiral.
The backlash brings its own risk. People start hunting for signs of AI in everything they read, and some refuse to read it at all. That is a catch-22, and it drives a negative spiral.
It also punishes exactly the wrong group: the people using AI to do things they could never do before. They’re connecting strategy in new ways, running analysis they never had time for, and bringing deep or niche insight to the table. If we treat all AI-assisted work the same way, we lose the best of it.
03Framework
Two rules to start with.
I propose two simple norms.
- Spend the time you expect your reader to spend. If a document will take someone thirty minutes, it should have taken you at least that much real effort to make, through the system you built, the context you gathered or the judgement you applied.
- Show how you used AI. Say briefly how you set it up and how it helped you reach your insight or conclusion: the data you used, how it was cleared and governed, and how you tested the output. That tells the reader this is not a forwarded answer to a prompt but the product of a deliberate setup, and that it deserves to be taken seriously.
The reader’s side: discuss the factory, not the car.
Etiquette runs both ways. Working with AI-generated documents, structures and strategies calls for a different kind of collaboration. The aim is no longer to change individual sentences or boxes. It’s to improve, together, the system that produced them.
Take quality control on a strategy deck. Using an agent to review it doesn’t mean I stop reading it myself. It means I can be far more meticulous than I could ever be alone in the time I have. I can check every calculation and every number, however large the deck is. That doesn’t come from one clever prompt. It comes from a set of scripts running in the background that trace the figures used across the slides. I’ve built a way of reviewing a deck through eight different lenses and stakeholder views, covering layout and formatting, usability, the numbers, business language and strategic drift, among others. These are basic checks, but they raise the baseline enormously.
So when I sit down with a strategist, the conversation is about how to use that agent and what its input and output tell us about improving the deck. Points of agreement or disagreement on individual slides can still be handled afterwards.
The best way I can describe it is that we are no longer discussing a finished car. We are discussing the factory. You build a better car by building a better factory, not by reworking one car at a time. It takes more effort up front, and it feels less direct than just making the edit. But the conversation shifts from the outcome to the process that can recreate that outcome again and again.
I learned this from composing music and setting up studios. When a sound is off, you don’t patch that one recording. You fix the signal path at its source. Fix it in the production line, not in post-production.
04Invitation to growth
A starting point.
These two rules are not a finished rulebook. They’re a first draft of an etiquette we will need to write together, as creators and as readers, as the tools and our skill with them keep improving.
My ask is simple: respect the minutes you ask of others, be open about how your work was made, and when you review, help improve the factory. If we get that right, the people doing genuinely new work with AI will get the attention they’ve earned, and the rest will improve because of it.
Where the numbers came from.
In the spirit of rule two: the argument and the draft are mine. An AI model helped edit it, a coding agent set it into this page, and the four illustrations were generated on a local Flux model, then chosen by me.
A hundred people listening for an hour is a hundred hours of their time.
Scott Berkun, Confessions of a Public Speaker (O’Reilly, 2009). The passage, as recorded in Derek Sivers’ book notes: “When 100 people are listening to you for an hour, that’s 100 hours of people’s time devoted to what you have to say.” The ten-minutes-times-five-hundred version is my own arithmetic on the same idea.
“Two or three minutes to produce, thirty to work through” is an author observation from my own inbox, not a measured average.
If any claim here is mis-cited or out of date, mail me at rt.nl/contact and I'll fix or retract.