The world of software today is in a state of enormous confusion.
So you're describing a bubble.
The expectations people have of limitless growth is not going to be sustainable.
How much would your workforce be impacted by AI?
There's no question it's going to massively change jobs and already in our company.
I'm honored to have Alan Trefler today who is the CEO of Pega Systems. You started the company yourself. I believe about 40 years ago. It's now listed on NASDAQ. And it's a multi-billion dollar capped company with offices all over the world. So, and you've just flown in from Boston. So, I'm really thrilled to have you on my podcast and get a chance to catch up with you.
Well, thank you, Richard. I actually like to describe myself as founder and still CEO, though, as you point out, it's been four decades, so I'm feeling pretty secure in that.
So, I mean, where do we start? What does your company do? How did it start?
The company started because I had been working as a software architect trying to plan out software systems for typically large New York banks. I lived in Boston but went every week to New York. And my boss who was a consultant sent us in to really try to figure out how to make both things at the time around money transfers and ATMs and other sorts of things were coming and we were deeply involved in trying to build those and I saw some opportunities in the way that we were building software in a whole new way that we could build it differently.
This is the early 1980s no
This would have been late '70s.
So you were a graduate. You studied computer science or something?
Yeah. So it's interesting because I went to university and I was recruited into the computer science department because I won a major international chess tournament as a sophomore. And we were trying to teach computers to play chess. And that was a really interesting problem at the time.
So you because you played chess, they assumed you'd be a good programmer.
Well, I think they assumed that I could explain to them how a chess master looks at a chess position and that at that era the computers weren't terribly fast. So, there was a strong belief that you couldn't play a game like chess with brute force that you needed to be able to really understand the logic and the sort of mental machinations somebody went through to figure out what to do. And I also learned of course to be a programmer and become part of my whole career actually.
So in the late 70s early 80s you were working in software and you spotted an opportunity and left the company that you're working for.
Yeah. The company I was working for was purchased and I needed to decide what to do next and I considered going back to university for advanced degree. I could have gotten another job, but I had this idea that I had approached my boss with and he had basically sort of rejected, but that I believed could make a lot of sense based on what I was seeing in the
Have you stayed in touch with him?
Yeah. He came back to me about a decade later and he said, you know, we really should have done that, Ellen.
So, what's an example of that, practical, single example of how someone would have used what you were doing.
So one of the things that I saw and was able to convince some other people to join me in was that many of these in particular banks but it also very true for governments and health organizations etc have lots of things you would consider a workflow. They have a new member of household you want to onboard or you have a product you want to sell that goes through a set of steps and stages and as you go through those you want to do certain things ask certain questions drive certain outcomes and what we did is we said we can codify these workflows we can make these workflows smarter than if somebody just wrote them in computer programming languages. And if we do that, it will do two things. One, it will make the workflows more effective. But the second, it will let business users, people who aren't very technical, understand them and even change them. And that was the original vision of the company. And despite all the massive changes in technology over the four decades, in many ways, that's what we do today, just for many larger situations.
Yeah. On a big different scale. But so what would be one example? If you've ever had a credit card dispute, you look at your credit card, you see something on the bill that you don't think you charged, the odds are about 80% that a PEGA system somewhere in the world will actually fix that for you. That was
So you just started with you and a friend,
Yeah. I started I went out on my own. The other two weren't sure, but after I was added and they realized I wasn't going to stop, they sort of joined in and we went in terms of bootstrapping something really creating an idea from scratch. You know, I think we're almost a poster child. We went nine months before we got our first customer and it was a very difficult period.
So you sort of financed from your savings and from friends and family. From family. I financed for money that had to be paid back. So, one of the first things I did is I took out a life insurance policy for half a million dollars because I had borrowed money from my mother and she was going to need to live on. I wasn't going to kill myself if it didn't work, but if it went badly and somehow I was unable to earn the money back over a number of years, she had to be paid back.
You know, how do you start winning customers? During the days I was the architect which meant I would try to figure out what the software should do. And then the second half of the day I was the salesman. So I picked four banks and went and called on them and told them what we were trying to do, why it would be better than what they were doing today and got them interested. And none of them of course would give us any money or buy because we didn't have anything that they could actually use. But as they saw it develop, we had a miraculous moment in the ninth month in which we went from having no customers to having the two biggest banks in the US as clients. And it was a pretty shocking transition from being starving to completely oversold
Yeah. Well, but with those nine months was that sort of expected because you were sort of building the software so you weren't really marketing, you know, you can't sell what you haven't got or did you already have a product?
No, we didn't have a product.
So really the nine months is developing a product before you can go and sell.
But we didn't wait until we had it to go talk to customers. I think it's important to understand whether the things you're talking to a customer about are going to be appealing to them. And so we would go we would say, 'Hey, we're building this. We're thinking of building this. Let me give you a little bit of a demonstration.' And it was a very different age in computer land because there was no cloud. We had to go buy a computer.
I mean, a big part of the capital we had we immediately exhausted because we had to get a machine that we could use that would be suitable for building this sort of technology. And the reality in business is typically when you need something nobody will give it to you. So all of the computer companies who subsequently were really happy to kind of throw things at us in the early days were only happy to sell things to us.
So how did you get your foot in the door with those banks?
Well, I had personally done some work for some of them. So I had a bit of a personal reputation and that helped leverage the meetings. I knew some people there and they were willing to meet again. It was not the case, however, that they were willing to go and buy something for hundreds of thousands of dollars from a company that had
No, no track record. So, they wanted to see things. But, you know, it's interesting. It was only many years later that I realized how much courage those initial buyers had to trust that we would deliver on the things that we said we would. And I would say that as a firm, the proudest thing I think that we feel as a company collectively is that those first two clients from 1984 are not just clients to this day, but they're huge customers of ours to this day across massive shifts in technology.
And you won like big customers to start with. So it wasn't like some early adopters who were just like a regional bank down the road where they'll give it a go.
We ended up picking and I decided the only way that we would pull this off is if we could focus our energies. And so we picked two massive banks and two regionals. And I wasn't sure which would be easier, but I knew somebody at each one of those for
And it turned out that we were kind of hoping we'd get one big bank or maybe two of the regionals have a chance to finish and show what we were capable of and then go back out to market and it turned out that both large banks within the span of like 48 hours signed up. So when you were introducing a new product in software as well, was there any faking it till you make it or did you suddenly like you've sort of overpromised a little bit to get the deal done and then we've got to get this software doing what it's supposed to do, we got to add this bit on, I've promised that, was there any of that or was it you very confident that what you'd already designed and had running was going to operate?
We were changing the design every week. We would get the feedback from the customers and it's interesting. Some organizations just can't operate that way. Some customers can realize that if they work with a more dynamic company and if they feel they can trust that company and if it works, they'll get something that lets them take a leapfrog. And those were the personalities of all the organizations we went after. We were looking for organizations that would be willing to put a stake in the ground because we didn't consciously overpromise. But the nature of these things is you just don't know sometimes.
And so you just don't know if it's going to take a month or two months to do. And so you do your sort of best guess. You can't be conservative and underpromise because you won't get a chance to do it. But if you overpromise and you miss,
It doesn't take more than one or two of those to make the customer say, 'Well, I'm not going to believe the next thing I hear.' So there's a very delicate balance that you need to go through. And part of the way of addressing that is to ask the question, and I think this can be true for a lot of businesses, a lot of industries, is to say to the customer, look, if I'm able to deliver this, will you buy it? And customers will kind of say yes. And then you say, well, would you be willing to put down a small deposit? And that actually helps you tell whether they're just being agreeable.
Yes, or whether they're willing to put in something not enough to keep it alive. Yes.
But enough to really validate the seriousness.
That's golden advice. So is there a way to prevent the not invented here when the in-house team of a customer will just try to do what you are doing themselves?
Sometimes they'll see what you're doing and they'll say, 'Boy, that's cool. I'd like to do that, too.' And it's even in some cases not even a matter of expense. It's sort of a matter of interest. What you have to do to combat that is you have to be able to pick a target that's hard enough that it will scare some people off and have a directional angle that creates anxiety that people say, 'Hm, maybe I could build the first generation.' But if these guys have been working on it for two, four, 10 years,
will their generations always be ahead?
And will that matter for the business?
Right? The other aspect of it is that if we can build it and sell the heart of what we do to 20, 200, 2,000 companies, then we can get an economy of scale and a level of insight from what those companies are doing that should keep the product both more cost-effective, but also more inspired and more able to keep up with what's going to change in the market. Well, you are, you know, Pega is now kind of the IBM. You know, you don't get fired for, you know what I mean? You don't get fired for
The world of software today is in a state of enormous confusion.
How do you see the technology market right now?
We're really at a there's a company called Gartner that rates technology firms. They talk about technology sometimes going through a peak of inflated expectations where people just expect everything from it and then a trough of disillusionment when sometime comes true and then usually with most technologies you find a place where there's a level of productivity you get etc. We're right now I think vacillating between wild expectations, mass delusion and a lot of confusion out there that is really, well at least to us it was evident that it was going to occur that I think the market is just now beginning to discover. I think Bill Gates said people overestimate what's going to be achieved in the next 12 to 18 months but they underestimate what's going to be achieved in five years or something like that.
And that's common and then there are some things of course where we keep thinking that in five years, how many times have we heard that fusion is five years away and at some point it will be it just
driverless cars, I mean we're getting them
We are getting them. I mean some of these things take they operate on their own schedule. But this whole new push with AI and what's going on in the market is creating enormous confusion and as a result I think there are some things that people should be worrying about that they really are missing.
What do you think is missing in the AI market right now? Well, we think when we saw the generative AI capabilities come out at the end of '22, we thought, boy, these are really interesting technologies. How would we want to apply them relative to our customers, etc. And I think we hit a really good balance. Now one thing we saw was that this use of the generative AI as you said was not always reliable and we operate in businesses and we operate in general with customers who'd like to be reliable or some cases by law have to be reliable. So we said how can we use the power of the AI to stimulate thought to help people reimagine things but then on execution figure out how to do something that will be reliable. And a good analogy here is let's say you wanted to open a restaurant you might hire a great chef, create a test kitchen, and that chef would go and distill down a set of recipes that you wanted to offer. And you take those recipes, put them in the recipe box, move them onto a menu, and you've got a highly creative business system, but operationally has the ability to scale. And when we wanted to model how we were going to use AI primarily, it was to say, let's use this AI to redesign the way the business process should work, to challenge us to think, do we need these steps? Do we not need these steps? How do we want to do things? Convert those into workflows. And then at runtime, when somebody actually wants to do something, find the right workflow, execute it, and not turn the world over to the language model at runtime, but use it when you're thinking at design time. Now, what's interesting is this is quite different from the way that a Microsoft or a ServiceNow or Salesforce have gone after this problem. They've been talking about just turning it over to the language model with the language model figure out what your business should do from some English language prompt.
And that has a couple of problems. The first of the big two
Is that it is not going to be predictable. You're not going to get a consistency of outcome because part of the beauty of the language models is they do introduce variation. Using that variation when it's trying to stimulate you to do good thinking that's awesome but you're curating it, you're figuring out what good is and what bad is. When you're actually running it, we think it would be nice to be able to operate off of as much as possible a standard set of procedures and only wing it when that made sense. The other problem which has just broken in the last six or eight weeks that these companies that were spending first they said billions, then they said tens of billions, then they said hundreds of billions.
You say we're talking serious money.
No, it's actually trillions of dollars being spent to build data centers
To run the AI and answer the questions that they wouldn't want to get paid back. And so for the first couple of years, these drug dealers began making it so that it was either free. You could open AI for free or for $20 a month, you can basically get pretty much all you wanted. And it was magical. It got better and better. Well, what's happened is they've now decided they want to go public, and to do that, they have to generate revenues. And all of a sudden dramatic price increases in what Microsoft and these other companies are charging for these AI tokens, which is the unit of measure they use to measure the amount of AI you use. And people don't really know how many tokens they're going to use
I mean, so it's just incredibly opaque. But it also can become, as we've learned, very expensive. Well, the only reason that's a problem is because they convinced the industry to go down these paths of reimagining how they wanted to handle a customer interaction or how they want to handle a situation from scratch. Every single time they did it, they convinced them that you don't have a recipe box.
That what you do is just a really good chef and he comes in and somebody shows up and he's going to design for you the perfect meal. You may not be able to have a price list on it. But you know, for some people in some cases in some situations that can work fine. But in terms of the basis for industry, I think there are some pretty serious problems if you don't have predictable outcomes and if you don't have predictable cost. So we've gone firmly down that path.
I'll just jump in there for a second because I don't you might know Demis Hassabis. Do you know Demis?
I know of, I was friends with him because of chess, and you're a chess player as well, but he's from here the UK and now he's head of AI at Google. I said to him, 'Oh, I like Gemini, his software, but ChatGPT is so friendly. It's so nice to you. It's kind of easy to use it.' And he goes, 'That's so bad for humanity. It gets people to overuse it.' So, are we overusing AI? Are we not accounting for the cost of it? Which is what you're saying.
Well, I think there's use that is adding value to people, and that's great. There's use that's just waste.
And rethinking the same thing over and over again.
Designing a new menu, the same menu every day.
Or designing the menu, redesigning the menu for every patron, right,
as they walk in the door, as opposed to saying, 'I'm going to have some inventory of things that reflect what we're capable of doing and have learned to do.' Well, that's not AI's fault. That's the people who harness it not using the right judgment.
well, I would say my examples are really mostly business examples
here as well. If you want to develop a relationship with ChatGPT or Gemini for that matter I wouldn't fault you for that but that also doesn't destroy the planet. It's the heavy-duty uses, the ones that happen hundreds of thousands or millions of times in a week that are the ones that are really driving it. When in fact, if they didn't have to use it, if you could just think about it once, record it, re-execute, that seems to me to be not just a more reliable way of doing it, but it certainly is a much greener way. So the world we're in now with AI, there's a lot of little companies pursuing particular niches and there's obviously the big guys. Who's going to win?
Well, I think different companies will win in different ways. I'm not sure that starting another software company is what the world needs this instant.
Yeah. Well, I was speaking at an AI conference two months ago and flew to San Francisco and been there for a little bit and was driving down the 101, which is this big highway that goes from San Francisco to Palo Alto and it was mind-blowing. There was billboard after billboard of AI companies, XYZ.AI and ABC.AI and whatever, 80% of which I've never heard before and I'm in the business. And there are just so many. Part of it is that the emergence of cloud computing has reduced the entry cost, but that will not in and of itself make those companies successful. It will just allow a lot more to sort of jump into the pool. To be successful, you have to have something that's distinguished. Now, the other end of the spectrum are the companies who are not taking advantage of ease of entry, but are taking advantage of being able to raise and spend vast amounts of capital to create these enormous and increasingly large models. And to be candid, a lot of that is just unsustainable. The amount of money being invested, the amount of capital being spent, the amount of energy being consumed, I think we'll get to a tipping point and people will say, 'Well, we've had enough.'
