Thrown away by one, made a legend by another: AI adoption in companies

The maker of the wah-wah didn’t know what the pedal was for. A user found out, after others had thrown it away and he kept experimenting. AI adoption in companies is following the same script: used as it comes out of the box, AI makes everyone sound alike. The edge is in the people who dare to use it differently, and in how long they have been doing it.

A few days ago I listened to a podcast called Model Behavior, from Substack, with Rick Rubin. Rick is one of the great producers of modern music, the man behind Johnny Cash, the Red Hot Chili Peppers and the Beastie Boys, and he has a view of creativity I admire. The story was about the wah-wah, one of the first guitar pedals, and it had been told to him by Johnny Echols, guitarist of the band Love in the sixties.

The wah-wah: the pedal nobody understood

According to Rick, Vox had just launched the pedal and handed it to a handful of professional musicians to try. One was Echols, a respected session guitarist. Another was a man named Jimmy James, who was technically Little Richard’s driver and played whenever the band was short a musician. The sales pitch was this: “with this pedal your guitar will sound like a trombone.”

Echols threw it in the trash. His reasoning was flawless: if I want to sound like a trombone, I’ll learn the trombone. A year later, a friend called him from San Francisco. There was a new guitarist just in from England, like nothing anyone had seen, named Jimi Hendrix, and he was playing the next day. Echols drove down. The lights went out, and the man who walked onto the stage was Jimmy James, the driver.

Echols says it plainly, as Rick tells it: Hendrix’s whole sound was built on the box he had thrown away. Vox had no idea what the pedal was for. It sold it as a way to imitate a brass instrument. A curious user, not the maker, found out what the tool could really do.

I can’t think of a better description of where we are with AI.

Who finds out what a new tool is for?

Today’s AI vendors are Vox. They sell the tool with the wrong pitch, because they don’t know what it’s for either. The promise is “this will make you more efficient.” Executives try it with that expectation, ask it to make their company sound like a trombone, get a mediocre result and put it in a drawer. “This doesn’t work in my industry.” They are Echols. They’re right about the short term, and they’re making the mistake of their career.

This isn’t just a rock anecdote. Eric von Hippel, a professor at MIT, has spent decades studying where innovations come from, and his conclusion is uncomfortable for any manufacturer: often they don’t come from the people who make the product, but from the people who use it. In a 1976 study of 111 innovations in scientific instruments, he found that about 80% of those that gave users a significant functional improvement were invented, prototyped and first field-tested by the users themselves, not by the companies that sold the instruments. The maker supplies the tool. The user finds out what it’s for.

And that takes time. Hendrix didn’t become Hendrix the day someone handed him the pedal. He became Hendrix in the year Echols spent playing the way he always had. You can’t get that year back. Nobody can buy it later. The advantage of experimenting early isn’t having the tool sooner. It’s having the craft sooner.

Why does everyone end up sounding the same with AI?

A recent study explains why so many people stop at the trombone. Anil Doshi and Oliver Hauser published an experiment in Science Advances in which a group of people wrote short stories, some with ideas suggested by AI and some without help. The stories written with AI were rated as more creative, especially those by people who were less creative to begin with. But they had a problem: they were more similar to each other. AI raised the average and flattened the differences. Used as it comes out of the box, the tool makes everyone sound like the same trombone.

Hendrix didn’t sound like Hendrix because he used the pedal. He sounded like Hendrix because he used it differently from everyone else.

Rubin puts it in a way that clicked for me: with any new instrument you first make bad sounds, and then someone figures out how to play it well. It happened with the drum machine, the sampler and Pro Tools. Each time there were serious musicians saying it was no longer music, and each time someone proved them wrong from the stage. The problem is that we are judging AI by the bad sounds. It’s true that most of what it produces today is mediocre. It’s true that automated content with no point of view is flooding LinkedIn. But concluding from that that the tool is useless is confusing the state of the instrument with the state of the musician.

Why is AI adoption in Latin American companies urgent?

In Latin America this has a particular urgency. Our companies rarely lead technology adoption; historically we arrive when the manual has already been written. With AI the manual doesn’t exist yet, and for the first time in decades that puts us on equal footing with any company in the world. A manager in Santiago has access to the same tools as one in San Francisco, at the same price, on the same day. The one thing they aren’t guaranteed is the year of practice.

How to start using AI in your company: one process, three months

I’m not saying you should believe the vendors’ pitch. I’m saying you should ignore it and try anyway. Pick one concrete process, only one, and spend three months seeing what happens when you apply the tool in ways the salesperson didn’t suggest. Most of those experiments will sound bad. One of them may sound like nothing you’ve heard before, and that’s the one that gives you an advantage your competitors can’t buy when they finally decide to start.

If you want a bounded way to do it, a proof of concept is built for that: one process, real data and a result you can measure. And if what holds you back is doubt about your data, I wrote about that in the doubt that costs more than the risk. I also covered, in AI isn’t the box, it’s the port, why the value is in how you redesign the process and not in the tool.

Echols was right about everything except the one thing that mattered. The question isn’t whether AI does the job you gave it well today. It’s what your organization will know a year from now that the one who kept the pedal in the drawer won’t. The pedal is on the table. Nobody knows yet how to make it sound good. That is exactly the reason to step on it now.

FAQ

How do you start using AI in a company?

Pick one concrete process and spend three months trying the tool in ways the vendor didn’t suggest. Most experiments will go badly, but one may give you an advantage your competitors can’t buy later.

Why does AI make everyone sound the same?

In an experiment published in Science Advances, Anil Doshi and Oliver Hauser asked a group of people to write short stories, some with ideas suggested by AI and some without help. The stories written with AI were rated as more creative, but they were more similar to each other. Used as it comes, the tool raises the average and narrows the differences.

Who finds out what a new tool is for: the maker or the user?

Often the user. In a 1976 study of 111 innovations in scientific instruments, MIT’s Eric von Hippel found that about 80% of those that gave users a significant functional improvement were invented, prototyped and first field-tested by the users themselves, not by the manufacturers.