This is a transcript of episode 285 of the Troubleshooting Agile podcast with Jeffrey Fredrick, Douglas Squirrel, and special guest Ulrik Lehrskov-Schmidt.
Software has become commoditised, but that doesn’t mean there’s a race to the bottom for efficiency. An accessible foundation means what you build on top can be much more ornate.
Show links:
Listen to the episode on SoundCloud or Apple Podcasts.
Introduction
Listen to this section at 00:11
Squirrel: Welcome back to Troubleshooting Agile. Hi there, Jeffrey.
Jeffrey: Hi, Squirrel. You know, I’ve got to ask you a question. I noticed this tweet of yours that read:
Software is becoming massively commoditised (Shopify, Salesforce) but that doesn’t mean there’s a race to the bottom for efficiency. An accessible foundation means what you build on top can be much more ornate–“prompt engineering” being the latest example.
Jeffrey:What did you mean by this? What do you mean by ornate? I’m very keen to know what you what’s the example you had in mind?
Squirrel: Well, I had someone claim to me last week that really everything had been invented now, that we really had all the major problems of software solved.
Jeffrey: Oh, really? That’s fantastic.
Squirrel: That’d be great if it were true. He said, “Look, we can do so many things now, that the basic problems a typical business has are ready solved. We have software that does it. We know the right ways to do it. What we need to do now is really focus on doing those things efficiently and well rather than coming up with anything new.” I said, “I can’t possibly disagree with you more, but we only have half an hour. So, um, I’ll respond another way.” So the argument starts from a good premise. If you look back, you and I, Jeffrey, can both remember when you needed a team of highly-specialized engineers, a fully specced-out, well-rigged rack in a data center, and a lot of specialist knowledge and permissions to set up a website where you could take payments. This was a hard problem in 1998. The world was not set up for someone to take a credit card over the Internet in a secure way, turn that into money, and then ship something out to someone as a result. Today, I can do that in five minutes with Shopify, literally. All the problems have been solved. Everything is commoditized to the point where it requires no skill whatsoever to use it. And if you stay with that application of software, if you stick with the idea that software exists so that we can sell basic stuff online, then yes, absolutely. There’s nothing more to do. All we need to do is build ever more efficient and easy to use versions of Shopify. And you could say the same thing for setting up a website that promotes your ideas. You could say the same thing for setting up a social network. All of these things have been solved and we don’t need to do them again. My interlocutor was correct to that point. The thing is that there’s lots more that we do now, and I referred to prompt engineering in the tweet. There’s lots of other examples. There are things that we can do now that we couldn’t even have imagined in 1998. An example is having a supercomputer in everyone’s pocket that can connect to an even bigger supercomputer cluster based on fancy GPU technology that certainly didn’t exist in 1998. And that piece of software that runs, say, GPT-4, can have a reasonable conversation with you and help you to practice and improve your conversations, to have a difficult conversation, which is what you and I keep talking about. There’s this other idea that you can use it to write your emails, and I think that’s stupid, but that’s probably another podcast. That infrastructure has been commoditized to the point that you can connect to it, but there’s still so much more that you can do on top of it, which is why we have this new idea of prompt engineering. How do you ask it questions? How do you talk to it? Not in a programming language. This supercomputer cluster that your supercomputer in your pocket can talk to, but how do you interact with it in the way that gets you the best answers, the most creative pictures from Midjourney, or the best research from GPT-4. That’s just one example of creating something even more ornate on top of an abstraction that we couldn’t have imagined in 1998. And for example, I pay for ChatGPT. I paid for it in a completely routine way, right? I paid for it using a credit card on the Internet, and it’s charging my card every month. And that’s great because I get to connect to the supercomputer cluster of the future.
Distinguishing Past and Future
Listen to this section at 04:55
Jeffrey: It’s a great example. I will say when I read it, the word “commoditized” just triggered something. Now you know that I love mental models and applying them all kinds of places. The model that came immediately to mind for me was Wardley mapping, link in the show notes. Simon Wardley lays out this idea about how software progresses over time, and he defines this evolution from genesis, to custom built, to product or rental, and then commodity or utility. He very explicitly makes the point that once something becomes a commodity or utility, then there’s new value built on top of that. He lays out this ten point formalized journey to get to point ten, “higher order systems create new sources of worth.” Exactly as you said, there’s still new problems. Sure, in the commodity space you’re looking for efficiency. But up in the genesis space, this new area that you’ve built on top, you have no idea what’s going to come out of it. Efficiency is entirely the wrong mental model. That’s kind of overall the message of Wardley mapping: you need to know where you are on this map and anticipate how things are going to evolve and make sure that you are approaching the challenge appropriate for where you are in the life cycle. Don’t be trying to treat something in Genesis the way you treat something in commodity. If you’re trying to be the most efficient person with your large language model, you are going to lose to people who are being less efficient but more creative, because we don’t yet know what problems are going to be solved by it. On the other hand, if you’re there trying to be the person who is the most creative credit card payment processor, that’s probably going to be much more challenging because you this is a pretty well defined space and efficiency is probably going to to dominate. So this is seemed like just what you were describing was a perfect mapping of this off-the-shelf model, which I strongly recommend to people to learn more about and apply to their actual world. It’s funny to me how it captured your view, and also why that person who thought the world ends at commoditization was wronpg about software. Really this applies to all technology: one of the canonical examples Simon Wardley uses is the invention of radio and how radio gets moves from invention to become a product—where people have furniture in their living room and that’s the radio—and then commoditized where it’s given away in your car, but there’s new value created on top of it. So really interesting to see that echo in this model coming up with prompt engineering, which is the new hotness out there.
Squirrel: Absolutely. And we would expect that over time prompt engineering is going to have quite a short life. If you’re just entering university or you know somebody who is and they’re thinking of studying prompt engineering—there must be somebody offering a course in this by now, I can’t believe it’s not happening—I’d say consider a different career, consider theoretical computer science or something, because this is going to get commoditized very quickly. The evolution is very fast, much faster than you and I saw with the evolution of credit card payments, which took maybe even 15 years before we had something like Stripe that can just do it for you in a couple of minutes. It seems to be going much faster. So my prediction, which I’ll leave listeners with, is that we’re not going to be prompt engineers for very long. That will become a known quantity and there’ll be something else that it will be in genesis that we can’t think of today, and that will be within the next few years.
Jeffrey: I will say prompt engineering will be like having good Google skills, right? If you are good with Google, you can find things that people who are not so good with Google can’t. So I think there’ll be a lot of things that you can do just fine in a kind of naive way. But I think there’ll still be returns for prompt engineering for several years now. That’s my prediction. We can check back in three years and tally up who was a more on the mark on that one.
Squirrel: We’ll be an episode 400 or something. We can check then.
Jeffrey: That’s right. People who learn it now, I think there is the ability to be part of something where we don’t yet know where it’s going to go. If you’re looking to innovate, now is a good time to be looking at that new frontier where there is so much uncertainty because that’s where innovation can happen is in the areas of uncertainty.
Squirrel: Absolutely. Thanks, Jeffrey.
Jeffrey: Thanks, Squirrel.