Automation and AI

Tullio Siragusa (00:13):

Good day everyone. This is Tullio Siragusa with Tech Leaders Unplugged. Let's get unplugged today, today's Tuesday, May 23rd, 2023. We're talking about automation and ai specifically asking and answering the question, does AI help with automation? My guest today is Tony Sumpster, is the CEO of Worksoft. Welcome to Tech Leaders Unplugged, Tony. Good to have you.

Tony Sumpster (00:39):

It is great to be here, Tullio. I'm looking forward to our conversation.

Tullio Siragusa (00:44):

Before we get started and dive into this particular conversation, there's obviously a lot of talk about AI for different uses, so this is going to be really good. Talk about automation. How did you get here? Tell us a little bit about your journey and about Work Soft. How does this topic tie in?

Tony Sumpster (00:59):

Yeah, so both <laugh>, both stories along, so, so but I'll keep it short. So Worksoft been around for, around, you know, 25 years. It has a very rich history in large-scale complex test automation really around the SAP and Oracle spaces done just a huge amount there and fortunate enough to land some of the largest corporations on the planet. So we, you know, really good there. The company was founded on a basic and very simple idea, which was how do we get business involved in testing our core applications? And along the way invented, not invented, but certainly created a no-code platform that became really good at large-scale, complex change. And that's where we started. Along the way, we learned some things about ai but I'll get to that in a bit.

Tony Sumpster (01:59):

For me, I'm a software guy. I, I, I've been and done many things. This is my second time as a CEO in the private equity world. Been a workshop for three years, and we're on a transformational journey. And that journey is to take automation where our customers need it, which is in three areas. One is how do we understand business processes and how do we do analytics around it? Why do that? Because we want to do pre-production or test automation better. And then the third piece of that jigsaw, if you will, is really to make sure that we can reuse the assets that we built for test in production. And people call that RPA today, but, but, you know, process understanding and discovery test automation or pre-production automation and production automation. And that is that world my background, again, software, 20-something plus years, maybe longer, but we're not talking about it <laugh>. But anyway, so, so just a bit of background for me.

Tullio Siragusa (03:03):

All right. Thanks, Tony. Thanks for the introduction. Let's dive right into this topic of AI and automation. The question we're asking is, does AI help with automation? Now, it would seem obvious that it does, but I'd love to get your thoughts on where are the better use cases today for how AI has evolved, at least as of today.

Tony Sumpster (03:25):

Yeah, and so I think it's wow, it's a very interesting time. You, you, you find yourself sometimes in your career when the whole you know, playing field that on starts to shift. We saw it in the movement to the internet, so around 2000, that was pretty interesting. We've seen, you know, things like Web three, we've seen a whole bunch of other areas, and right now we're at this space where AI has broken out. Now if you talk to the AI researchers, they'll tell you, we've been on this journey for 10 years, and that's all good, and for the rest of us, we're just catching up. And some of that's true but if you look at things like chat, GTP and its variations, you know, three, 3.5, four, all those different types of things, what we're starting to see is in the large language models, and you see this as L L M often, is that we are able to consume and train data sets on vast amounts of data.

Tony Sumpster (04:28):

What does that mean? It means that we can ask questions or prompts if you will. We can ask questions of these large data sets to give us information back. And so there are many places where this is going to work. So think of support calls. I can, if I train my model correctly, I can ask a support system to give me the answer to how I do something. Training is another great example of if I have my manuals and they've been consumed or my videos or whatever happens to be, and I can ask questions about how I do something on the fly, I can make my users more effective about the way they do things. There are places where if you are a knowledge worker, your world is going to be very, very interesting for three to five years as these models are going to come to more accurate answers than a human being will do. One that particularly interests me is transactional lawyers. You know, it's a pretty set in a pretty safe set of instructions, rules, the way that the law works. And if you're in that sort of stuff, I think you're going to have to, you know, reskill yourself about the way you do things. So, lots of use cases coming down the pipe that is, you know, really going to encourage people to be more effective in their roles and is going to create a whole, you know, a new set of roles as well.

Tullio Siragusa (05:57):

You said something interesting about the legal profession, and it sparked the idea that that's a highly regulated space. So are financial services, and healthcare, they all have some compliance requirements, some of them more restricted than others. Do you think automation will enable less errors and more, you know, the ability to be compliant with certain laws, et cetera? And are you seeing use cases that, you mentioned legal where that's being done today?

Tony Sumpster (06:28):

Yeah, I mean, I think I was on a panel last week in New York, and we were talking about AI and automation. There was this you know, idea about this generative ai. There were speakers from United Healthcare. There were service providers, big service providers who service a, you know, a good number of vertical industries. And the reality is, is that all, almost everybody sees the ability to give large language models and the models that will come next guardrails, all right? And those guardrails can be rules that can, that can fit into a regulatory framework. They can be rules around the way that we ethically say something. If you think about large language models today they're pretty much without guardrails. You stick stuff in, you train it, you ask for a, you know, question, how to make a nuclear bomb.

Tony Sumpster (07:24):

Well, yeah, probably shouldn't give the answer to that question, but today, if this trained on that sort of stuff, you could. And so what's happening now is the, the, the models that are to come are going to have guardrails around what it gives, how it gives the information, and what happens with that type of information. And so anything is regulatory compliance, anything in those sorts of areas. They will be the first of the new set of ai, if you will, if you can believe this, maybe a year from now, we'll have people who are just doing legal. They'll have the regulatory frameworks, they'll have all, all the guardrails in place so that when you ask a question, it's legally sound, it would stand up in court, all those sorts of things. Now, whether that's a year away or not, I'm not a lawyer, so I can't tell you that, and it probably needs to go to a court for precedent, but all these things are very, very possible.

Tullio Siragusa (08:21):

Yeah, interesting. I can imagine asking a platform a question on a generativity ai, a question gives you the answer and then gives you a very long disclaimer, <laugh>, so <laugh>, this kind of thing. Yeah. Well,

Tony Sumpster (08:36):

So that's true. One of the things that came up in ChatGPT 4, was, the ability to have, where… did the source come from? So in the previous versions, it didn't say where the source came from. So you've got an answer, but, do you trust it? Do I mean, you know… here's the thing about AI. Do I trust the data that I'm being given and giving things like sources? So here is my answer. I got this from here, here, here. You know this is a health-related question. I got it from the Mayo Clinic. All right. The… I believe the Mayo Clinic, if I read it on a website, I would believe them because, you know, they're trustworthy. And that sort of thing is going to start to underpin the, the broad sets of information that we're going to get in the future. And I think, you know, as we get onto how automation and AI work together, you'll start to see the training, the guardrails, all that sort of stuff come in. And we're starting to see that now in the broad context of general ai.

Tullio Siragusa (09:39):

Let, let's talk a little bit more about the automation piece. I know that at least for the past 10 years, maybe 15, we've adopted machine learning in the context of, for example, MarTech. But it requires massive amounts of data in order for it to do the right subset of permutation is basically thousands of AB tests, right? That is learned based on input. And through that, it informed the ability to execute a campaign. Some of that has been automated, and some of it hadn't, but definitely provided the capability, for example, that you wouldn't normally need an Army analyst to do, right? So, yeah, I'm just curious right now, most of it is very informational. You get to the information quickly, you can generate content quickly, and you can analyze information quickly. But where are the uses when it comes to automation? You know, how are you guys putting it into play, with automation? Maybe you have some use cases.

Tony Sumpster (10:36):

So, great question. So, we have been working on AI machine learning for probably the last six or seven years. The, the, in, in our core automation, we have, you know, dynamic learning you know, relearning. As objects change in an application, as, as the flow of a process changes, we can learn that we can adapt our tests you know, one of our clients uses our technology, to do 110,000 tests a night. You can imagine that what happens in that particular case is that if an object a, a social security number of field changes and the ischemic behind it changes, you really want to ripple that through all of the tests that have that particular object. The other place where we see huge amounts of things going on where, where AI is, is being super helpful, is around a problem we call the process visibility problem.

Tony Sumpster (11:39):

And that is, most organizations don't fully understand what their processes are. And the way you used to go about it was you'd get a bunch of consultants and subject matter experts in a room. You use VIO or graphics or something, you'd build models, you'd, you know, map it out. Or if you are even more old school, you get some brown paper and some sticky notes. And your human beings who are the <inaudible>, they would put things up on a board and you, you would come out with a view. Now it's unheard of, but apparently I, you know, humans are fallible. It's, I know it's hard to believe, but apparently, we don't, we…

Tullio Siragusa (12:16):

Don't say <laugh>.

Tony Sumpster (12:17):

I know, I know. It's you know, it's hard to say. So, what we've been working on is how can I get a way to collect the way that human beings do things. Mm-hmm. And we have some great technology around process capture, which uses machine learning to be able to help us understand what is going on with multiple sources that are capturing what human beings are doing. There's some great technology in the process mining space, which is a, you know, where we correlate data together and we use machine learning and AI to look for a variance of processes for, you know, how far something is going, whether something's, you know, tested or not, whether something is automated or not. So, that process visibility piece is where we're seeing the strongest growth in our business, because customers want to know what, not, what is my process ideally, because when we designed it two years ago, the world changed and covid changed. We bought a company, we've introduced new products, but actually what is really going on? And help us do that because then we will become more efficient in what we do. And that's just one example of how AI today is really changing the game. We're going to get more efficient if you just took it from a test automation perspective or in production, I'm going to know more at the start of where I'm going to automate something or why I should automate something that we've ever done before. So I'm going to be better than ever before

Tullio Siragusa (13:47):

This topic. Thanks for sharing. That begs the question, and I can't help but ask the question because what, what it sounds as, so for example, going back to the MarTech example, you can enrich data with external inputs and essentially between your contextual data and your static data and the external inputs, you can massage it and come up with some better, better outcomes. But it sounds to me, in order to apply it to processes you know, how does it learn what's going on, for example, in the industry and what other companies are potentially doing that's working better? And I can't help to of this, you know, where you have that transparency and smart contracts and the ability to share, you know, amongst companies that could enable getting there faster and improving processes, or is that not necessary?

Tony Sumpster (14:41):

So, so I think I think you're going to end up with two, two types of ai, right? You're going to end up with this generalized base of ai and then you've got to end up with private models that will be right for your organization, right? So if you are, you know, if you are Nestle or Microsoft or Google or, you know, whoever the, you, you know, your data set is, is so rich because you're in so many countries with so many product types and, you know, so many skews. Your, supply chain is enormous. And your data set, therefore, is huge. And I think for those sorts of organizations, their competitive advantages will be building out models that are specific to them. I think if you are you know, a small-town fabricator in downtown San Francisco you can't afford to do that.

Tony Sumpster (15:31):

So you're going to look to companies who just do benchmarking on ai and yeah, blockchain could absolutely be part of that. You know, smart contracts could absolutely be part of that data set feeding in saying that, look, you know, for a company of your size with this revenue, you know, your time to market for a new product should be x your, you know, movement of a product from ideation through to sitting on a customer's desk. You know, for the top quartile is, you know, five days or whatever happens to be. And so, you know, we're definitely going to see two different things turn up generalized huge amounts of data with big players and those who will put the foundation in place. I think the internet thinks Wikipedia and then there'll be people who've got very private large language models or whatever the next generation is who will have just their specifics with the guardrails in place or the compliance, the regulatory stuff. And that will give them the data that they need. And then they'll be third parties who will sell that data in the same way they do today.

Tullio Siragusa (16:44):

It's interesting what you're bringing up, this idea that most likely the way AI will be most effective is truly decentralization and specialization. I I kept thinking like, that, when it comes to QBRs, for example, they're very time-consuming to prepare QBRs for a client. And if you'd want to do it right, especially if you not only want to present performance for the past quarter but also provide some intelligence about here's where you stack against other companies in the industry in terms of the maturity of adoption of technologies or processes against other companies in your industry, to be able to provide insightful, valuable data to a client to guide 'em through what they should be focusing on. That's not an easy thing to do. And I kept thinking, man, this would be so awesome if there was an AI module, you just plug it in and you put in your currently your client's parameters, how things were done, the metrics from the last quarter, and you ask it to give you a recommendation based on what's going on in the industry IT companies in the similar size and space and you know, where they are in terms of their maturity level as an organization.

Tullio Siragusa (17:51):

Is that kind of what you're talking about, where eventually we could have these modules where they could be applied for a number of business applications where it's not just one big model and you kind of have to customize it. It's really, it's really like that's ultimately going to be a bunch of products skews if you will, that have different uses. How is that going to be managed in your thoughts?

Tony Sumpster (18:16):

Yeah, so, so I mean, we're going to find ways of, of, of connecting specific specialist models on top of this. This sort of, you know, foundational layers where I think about it. We don't see that today, but there will be this big foundational layer to be managed by, you know, the Googles, the Microsoft, those who've got the other resources to do it. The event I was at, the HFS event, that I was at, within a month or so, they're going to release a ChatGPT model, a large language model on all of their research. Now imagine that you're building your QBR. There'll be some AI that will clearly pull data from many places. They'll put it into a you know, a PowerPoint or you know, something similar. And you'll say, okay, but in this, I need to have this specialized data, which I pay for through a subscription, and I can, you know, I've got an API that sends a prompt to, you know, my analyst you know database effectively, and it's going to say, hey you know, my customer.

Tony Sumpster (19:23):

So, so the prompt is, you know, give me the latest benchmarks for fabricators or, you know, plastic fabricators in, in California you know, give me these things I need to know that's sort of doable. Now, <laugh>, I mean, scary later. You know, we're going to find people who just find a way, to get pieces of data and then bring it together in a consolidated fashion. Because for us talking to customers, it's not about the data, it's about the conversation. What's the data telling us? How do we have that rich conversation about how we make the customer better? And for us, that's around process understanding, using machine learning to make them better so that they can move faster so they can innovate faster, they can be more competitive, they can you know, reduce their cost so they can, you know, get products out to market. But all of these connection points, they don't quite exist today in a way that all of us could use it, but we're not very far away at all.

Tullio Siragusa (20:25):

Interesting. I, I can't help but think about you know, some of the large software companies have marketplaces where you can acquire certain modules and capabilities. It sounds to me as though for startups at least, you know, really specialization and niche for AI is going to be key, and then hopefully those will get acquired or they'll be part of a marketplace someplace. Is that where you're envisioning? I mean, everybody would never imagine they had 30 or 40 apps on their mobile phone, you know, 20 years ago. It's like, there's no way I'm going to have an app for that. You know, I just go on the web and then we all have a bunch of apps, you know, 20, 30 plus different apps for different uses. There's not a single app that does it all. It, it's kind of, is that kind of the, where we can borrow the success as we move forward with AI applications that are specific to certain use cases?

Tony Sumpster (21:17):

Yes.

Tullio Siragusa (21:18):

Is that what I'm, I'm hearing? Yeah.

Tony Sumpster (21:19):

Yeah. I mean, I think you'll bang on, I think I think your ideas there are exactly where it is. If you look at the amount of funding going into ai, you know, I sat on an investment dinner with a bunch of AI VCs, I don´t know, six weeks ago. And you know, the number of things that they were betting on was pretty much down in this road, which is to get people very specialized around the data, find the data to do training on, and then being able to work at how to, how to find a way to monetize that data. And I just think there are just so many methods that you could use. I mean, I think, I mean, again, I'm super excited about what the possibilities are. We are right at the beginning.

Tullio Siragusa (22:01):

Yeah. I'm curious whether the winners will be the big platforms, or those platforms transformed into entry points for different platforms. You know, like will Microsoft SAP org or what have you, will they be basically, you know, foundational entry point to the very specialized modules that you could either license to their marketplace or if they've acquired those companies licensed directly through them? Or is it going to be more of a, you know a collection of a bunch of independent software companies that you end up adopting to get what you need, get to that to be seen? Right. <Laugh>,

Tony Sumpster (22:40):

I wish you the answer. I mean, I think, I think…

Tullio Siragusa (22:42):

That's the million-dollar question or the trillion-dollar it is.

Tony Sumpster (22:45):

It's, I think you know, one of the things about large companies is they struggle to innovate at the pace of other people doing, you know, do that. If you listen to some of the, you know, the, the talking heads, if you will you know, Yan Lekun or you know, Geoffrey Hinton, you know, they will tell you that yeah, there's huge progress on these big models, all the rest of it. But they'll also tell you that you know, that small organizations, small teams can move incredibly fast. And so I don't think we know the answer. I think there are some new molds around decentralization, which you mentioned earlier, which I think to be super helpful. I do think that there is a place for a foundation layer, which, which I think is going to happen. And again, I, you know, in this bottle, this Wikipedia with things on top. So we'll see. I, I, I don't know how it's going to work out. What I do know is we're in a fantastically interesting time and the AI is going to make a big difference, certainly to our automation, the way we help customers. But I think overall to society, I think we're in an interesting spot.

Tullio Siragusa (23:49):

Tony, it's been great to have you with me. This today. It gave me a lot to think about. Hopefully, those watching will get a lot to think about as well in terms of the future of AI and where the opportunity sounds like a great time to be in the AI space, if you want to be a startup, whether you go at it alone or build a module that some big platform's going to acquire, that's, that's the future. I mean, it seems pretty clear in terms of where, where all the attention's going to go. Thanks for being with us again. Just stay with me as we go off there in a second. I want to announce what we got coming up. We've got a few more guests coming up. We've got one Thursday; we have to move tomorrow's appointment because I'm hosting a brain date in Ascend here in Ascend Summit in here in San Francisco.

Tullio Siragusa (24:32):

So, we got Thursday with Amanda Blevins, who's the VP and CTO of America for VMware. We're going to talk about beyond multi-cloud, and then we're going to talk Friday with Piyush Malek, who's the CTO of Veridic, about Decoding Tomorrow's Digital Frontier. And then one more coming up next week on Tuesday will be with Lisa Thill, who's the Managing Director of Data and AI at Launch. So very interesting guests. We're going to continue to see what we can learn and thanks for being with us today and enjoy the rest of your day, wherever you might be. See you again Thursday at 9:30 AM Pacific.