The best is still hard to be

And we pay a lot of money for the best.

The divinely discontent.

You could have two theories about stuff, which are really two theories about people:

1. People want stuff that is good enough. If they are buying a coffee mug, they want it to be the right size and shape; they want it to be a pleasing color; they want it to feel sturdy; they want it to keep their coffee warm. If a coffee mug manufacturer can make a mug that delivers on those things, then there’s little commercial point in developing more “features.” The mug suffices; people are satisfied; the market is satisficed. Why bother doing the hard work of making an better coffee mug? What does making a better coffee mug even mean?

2. People want their stuff to be better. Fifty thousand years ago, they wanted a cup that was not their hands. Seventeen thousand years ago, they probably wanted one that was not their friend’s head. Then, they wanted one that was clay. Then, made out of glass; then, glass, and with their favorite team on the side. Then, pretty ceramics. Then pretty ceramics with handles. Eventually, pretty ceramics with handles, with your favorite team on the side, and microwave safe, dishwasher safe, freezer safe, and oven safe up to 480 degrees Fahrenheit. And now, coffee mugs heat themselves; they heat themselves and are pretty ceramics; they are normal mugs, but the coaster heats itself.¹
   
   Perhaps, this is as far as the coffee mug can go. Perhaps this is coffee’s final frontier; perhaps we are the privileged generation that arrived at the far edge of this particular universe—but that seems awfully unimaginative. In the future, coffee mugs might be made out of indestructible ceramics. They could come with a straw. They could come with a suitcase. They could wash themselves. They could be covered with a non-nutritional, semi-permeable, non-osmotic antibacterial varnish that never requires washing. They could, through some as-of-yet undiscovered magic, keep your frozen tahini cold brew slushy that perfect consistency of slush. And when those mugs exist, will our crude clay bowls still be “good enough?”

   
   In other words, sufficient is relative; satisfaction is defined by what else is out there. What I want from a coffee mug is whatever is possible from a coffee mug, plus a little bit more. If you can invent a better mug, you can teach me to want it.

On the one hand, if you believe the first theory, you might win a Nobel Prize and a Turing Award. On the other hand, the second theory seems almost obviously true. Give us something new; we love it today; we are frustrated tomorrow. We spent millennia dreaming that we could fly; now we can, and we whine about the wifi. We loved ordering taxis from our phones; we immediately began complaining if we had to wait more than a few minutes for it to show up. We were amazed when ChatGPT could imitate C-3PO; soon, we were yelling at Claude Code because it could only conjure working software applications out of thin air 99 percent of the time. “As Jeff Bezos put it,² customers “are divinely discontent. Their expectations are never static—they go up. It’s human nature. … People have a voracious appetite for a better way, and yesterday’s ‘wow’ quickly becomes today’s ‘ordinary’.”³

Anyway, software is dead again, we’re told, because apps that used to take years to build can now be created in a weekend. From the harbinger du jour:

[By late 2026], AI had made it easier to develop and ship new features, so differentiation collapsed. Incumbents were in a race to the bottom on pricing—a knife-fight with both each other and with the new crop of upstart challengers that popped up.

…

DoorDash (DASH US) was the poster child.

Coding agents had collapsed the barrier to entry for launching a delivery app. A competent developer could deploy a functional competitor in weeks, and dozens did, enticing drivers away from DoorDash and Uber Eats by passing 90-95% of the delivery fee through to the driver. Multi-app dashboards let gig workers track incoming jobs from twenty or thirty platforms at once, eliminating the lock-in that the incumbents depended on. The market fragmented overnight and margins compressed to nearly nothing.

I do not know if you can build DoorDash in weeks, and I certainly do not know if dozens of vibe coders can recruit a network of 9 million delivery drivers to use their app. But even if they can, there’s a third assumption in this story—and in prior versions of it—that has always felt curious to me: That DoorDash, or any soon-to-be-replaced-by-a-thousand-weekend-projects product, has been successful because it built an app that crossed some threshold of satisfaction, and its continued success depends on it being one of the few apps on the far side of that threshold.

Which seems…strange? Because, maybe today’s software businesses are doomed. Chatbots built on top of spreadsheets may well be the next CRM, and perhaps Salesforce is too calcified to rework itself to keep up the modern methods of software development.⁴ It is a hard and brutal transition, I imagine, going from making mugs by hand to making them in a factory.

But once everyone has a factory, then what? Having one will no longer be a competitive advantage; everyone’s CRM will be a chatbot and on top of a spreadsheet. Then, inevitably, someone—maybe Salesforce! Maybe some former Salesforce employees!—will spend a bunch of time on the small details of that chatbot and that spreadsheet, and they will make the best one. And we’ll all decide that we’re unsatisfied with everything else, that that’s our new standard, and actually, now, we have some complaints and want something even better. Sure, success may be on the far side of a threshold—but that threshold is not static.

Prior to the internet, we could get food delivered to our house by calling restaurants on the phone. When the internet became popular, lots of companies tried to come up with ways to order food online. Then, as now, each company had access to the same foundational technologies that let them build their business cheaper than ever before. Then, as now, food delivery seemed like a “commodity:” You look at pictures of food; you push a button; it arrives at your door; do you care about the app that you do it on? And yet, DoorDash figured out the best way to do it, and most people use DoorDash.⁵

Why does AI change that story? Sure, it might change how we order food—maybe the food is made by a robot; maybe it’s delivered by one; maybe it is ordered by Oura rings sensing that we’re hungry⁶—but if anything, that seems to create more opportunities for differentiation. Someone will be the best at that new thing, and being the best at anything is almost tautologically hard to be.⁷

The “cost of creating content going to zero” didn’t kill content, nor did it bankrupt the business of content creation. We still like watching videos; we still have opinions about videos; it’s still hard to be someone who makes the best videos; if you are, we will still pay you a lot of money for your videos.

It seems inevitable that lots of people will also continue to have opinions about software—defined broadly, at least, as a thing on a computer that does stuff for us. And if we have opinions about it, it also seems inevitable that we will be, as always, divinely discontent about what it can do.

---

Department of Talent War

If you are a normal company, you hire people to help you sell stuff. Though those two things might periodically overlap—you might try to impress prospective employees by telling them that you won a big contract with an important customer—that’s the order of operations: recruit people, build things, make money.

If you are an AI company—and if you believe in some degree of science fiction—that formula is backwards. Your plan for market domination (world domination? Galactic domination?) is not to hire people and then make money from what they build; it is to be the first company that creates an AI model that is good enough to improve itself. That model will begin compounding its own capabilities, you will accelerate far ahead of the other AI companies, and the model will then tell you want to do. “Once we’ve built this sort of generally intelligent system, basically, we will ask it to figure out a way to generate an investment return,” Sam Altman said, a few years ago.

So, your order of operations is reversed: You want to build popular things and make money so that you can recruit AI researchers. (If you aren’t an AI researcher and you want to work at OpenAI, come be a recruiter, Sam Altman said yesterday.) Customers are useful, mostly, because they make your company an important place to work.⁸ Knowing that millions of people are using your research to do good and meaningful things is probably rewarding, for many AI researchers.

But of course, many AI researchers might prefer that some people don’t use their research to do certain things. Some of those people might be the United States Department of War, and some of those things might be “autonomously killing people without human oversight.” And so, if you are an AI company, you might want the Department of War’s $200 million—though, again, not that much—but what you really want is a bunch of AI researchers to like you and want to work for you.⁹

Anyway, last night, Anthropic CEO Dario Amodei said that they would not allow the U.S. government to use their AI models to surveil U.S. citizens or power fully autonomous weapons. This morning, 404 employees from Google and 74 from OpenAI signed an open letter asking their employers to be more like Anthropic.

If you aren’t an AI researcher and you want to work at Anthropic, maybe try being a recruiter. They are probably busy.

1. And doubles as a lamp! And a fondue pot!

2. Via Ben Thompson, via Eugene Wei.

3. “Never test drive the nice car,” I frequently remind myself, as I am test-driving the metaphorical nice car.

4. Or, perhaps not.

5. The app may not be the reason people use DoorDash; it might be the selection of restaurants, or the prices, or “business practices,” both Bad and Good. The point is that many companies wanted to be DoorDash, and only DoorDash figured out how to do it.

6. The future of commerce, according to Stripe:

The system already knows the school calendar, your son’s preferences, and your typical budget. All you do is receive a notification: here’s the back-to-school list of everything that’s been purchased. This is the most futuristic vision, where the things you need show up right before you need them, without you having to ask.

7. You could argue, as some of the “death of software” posts do, that, if the world is run by robot assistants that use software, then everything is the same and nothing is the best. But the robots aren’t just a generic model; they are models and, still, a lot of specialized software. Some robots will be good; some will be bad; we will still want things to be better; we will stick pick the ones we like best. And even if we don’t, and everything is a text box for talking to a generic model, the models have opinions too.

8. Making money from customers is also useful, insofar as it helps you pay your researchers. But when you can raise $110 billion directly from investors, their money doesn’t matter all that much.

9. Of course, if you run an AI company, it is also possible that what you really, sincerely want is to not destroy society, or humanity. It is possible for people to also have principles (and, you know, reliable AI models).