The Problem With Enterprise AI

Key Takeaways

This next segment is brought to you by our sponsor, Outsystems, an AI platform for developers. Companies like Toyota and Ro Pharmaceuticals use Outsystems to build custom apps and AI agents across their organizations. I spoke with CEO Woodson Martin about what the company does, how he evaluates ROI as businesses look for real AI impact, and also how he sees the race for the best coding models unfolding. Here is that conversation. WOODSON, WELCOME to TITV. It's great to have you here. >> Hey Akos, great to be here. So, Outsystems uh is an AI platform for developers and I'm hoping you can help us understand how it's different from all the other developer platforms that are out there that are working with AI that we hear so much about. >> Yeah, outsystems is the AI development platform built for the enterprise. So companies like Toyota Motors and Ro Pharmaceuticals, Axos Bank, they use out systems to rapidly build custom apps, agents and for missionritical core functions, but also to modernize kind of legacy processes with AI and Agentic solutions and manage and govern that full life cycle of a portfolio of apps and agents all on a unified platform. So Al Systems is the only AI platform that's unified, agile, and enterprise proven. Now, one of the topics that we've been talking so much about on this show are the barriers to adoption that enterprises see with AI. And one of the things that has come up this week in particular is how do you measure ROI when you're not even really sure what to measure and how to sort of assess the effectiveness of AI after you buy it? I wonder how you think about that at your company when you're pitching your platform to customers. >> Yeah, I would say you know maybe at the headline level this year feels very different for enterprise AI investment. You know as organizations are kind of shifting from a mindset of experimentation uh toward accountability for real business impact. It's kind of changing the story in the way our customers are thinking. And I think one of the things that's become clear through the experimentation so far is that you kind of have to expect failure early and plan for it because most AI agents fail when they get into production. And I don't mean they fall down and break. I mean they don't deliver on the promise. Uh and that's kind of normal. It takes iteration, experimentation, and a lot of change to get things exactly right. you know, the work that you want AI agents to do versus the work you really want to rely on the human in the loop. That's kind of hard to predict when you're first designing one of these systems. And so, you need to be able to experiment. And that's why you got to really kind of treat AI as a system and not just a model. So, so, so you're saying you're saying, you know, don't expect the thing to work out of the box. I mean, you know, it's going to have to take a couple iterations of of using the tool and and training your people on the tool. >> The whole idea is these are learning systems and they rapidly evolve and change, right? Part of that change happens because the models learn, but a lot of that change happens in the enterprise because we're improving the context engineering. Um, or we are evolving the way that users, maybe our customers or our employees are interacting with the system because we're driving change. We're trying to drive behavioral change. We're trying to drive change in outcomes. And so this whole systems evolve fast, but the whole systems not just the LLM. It's not just the reasoning of the agent, right? It's also everything else. It's the data, it's the workflows, you know, it's the UX of the applications. Maybe those are mobile apps for your customers. Maybe it's web apps for your employees. And all of those things need to be able to evolve quickly uh in an enterprise to be able to iterate to get toward that ROI. And that's why platforms are so important as part of the mix here uh in AI and that's really a difference that we're seeing. How how long though? You know, this is a question that that we ask people is okay. So, it's it's going to take a while and we've had some people say, you know, uh broad adoption is still 5 years away, right? And I I wonder, you know, just when you're thinking about ROI with your clients, I mean, surely they must be asking you, well, how how long should I, you know, wait to expect to see that? >> It's a great question. Like we're super excited about the pace at which customers are starting to see real value. One of the first customers to work with our agent platform, we call that agent out systems agent workbench was a company called Travel Essence. They're based in the Netherlands. They run luxury travel trips. Um 3 weeks from the very beginning of the project to the achievement of ROI. An incredible story of innovation and pace. But really what they did, they didn't reinvent the process. They have applications built on our platform that they've used for years to manage uh the work of their travel planners uh in interacting with their customers to put together these custom holidays. That was a process that would take them 2 to three hours per customer to research destinations, book hotels, restaurant reservations, etc. and they've built a series of AI agents now on our platform that use a variety of different LLMs on the back end. Uh, you know, they've got a booking agent. That booking agent makes the reservations. They've got a research agent. That research agent, uh, explores the venues and matches those to the preferences of the customers they've collected. They've taken a process that used to take two or three hours per client to develop a proposal for a custom holiday and turned that into a process that takes three minutes. And now that travel adviser just reviews that, makes any final tweaks and presents it to the customer. A dramatic savings has helped them achieve a 20% acceleration in growth of that business in terms of revenue in just the first few months. Right? So that's rapid ROI. So it can happen quite fast. Let let me ask you about AI coding broadly speaking you know when I think about the opportunities the work gets done faster that's very clear as as one opportunity for people on the ground what do you think about the risks here I mean as these AI coding tools become more and more ubiquitous and people start to use them a lot I mean is it concerning at all in some cases at all I mean sure at scale you know aentic AI can either kind of drive sustained progress or introduce real operational risk kind of depending on how deliberately it's designed, governed and run. Um, so you know, I would say just because it's easier to generate code today doesn't mean that's any good, right? Like just imagine, right? These agents are outcome driven, right? So imagine that code decides to sell all your confidential data about your customers to your competitors. Maybe it thinks that's the fastest way to turn a profit in your business. Then maybe >> that's the way it's been that's the way it's been bycoded essentially and nobody was really paying attention to it. >> Could be. Right. And so the question is how do we develop real enterprisegrade architectures around this new technology so we get the benefits without all those risks. So we need the built-in safeguards to unlock scale without the chaos. And that's really where platforms like out systems help. We're a deterministic platform. For years, right, customers have been building deterministic workflows where you get the same input 10 times, you'll get the same output 10 times. Now, we have LLMs and by their very nature, they're probabilistic, which means that you put the same input in 10 times, you'll get 10 different responses, right? And that can be great for reasoning and creativity and that's awesome, but it's not exactly what we want in our most like say regulated businesses on the planet. We actually want a mix, right? You want some of the reasoning to help accelerate decisionm, but you want deterministic processes for things like communicating outcomes to customers. Imagine you're a bank, right? And you're making loans to consumers. That's a well- reggulated industry. You don't want the responses to your customer about why their loan application was denied written by an LLM in a fully autonomous process. That could create legal risk for you. When you let me ask you, you want a form letter, right? >> Let me ask you a question about all these different coding models then that we see coming out. I mean, every month there's just uh there's a new toy for people to play with, right? And and you you see the benchmarks moving up and down. uh you see people talking about uh who is winning the race. How do you see the conversation around these models playing out in the long term given that you integrate with all these players and allow your customers to use so many of them? >> Yeah, I mean one of the things we're focused on is like being agnostic to the model that customers choose. So in the agentic systems customers build on our platform, it's typical to have four or five agents being orchestrated in a single system and often those agents are using different LLMs that are optimized for the use case. And we're starting to see customers actually avoid LLMs for some of them and go to small language models that are more tuned to the specific task at a lower token cost. Right? And so we think that customers and we see that customers are going to want to evolve and change the models they use. You know, as Gemini releases a new as Google releases a new version of Gemini or Open Eye comes out with a new version of chat GPT, you want to be able to play with that and hot swap those models without reinventing the entire Aentic system and just tune and optimize. And that's what we're making possible by allowing you to bring your own model, bring your own tokens and essentially use our platform for orchestration. So we really think that this world evolves to the point where what matters is the platform and the orchestration and not the specific model. >> Right. Well, Woodson, I want to thank you for joining us on the show. Uh we will talk to you again very soon. That is Woodson Martin, CEO of Outsystems here on TIV.