In New York in 1919 a young copywriter working at N.W. Ayer was handed an account that few others in the agency wanted to work on. The copywriter, Raymond Rubicam, was asked to write three full page magazine advertisements for Steinway & Sons. The piano maker had long thought that advertising was somewhat beneath them. Ayer himself had spent the best part of twenty years trying to talk them out of that view, with little success. Charles H. Steinway, grandson of the founder, had run the company since 1896 and saw no reason to change a position that the family had held since the company was founded so the brief that Rubicam got was to update what had been done before.
Rubicam struggled for days to come up with anything interesting to say until he looked in the company’s information file and found a proof book of old Steinway adverts featuring page after page of elegant women sitting at elegant pianos in elegant drawing rooms. Minimal copy on the ads said almost nothing about the instrument. In the same file Rubicam read that Steinway pianos had been played by almost every great pianist and practically every great composer since Wagner. This had been known by the company for almost fifty years but nobody had thought to talk about it. The resulting advert that ran in January of 1919 read ‘The Instrument of the Immortals’. Sales went up 70% and the ad is often singled out to this day as one of the most effective adverts of the 20th Century. Four years later Rubicam left Ayer to found Young & Rubicam.
What the copywriter had stumbled upon was no accident of fate or taste. The reason that all those renowned pianists played Steinway pianos was because the pianos were genuinely better than any rival could produce at the time but also due to something that William Steinway had started building back in 1866. Henry Steinway Jr’s original 1857 patent for a piano action was the first in over a hundred patents that the company would accumulate. In the years that followed William’s brother C.F. Theodore struck up a friendship with the physicist Hermann von Helmholtz whose work in acoustics led to the company patenting a unique design that let sections of string that had previously been damped and useless vibrate in sympathy with the notes being played. The company had literally engineered a fuller, richer sound into their instruments.
But the company still had the problem of how they could demonstrate the superiority of their pianos when most people can’t tell the difference between an excellent instrument and a very good one. William Steinway’s answer to this conundrum was to build a concert hall. Steinway Hall opened on East 14th Street in New York in 1866 at the cost of $200,000. It was the second largest auditorium in the city and it became the home of the New York Philharmonic for twenty-five years until Carnegie Hall opened in 1891. The venue was hugely popular with music lovers but to get to their seats every member of every audience had to pass through the Steinway showrooms. In the year after it opened the sales of Steinways increased by more than 400 pianos.
At the time it was quite common for manufacturers to donate a piano to famous pianists in the hope of sales by association. But a few years after opening the hall, William took a different approach. He engaged the Russian pianist and composer Anton Rubinstein to undertake his only American tour playing Steinways in front of thousands of American enthusiasts at 215 concerts over 239 days. Rubinstein became the first Steinway artist, and the tour signified the beginning of a system which would cement Steinway’s position as the most sought after piano maker in the world. Every great pianist ended up playing a Steinway because the company owned the hall they played in, paid for the tour that made their reputation, and spent decades ensuring that the instrument waiting for them on stage was the best it could possibly be. The company had built a system that continually manufactured the evidence for its own superiority. Every concert was a demonstration that seemed like an independent endorsement.
Today Steinway maintains what it calls the Concert and Artist Bank, a pool of around 300 concert grands held across its dealer network which the dealers are required to maintain to a concert standard so an instrument is available whenever an artist needs one. This is run alongside the Steinway Artist programme (which covers 97% of the world’s active concert pianists) and its All-Steinway Schools, a designation which conservatoires and universities earn by committing almost exclusively to using Steinway pianos (and which some have said is anticompetitive). No Steinway artist is paid to be on the programme. They have to own a Steinway themselves to qualify, and perform on the company’s piano’s because they want to.
The system is a closed loop. Pretty much every artist of repute wants a Steinway and wants to perform on a Steinway because it’s the instrument they learned on, and it’s the one that they can guarantee will be there at the concert venue, tuned and ready to play at the highest level.
The same kind of closed loop is forming in AI right now, except that it runs on context rather than artists and concert stages. The more we use particular AI models the more they learn about us as individuals and the quirks of our specific ways of working. Every chat, document, revision or correction is an opportunity for the model to learn about you. Memory, connectors, and linked systems all add to the closed loop. Meaning that every interaction becomes slightly better than the last in a compounding loop that makes it harder for us to switch (even when some models allow you to import your history). Like Steinway, the conditions under which preferences are forming were arranged by someone with a vested interest in the outcome.
How do you compete with a closed loop? By changing the territory on which competition happens. Steinway competitors Yamaha recognised how difficult it would be to overturn decades of Steinway dominance on the concert stage. Instead they launched the Yamaha Music School which grew into a global teaching system that the company says has now run in more than forty countries and taught over five million students across thousands of locations. If Steinway was going to own the soloist stepping onto a stage, then Yamaha could own the first moment that a child touches the piano keys. The company went on to become the largest musical instrument maker in the world, putting millions of pianos into homes and classrooms around the globe.
The frontier labs are built on the principle that bigger is better. More data, more compute, more energy-hungry infrastructure. And the competitive response that often gets discussed is a smaller, cheaper, faster model that eventually catches up in capability. I think this is less convincing because it’s competing on the same axis that the incumbents have arguably already won.
The more interesting possibility is a hybrid one. The frontier models possess deep general knowledge, can handle genuinely hard reasoning, and they make excellent orchestrators. They aren’t going anywhere. But there is a plausible future in which a growing share of everyday work runs on smaller models held close by, on your device or inside the organisation, handing off to the cloud only when the complexity of the question warrants it. The reason that would matter is less that the local models are faster, cheaper or more private (though they may well be all three), and more that they would sit on much better context. A more detailed accumulated understanding of how you work – effectively the thing that has kept us locked in to particular frontier models to this point, but much deeper. It’s a similar advantage but operating at a different layer, the layer where preference forms.
Yamaha never did take the concert stage from Steinway and in many ways they never needed to. Steinway competes for the moment that is rare, visbile and consequential. Yamaha went for the one that is frequent, less glamorous, and repeated thousands of times before anyone gets really good. Two different layers rather than different positions on the same one. Which is a more useful way of thinking about the AI market. The frontier models keeping the irregular but harder questions, the local models working on the continual, everyday ones where habits form. So rather than asking which model is best, a better question for this hybrid world might be which layer matters more to how you actually work, and who gets to own it.
A version of this post appeared on my weekly Substack of AI and digital trends, and transformation insights. To join our community of over thirteen thousand subscribers you can sign up to that here.
To get posts like this delivered straight to your inbox, drop your email into the box below.
Leave a Reply