No. 27AI Productivity tools

Execution is the moat

I was watching a YouTube video about making sleep videos with AI. You’ll know the kind of thing it was teaching: three hours of rain on a window over a moody generated scene, soft piano underneath, and often only a few minutes of audio on a loop. The trick is in the title, which promises no mid-roll ads so nobody gets woken at three in the morning, and the money comes from the adverts at the start. YouTube lets a creator put mid-roll ads in anything over eight minutes and switch them off per video, so a sleep channel’s whole pitch is that it has switched them off. The presenter showed real channels earning serious money this way. When I watched, the video had around 300,000 views.

Underneath, someone had left a comment: execution is the moat.

My first reaction was that it’s the wrong way round. It’s a startup saying, the sort of thing that gets repeated at pitch events, and AI has made execution about as cheap as it has ever been. If anyone can do it, how can doing it protect you from anything?

Then it clicked.

Too easy to bother

I’ve said before that maybe one in a hundred people who watch a video like that go and do the thing. I now think it’s far fewer, and the nearest thing I can find to evidence agrees with me.

The videos are fun to watch, and they make it look almost too easy. That does two things. Some people think “I’ll do that another time” and never do. Others think “if it’s that easy, everyone else is doing it” and don’t start at all.

And they’re half right. Do the maths. If all 300,000 viewers went off and made a sleep channel, there would be 300,000 sleep channels and none of them would earn anything. It only works because almost nobody follows through.

The best numbers I know on follow-through come from free online courses, which isn’t the same thing but is close enough to be uncomfortable. When Justin Reich and José Ruipérez-Valiente looked at Harvard and MIT’s courses on edX for Science in 2019, about 3 per cent of the people who started a course finished it. Of the ones who said at the outset that they intended to finish, roughly one in seven did. Most never came back for a second year. Those people had gone as far as signing up, which already puts them well ahead of someone watching a how-to video on the sofa. If the sofa crowd behaves anything like them, my one in a hundred was generous.

Who gets paid

The person making the tutorial is selling shovels in a gold rush, and the original shovel seller is worth a look. Sam Brannan had a store at Sutter’s Fort when gold was found nearby in 1848, and he never dug for any himself. He sold supplies to the people who did, up to $5,000 of goods a day, and by most accounts he was the one who ran through San Francisco shouting about the gold in the first place. Historians treat the shouting as folklore, but it’s the right folklore. The shovel seller is the one telling you there’s gold.

The presenter is in the same trade. They may well have made their money from the thing itself first, but now they’re paid by every single person watching. A hundred per cent of their audience contributes to their income. Only a sliver of that audience goes on to compete with them.

Fig. 1 · Who the tutorial pays

Everyone pays, almost nobody competes. The presenter is paid by all 300,000. A generous guess says 3,000 have a go, and if they behave like online learners, about 450 see it through.

Ebooks are the clearest case. There are endless courses on finding a niche and then filling it with a book written by AI. If the niche-finding really worked, the people selling the courses would be writing the books in those niches, not telling you where they are. They’ve worked out that teaching it pays better than doing it, probably because there aren’t many niches left. By the time you’re watching the video, you’re late to the party, picking over whatever’s left.

It’s worth noticing where the big numbers come from, too. Most of the “ten thousand a month” figures you see on thumbnails come from people selling courses about it, which tells you something about where the money actually is.

Not free, not effortless

AI looks cheap, and it looks free. It isn’t free: generating the pictures and the music costs money, even when a few minutes of it is looped for three hours. And there’s still real skill in it. The presenters make it look effortless because they want you to enjoy the video.

I’ll give them credit, though. The good ones are open about it. They name the channels and you can check: the channels are real and the view counts are real, and the money adds up. Fortune profiled one such network at the end of 2025, a mostly AI “faceless” operation its owner said grossed about $700,000 a year, whose biggest channel posts six-hour history videos to fall asleep to. Fortune saw payout records, though much of the rest is his account. The channels the presenter named are still up and still running, too, not quietly pulled.

And the better ones aren’t just a static image over a loop. There’s craft in the visuals and the sound, and that’s part of why they’ve lasted. It may also be what keeps them up. In July 2025 YouTube renamed its “repetitious content” rule “inauthentic content”, and its examples now include AI-made videos built from generic templates that give the impression of mass production. It doesn’t mention sleep videos, and I’m not saying anyone is about to be banned. But a channel that is ten interchangeable loops is plausibly at risk under that wording, and a channel somebody has actually made is less so. The craft is what makes a channel look like a person’s work rather than a template’s.

Watching instead of doing

The bit I keep coming back to is the watching. I’ve noticed it in myself: a lot of what feels like progress is consuming advice about it. There’s a name for it, action faking, a term popularised by the entrepreneur MJ DeMarco, and it covers everything from buying the domain to watching the fourteenth tutorial. These how-to videos are addictive, partly because they’re made with the very tools they’re teaching, with slick editing and lovely motion graphics. They’re entertainment now.

It’s also exactly the kind of habit a room makes easy. I argued in Stop waiting for willpower that the people who look disciplined aren’t resisting anything; they’ve set things up so the right thing is the easy thing. The sofa and the autoplay are set up for the opposite. Watching another one is the path of least resistance, and it feels productive, which is the dangerous part.

I can’t prove more people are watching passively than before. You could in principle compare a video’s views against the number of channels that copy it, but YouTube is far too big to measure properly. It certainly feels that way, though.

Worth the punt anyway

None of this means don’t bother. Even if the gold rush is mostly over, it’s worth the punt when you enjoy the making, and the craft is still where the difference shows. Brannan did well out of the miners, but some of the miners found gold.

The one thing I won’t let go is taste. I argued in the seven pillars that skill has narrowed to taste, and I still think it’s the most important thing. The day I think mine has gone off is the day I stop making things.

Which brings me back to the comment, and why it’s right after all. Derek Sivers wrote in 2005 that ideas are just a multiplier of execution, and an idea with nothing done about it is worth almost nothing. That was true when execution was hard. It’s still true now that execution is cheap, because cheap turns out not to mean done.

Execution isn't the moat because it's hard any more. It's the moat because so few people bother.

Execution isn’t the moat because it’s hard any more. It’s the moat because so few people bother.