Low-Code AI: From Requirements to Apps in Minutes | OutSystems' Rodrigo Coutinho

Speaker 0:00 Now you can ask someone what they want. You do it live and you have in seconds a prototype of what the person told you they wanted. The fact that you can do these iterations in seconds in a meeting room, it completely changes the way you work. So I think that's one of the great superpowers of these new technologies, is the ability they give you to iterate very fast and therefore innovate much faster.

Conor Bronsdon 0:30 Welcome back to Chain of Thought, everybody. I am your host, Conor Bronston. Today, I'm joined by my fantastic cohost, Atin Samuel, co founder and CTO of Galileo, and our guest, Rodrigo Cotino, co founder and AI product manager at OutSystems. Rodrigo, welcome to the show.

Speaker 0:46 Thank you so much for having me. It's a pleasure. Absolutely. I think Ottin and I are really excited for this conversation because

Conor Bronsdon 0:52 you co founded Ottin Systems nearly twenty five years ago now. In 2001, you've been the driving force behind Ottin's product vision, the focus on enterprise applications, which I know is something that Ottin thinks a lot about as well. And now you're obviously responsible for leading the charge in bringing AI to the OutSystems platform and helping your customers take full advantage of AI in the process.

Conor Bronsdon 1:16 And there have been so many changes that I know you and Aten have both seen in the AI landscape over those years. What are the big ones you've been seeing as you've evolved your work at OutSystems?

Speaker 1:28 Yeah. So so so it's been it's been terrific the way things have been involved on the AI side in the last few years. So when we began our journey at OutSystems using AI, we we already envisioned this future where AI could be used to help developers be more productive, to help them create better applications, specifically enterprise applications. And so we started investing that as you mentioned back in 2018.

Speaker 1:54 At that time, it was the era of deep learning neural networks. We also have some some other learning abilities that we added to to the the algorithms that we have. But it was essentially those old technologies, you know, that are seen today as old technologies that were helping us add these capabilities to the product to help developers. And some of the things we added

Speaker 2:18 included things like trying to understand what's the next thing the developer wants to do. So give suggestions while the developer is building the application, what should they put there next. Other capabilities we've added were validation capabilities. So we would look into the, well not the code, we use a visual model to develop applications. So we had a mechanism that used automated reasoning to look into these applications and make sure they fit the standards in terms of security,

Speaker 2:49 in terms of performance and all of that. But it was pretty much what nowadays would be called old school AI, let's say like that. And then LLMs showed up. So these new large language models and specifically with the release of OpenAI that changed completely the way we think about these problems. And it was super interesting because very early on we saw the potential

Speaker 3:17 of these technologies and we added a capability to our software that would enable a developer to add the sentence in English and it would translate into a database access. So that was the first time we tried to use these LLMs to help developers be more productive. This was before all this craze, so OpenAI wasn't open yet. It wasn't available for everybody, so we actually had the large linkage model running in house in one of our we had an instance here, everything was tweaked by us to to work, and so this was our first dive into this modern GenAI technologies. So it was obvious for us that this would change everything.

Speaker 3:56 So it was a huge step that we took. And today that feature is still there, by the way, so it's something that you can still use. But then of course things accelerated and we started to make even more use of it until we came to the point where we are today with with Mentor, OutSystems Mentor is our AI technology, where just by adding a prompt or giving a requirement document,

Speaker 4:19 you can build a full fledged enterprise application in just a few minutes. So it's really amazing to see where all the advances we've been having just a couple of years using these technologies.

Conor Bronsdon 4:33 Yeah, Aten, I feel like I've seen you have a similar journey over the last several years as AI has evolved, where you've had these distinct experiences across different companies, and obviously now on your founding journey at Galileo, where I'd be curious to just hear how your experience parallels or or intersects with the effects that Rodrigo's talking about.

Speaker 4:58 Yeah. I think in many ways, my journey is almost eerily parallel to to Rodrigo's, especially with being in Galileo because we kind of started our journey with the old technologies that Rodrigo was mentioning, more from a data quality perspective. So, you know, the definition of evaluations in the erstwhile era before LLMs meant that, hey, you're training some kind of a language model in house and you have

Speaker 5:28 training data that with labels and a lot of mistakes and noise in the data, which is dragging the performance of your model down. So which is what we primarily worked on before LLMs hit the scene. You know, we would build algorithmic ways to detect noise in your training and evaluation data to help to improve the quality of your in house models. Of course, those things, like Rodrigo said,

Speaker 5:53 the workflows have changed now where you can just prompt and get a very smart output from an API or an easily hostable, very large language model. But the problems have kind of evolved to the point where you're now trying to basically manage the quality of your inputs and outputs, whether it's the prompt quality that's leading to better or worse outcomes, or just the quality of the output itself, whether it's grounded in the context or whether it's complete and relevant to the input.

Speaker 6:27 So the problem the baseline problem remains the same, but, you know, just how you measure it and what you care about has changed. And in that line, actually, question for Rodrigo. Have you kind of seen the same trends when it comes to evaluations and thinking about the quality of these systems? Oh, yes, definitely. So one of the things that we are proud of is that whatever we generate is a valid application.

Speaker 6:55 So when it's done, it needs to compile, it needs to do something that's useful, it needs to be valid. And so we do all of those checks. So we use these tools in margin language models and we play with a lot of different models, now I was kind of smiling when you were talking about the evaluation, we keep testing and changing what's going on even without making that public. We're just testing it out and using

Speaker 7:19 different models. But then we always do this validation. We use, as I mentioned, a model language. So that helps because we can in one step validate if the model is alright. But more than that, we also use other AI tools to validate if it's secure, if it's fast, if it's doing the right thing. But that's a big part of it because, I mean, we we we've all heard the funny stories of AI hallucinating, producing things that don't make sense or or the horror stories where it actually produces

Speaker 7:47 problems in security and so on. And and so we have all these steps in the pipeline to avoid that type of thing. Yeah. And obviously, this aligns directly to OutSystems'

Conor Bronsdon 7:57 long term philosophy around both low code and AI of empowering developers to deliver custom apps in an easier way and empowering folks who maybe aren't senior developers to do more. Something that's really happened here in last couple of years is so many more folks are now enabled as builders, both through low code tools, like what AltiSim has done, and through AI powered coding and these other opportunities.

Conor Bronsdon 8:21 How have you seen this current explosion of tooling and, you know, new reasoning opportunities align to your long term internal company philosophy, and how has the company transitioned over those years to adapt to these new technologies? The good news is that these tools are totally aligned with our vision. So our company vision is to allow every company to innovate through software,

Speaker 8:46 and so the faster you can create this software, the better the software is, and the more people can create it, the faster you get to innovation. So this ability that these new tools provide for you to experiment very fast, and that even allow you to improve the communication between the people building the software and those that are going to use it, Because what used to take months is before where you had the endless discussions and meetings writing up requirements.

Speaker 9:14 Now you can ask someone what they want. You do it live and you have in seconds a prototype of what the person told you they wanted. And so you can immediately validate if that's the because well, let's face it, sometimes communication fails, right? So so a lot of times what happened was people would ask for something, a month goes by, the engineers come with something that's totally different or that doesn't make sense. The fact that you can do these iterations in seconds in a meeting room, it completely changes the way you work. So so I think that's that's the actual one of the great superpowers of these new technologies

Speaker 9:50 is the ability they give you to iterate very fast and therefore innovate much faster. So for us, this was amazing. This was terrific news. And we were really in the right place because having this low code capabilities, as you mentioned, enables you to have people that don't have that many knowledge about software development to also create applications. Now, we still believe that the tool is for pro developers and that they are the ones that that will reap the most benefit from it. But now other people can also participate in the loop. It's not an exclusive

Speaker 10:24 benefit of pro developers. But most of all, this helps pro developers focus on the important things. So the old things where you you would take a very long time just trying to understand how small details of the technology work, being it tweaking CSS or or doing integrations or or doing that type of low level things, now that's instantaneous, and developers can really focus on what matters.

Speaker 10:52 That is, how can I help my end users use these to innovate and to bring the the business to a new level? So I think all of these things together are completely aligned with the vision that we have since 2001, because this is something we've been wanting to do since 2001 and allowed us to give this big jump with OutSystems Mentor. So we're very excited about this. Rodrigo, on those lines, I kind of have a question for you from a quality assurance standpoint. I'm sure you think about, you know, how end to end,

Speaker 11:24 you know, the the user behavior of your application gets affected by the quality of the outputs. You mentioned the point about faster iteration. Couldn't agree with you more. Now it's much easier to, you know, have choices of LLMs, choices of tools, really it kind of comes down to how can you find the ideal combination that makes up your application in the least time possible.

Speaker 11:47 But of course, there's a penalty you pay from my perspective when it comes to speed, which is you have to figure out how to effectively, you know, quality control and evaluate these systems. So my question for you is, have you seen this trend that now with the more and more powerful models, the the inherent mistakes in the models are low, but the the onus has kind of shifted to the the surrounding components of the software and, you know, mistakes in those kind of lead to bad outcomes. The focus for quality is more around that and less around the specific LLMs

Speaker 12:28 making mistakes. But I'm curious if you've had the same observations from your experience or have you observed something else? We did, but interestingly enough, more when we use these tools or when our partners use these tools to generate traditional code. So so we have this great benefit of having a language model that that abstracts a lot of the complexities

Speaker 12:51 of what's happening in the details, which means it's also much easier to validate and make sure everything is alright. But we have witnessed some situations and we have some partners that even went to the to the lengths of of turning off these helper tools to generate code, because in the end, if you're only generating code, then you have a problem, which is who owns that code. So they had

Speaker 13:18 an accountability issue where they had a lot of generated code that nobody understood or knew exactly what it was doing and it worked in the situation and suddenly stops working and there's nobody to maintain it or fix it because it's like nobody's responsible, nobody understands that code. So I think that's kind of a limit situation, but it does show that there's kind of this promise where with AI anybody can do applications.

Speaker 13:47 Not yet, right? You still need that expertise, as you mentioned, to be able to look at the code, understand it's doing the right things and be able to tweak it and to make it unique. And ensure it fits your architecture, ensure it fits your compliance requirements. All of this becomes much easier with the low code, so I think the combination of the two, it's really the secret here.

Speaker 14:08 So now we generate full code, which means it's much easier to read and it's much easier to validate if it's okay. But yeah, we've been seeing all those requirements of looking, okay, so this was generated by AI, how do I make sure this is what I want? And and in in, and there are several ways to do that. On our side we do all the pre validations. We also do something that is when we finish generating an application,

Speaker 14:33 we also use the AI to generate sample data. Because that's also a great way for users to validate immediately if that's the result they want, right? So when you build an application, you run the application, you don't have just tables saying no items to show. You actually have a lot of data so that you can check it to see if it's alright. And then of course we have also AI doing the validation for the security and all of that stuff that I mentioned earlier. But yeah, definitely

Speaker 14:58 the role of the developer right now, it's much more about being a team leader and doing code validation and making sure everything is alright, doing the communication with the business and orchestrating all of these parts rather than being a big pusher trying to make an integration work. Can't help but kind of throw a shameless plug here. This is something that Galileo

Speaker 15:22 focuses on day in and day out, to think of us as an evaluation co pilot. And like you said, there's two aspects to this. There's the quality of the generated code itself. A lot of people are using Cursor and similar tools to pretty much prompt their way into an application. So a lot of code is just auto gen by MLMs. So how do you validate the code that's being generated? Almost like a real time evaluation

Speaker 15:48 on a copilot that can not only validate and give you insights on the generated code, but also create unit tests for you automatically for that code. So that's one aspect. I see it as like the micro benchmarking aspect of Evalves. And then there's the macro benchmark, is testing out the end to end quality of the application. Like you said, data sets are a big part of it. You're rating the data that you want to test your app with. We are kind of in the business of automating a lot of this, not fully, but really assisting the team leader,

Speaker 16:19 being able to do this efficiently. Create datasets, test your system end to end, and also test the quality of the code that's being generated. I know Google had, I think, put out this, statistic and this is dated at this point, probably like eight, ten months ago, that about 20 to 30% of Google's code was generated by an LLM, which is kinda crazy to think how big, you know, Google's code base must be. And this trend is only going up and up. So the need for, you know, robust real time evaluation as code is being generated

Speaker 16:52 is much higher than what it was yesterday.

Conor Bronsdon 16:55 Well, and this is where I think Rodrigo's point about low code and AI is super interesting because not only are we seeing just more code be generated in general, but as I talked to Charity Majors about on the podcast last week, we are seeing this shift that Rodrigo is also alluding to around devs who are even at like a junior or mid level are kind of moving into the role that was formerly taken

Conor Bronsdon 17:20 almost exclusively by senior principal staff engineers who were, Hey, I'm solving business problems. I've moved from just being a cogeneration person to, I am thinking about the business problems and I'm solving those problems. That's obviously always been an aspect of software development, but often as you were more junior, you would be supporting a team in doing this. And now you're seeing developers,

Conor Bronsdon 17:42 whether they're individual devs who are just building their own apps, who are being enabled to do that faster and actually really go solve a problem, or on a team level, they're being challenged to really up level themselves. And this is where I think you also see a lot of startups hiring a lot more senior engineers and saying, Hey, look, I want folks who are great at solving business problems because,

Conor Bronsdon 18:04 you know, we already knew that elite teams didn't have a problem generating code, and that the problem was more about the life cycle of software, identifying the problem, aligning the problem. But increasingly, we're seeing that be even more true with these tools that are enabling us to generate code faster, generate applications faster. This is where I'd love to ask you a bit, Rodrigo, about Mentor. Obviously, this is another take on how to approach AI application development and

Conor Bronsdon 18:28 building applications from the ground up. This falls under this idea of aligning low code with AI tooling, and this this mission that OutSystems has been on for a long time. But I'd love to understand a bit more about it, and and how you see Mentor

Speaker 18:43 playing within this ecosystem of AI tools that are enabling people to build applications today. Mentor is essentially the umbrella for the AI tools that we made available on top of the OutSystems platform. It targets several stages of the process. We have first a tool that with a simple prompt or a requirement document, it will go from zero into a generated application.

Speaker 19:08 And the way it does it is is in it interprets everything that you you type or that you send it and then kind of gives you a summary of what will happen. And this is an important aspect because we are true believers that the magic occurs when you mix AI with humans. So at every point in the process, we believe the human should know exactly what's going on and should have control to change or to influence the result. So this is like the first step where they can go there and influence that result.

Speaker 19:38 Once you're happy with that, you just click a button and a full fledged enterprise application is created for you. So this includes the UI, it includes the business logic, it includes the database, everything is ready and it's ready to publish. So we make sure the application is valid, you have what we call the one click publish. You can click one button, we generate the container for you with the application ready, with the database ready, all the sample data all ready to go. So this is like the first step.

Speaker 20:07 Now of course we know that, I mean, people forget to write stuff on the requirement documents or or the AI might miss it. Yeah, I know. Right. We all do. Or even as as I mentioned before, because you can do this in in the middle of the meeting with your business user, they can look at the application and they say, Oh, I just had a great idea. I wish we had this type of data so that we could add more capabilities to the app. And so we have this other tool that allows you to do this also using AI.

Speaker 20:39 And here, I think that there are a bunch of innovations. It's visual, so that makes a lot of difference and all of that. But one of the things I love is that it gives you suggestions. So instead of demanding that the person knows exactly what they are going to do, so the developer needs to tell the machine exactly what he or she wants, we give suggestions. So based on the context, based on your factory, based on the application,

Speaker 21:06 we we have an idea of what you wanna do next. So you can of course type your own thing and do your innovative thing, but for the usual patterns, it's already there. You just have to click it. It does it all for you. And each time you do one of these iterations, the application is ready to run. And so this is like the, let's call it the design in the iteration parts of of building the application.

Speaker 21:29 And then I'll also, as I as I mentioned before, what what we did was we added a whole validation step on top of this. So when you finish building an application, your application is running, we have AI agents that validate for performance problems, security problem, compliance, all of that stuff. And so you have a report that you can go there, check what's working good, what can be improved, and then improve it.

Speaker 21:54 And then the final thing is, it's not only about using AI to build applications, you actually want to add AI to your applications. So we have what we call the AI agent builder, which is a tool that allows you to, by connecting to several models, OpenAI, Bedrock, Google, all of that stuff, you can create agents that then we we transform into a a low code element. So you have the visual element where you can plug AI throughout your application

Speaker 22:26 to do simple tasks like summarizing things or even so we we had this this scenario where we actually put an AI controlling a business process where based on it, they they would decide if something is approved or not. So we also make it very easy for you not only to use AI to build applications, but also to put AI on your applications to benefit your end users. And so I know you mentioned

Conor Bronsdon 22:50 OutSystems had your own in house LLM that you were developing. Is that what's powering this, or are you leveraging, you know, an API from OpenAI or something else to kind of drive the creation behind this?

Speaker 23:04 We use multiple models. So we do use our own models, but we also use OpenAI. So but but this is one of those things that maybe in a week this answer will be outdated, because we keep testing and validating and making sure which one works best and we keep changing it to give the best possible results. So this is something that the team is always looking at improving.

Speaker 23:25 Rodrigo, a related question to what you just said because I found it very interesting that you're kind of one of those rarer organizations which do both in house model building as well as leveraging what's out there. From a cost perspective, how do you see the trend moving towards? I mean, you're trying to figure out that, at least in my experience, we've seen this triad between cost, accuracy,

Speaker 23:51 latency, and different combinations of those kind of suit different applications. But you've seen both sides of the coin. You know, the cost of training and fine tuning an in house model with GPU resources, etcetera. And then there's the cost of shipping your data to a different cloud and having rate limit impediments and all of that. Having seen both sides of the aisle, where do you see the trend going? Are you more seeing more bang for the buck,

Speaker 24:19 you know, in house and seeing efficiency there? Or you feel like we're going to move to a place where, you know, fine tuning may not potentially be a thing that most organizations do? So it depends a bit on the problem that we're solving. Whenever we can, we try to use this more public models. Because of what you mentioned, right? The fact that we don't have to host them, the fact that we don't have to train them and all of that,

Speaker 24:44 it's it's it ends up being less expensive, mostly because also of the investment we need to make in people to make sure that those things happen. But then there are models that either because they use sensitive data or or because they are very tweaked and specific to our systems that then we need to make them in house. But again, it it depends a little bit. We we

Speaker 25:05 we tend to focus more on than the result that we have for our customer rather than the cost. The cost is something that's also changing a lot, funny enough, because our AI team is really very creative in finding ways to save money. It's something that they are always thinking about and so this is something that we've been able to shift back and forward. But I would say right now, whenever possible,

Speaker 25:28 we use these public models because of the way they've been evolving. It's amazing how how you go from from Chattypiti 3.5 to four and suddenly there are a whole set of problems that that get amazing results, you know? So we want to take advantage of that and the capabilities of big corporations to train these models. But then of course, whenever we have compliance issues or more specific things, that's when we go in house. Got it. Got it. And also, would like to clarify that when I said fine tuning,

Speaker 25:58 I meant in house fine tuning, not just fine tuning in general. I usually wear a T shirt that says fine tuning is a big part of the future. So I just want to clarify that.

Conor Bronsdon 26:08 Love to get some of Rodrigo's thoughts about the future as well, because it's it's very clear that OutSystems sees this opportunity to enable people across the world to build. Obviously that's been a huge part of your mission, you know, the entire twenty five years the company has been in existence. But what are the key challenges or opportunities that you see, Rodrigo, in the intersection of kind of AI, software development and creativity, which we're talking about here? I see a bunch of opportunities in this ability to to change

Speaker 26:41 completely how how you think of software development. So this idea that you can do things immediately, so when you start thinking about this, right, when you can build applications almost instantaneously and have the right feedback and improve them on the fly, I think that's really going to change the way people relate with building software, right? Because right now it's seen as something that's

Speaker 27:04 it takes a long time, it is expensive, and many times it's not seen as a possible solution. But if you truly want to be innovative, if if you truly wanna be unique and differentiating, you need to go that route. Because if you're using packages or something that everybody else is used, you're never gonna be able to to stand out from the crowd. And I think these tools, low code, together with AI,

Speaker 27:29 bring this to a whole new level where creating software is much faster, much cheaper, and allow companies that before thought that custom software was prohibitive in terms of cost and time, Now it becomes a possibility and they can go that route. So I see a lot of promise in this new world where it becomes much easier and faster to create the software. And also because

Speaker 27:51 even companies that already were building, suddenly they can start experimenting much faster, which is also a big part of this. So if you can do five different softwares for the price of one, then the cost of failure becomes acceptable and you can start experimenting and doing crazy stuff. That's hard to see into the future exactly where that will bring us, but I'm very curious because I'm sure it will bring some very exciting things. It will keep getting better

Speaker 28:20 as these models start. So I'm very excited with this ability of models, take several types of information where you can get prompts together with requirements, together with drawings, together with photographs, together with, I don't know, a recording. It would you can have a meeting with someone from the business recorded, give it to the to AI and get an application from there, you know, or you can just catch on a whiteboard what an application would look like, send it to AI and create those applications. So I think this this multi model ability

Speaker 28:55 of these models will grow to accept different types of inputs more than they can accept nowadays. And that will also completely change the way in which you think about and ideate around software. I see a really bright future here in terms of what you will be able to do. I think the major challenges, one of them is this that we already talked about and the team was also addressing, which is when you are able to build a lot of things, it's also easy to do a lot of not so good things, you know. So it's

Speaker 29:30 much faster to do the wrong things also and not in terms of wrongdoing, but really people being very well intentioned, but suddenly for some reason creating software that doesn't help or that spoils things. So this attention to quality, this checks and balances, making sure the software is properly done and secure and all of that, I think that will grow in importance.

Speaker 29:56 And finally, I think another one of the big challenges is human nature. So we are in It's one of those situations where things change faster than people can manage. And so, yeah, great. So you can make five pieces of software in five days, but who will adopt it? Ability to then take these great pieces of software that are being created and these awesome ideas and get them adopted and used, I think it will also be the next step on

Speaker 30:25 the challenge of then making true use of these AI technologies. I'm personally very excited about multimodal. Like Rodrigo said, this is just the beginning and there's so much that's already kind of the foundational layer of multimodal at the LLM side has kind of taken shape quite a bit as agents have picked up more at the application layer. So it's going to happen in the next ten to fifteen months. We'll see the first set of true multimodal applications

Speaker 30:57 to see the light of day, a lot of interesting technologies which have been developed, multimodal embeddings being one of them, you know, trying to make sense of information which is kind of baked in with text and images and video and building like a question and answer system. I'm personally very excited about those kind of sort of augmented Q and A systems, retrieval systems, leveraging multi modal.

Speaker 31:27 But also, think, like Rodrigo said, we're moving fast. In fact, I would say there have been instances in history of technology where things have moved fast. But my personal take probably is like, it's the first time that the foundational layer, like the ingredients are truly so good that, you know, the pace at which we are moving on top of it is almost like very real. Like we're trying to build something that truly works. Often we've seen, you know, back in the day with the cloud technologies,

Speaker 32:01 they they they kind of had step function increases over the years. But here it's like you you have the ingredients ready to truly build something super exciting. Again, like we're just getting started with with agents and multimodal. All this will come together to truly create something that we've never seen before. And to me, that's the most exciting bit. I'd be curious, Rodrigo, how

Conor Bronsdon 32:26 OutSystems is thinking about incorporating these multimodal AI capabilities or or what your approach is there.

Speaker 32:33 From a software development perspective, the idea would be to, as as I mentioned, to extend what we do now in terms of application generation and in the application changing, to allow, instead of just prompting in a text based requirement, to also allow you to add things like hand drawn designs or or a recording of the meeting or even a video. So so you could make like an animation showing exactly how you want an application to behave or you could record an application. Right? So so I could record myself,

Speaker 33:05 I don't know, using a CRM and then I'll say, look, I want something like this, but now make it specific for for this industry. So that's the type of thing we are we are considering is is how can we use this multiple elements, put them all together, and in the end create something that's an enterprise application ready to go. More on the side of of how our customers can use it, what we are thinking is that there are a lot of amazing examples specifically with vision that have been happening

Speaker 33:38 more in the old tech that we think it can be bring, and so the let's call it traditional AI that we think we can bring to new world. So there are a few, for instance, we have customers that use AI on top of video images to understand the condition of a specific warehouse and make sure if it's full, if it's empty, how much storage you have and all of that type of thing. But it was a very manual process and with the training that Etienne was mentioning and all of that, so I think there is an opportunity to improve those models and to make them more accessible to more people.

Speaker 34:13 And so it would be super interesting to see if if you had if you were able to build an enterprise application where you send, I don't know, the photograph of the building and immediately it understands what needs repairs, for instance, or of a warehouse and it knows what needs to pack. So this type of ideas. The same thing, of course, for audio and all of these things. So I think there is a lot of possibilities here.

Speaker 34:36 And also, so most of the real world is not text, right? And so as soon as you can have this video audio sensors, that also brings the door to something we haven't mentioned here yet, but that's also on the rise, which is the Genetic AI. So having AI actually be proactive on some of the things it does. So instead of just being a question answer type of thing, where you ask for something and it gives you the result, it could actually based on

Speaker 35:05 something that's happening in the world be proactive. So going back to the example of the warehouse, I mean, it should be able to detect that something's missing and immediately put a purchase order for toilet paper or or whatever. So so this is the type of thing that we are thinking about. How can we combine all of these pieces that are showing up and bring value not only software development, but also for our customers.

Speaker 35:28 That's super interesting. Actually, I had a question on that note. So you're already offering frameworks and tools like AgentBuilder to your customers so that they can create their own agents and From your perspective, I'm curious to know some of the challenges that you're seeing both from a customer standpoint where customers are hitting certain challenges using AgentBuilder,

Speaker 35:53 for example, but also you offering an agentic platform. What are the top of mind challenges for you? The biggest challenge for me is funny enough, the fact that so this was a very particular technology revolution in the sense that it kind of mimicked the old technology revolutions where technology was more important than use cases, and that's kind of happening.

Speaker 36:22 The biggest challenge right now for a lot of our customers is, okay, we know this is important, we know this is the future, but we still don't know how to use it or how to what are the use cases that can really benefit from this type of technologies? And this of course depends on the level of maturity you have regarding AI. We have a few customers that of course already know and are using in things like

Speaker 36:47 for ticket deflection on their support processes or for automating processes, having agents deciding if if a specific thing is approved or not. We have some interesting use cases of customers matching information between things like job offers and CVs, you know, that type of thing. But a lot of them are still trying to find the use case that will truly make this technology shine.

Speaker 37:14 So funny enough, biggest challenge that we have is this need to educate people on exactly where this technology is shining and what they are good for. Because sometimes there's also this wrong perception that these technologies fix everything, which is not unfortunately correct. But they they are better at some things than others. And and so we need to give that education and then work on them in ideation

Speaker 37:40 to come up with use cases that actually benefit their their business. So the big right now, the biggest challenge is not the technological. It's it's actually this this ideation process and coming up with great use cases where they can can benefit from the technology. I think that's a fantastic note for us to close out on. And, Rodrigo,

Conor Bronsdon 38:00 we're very excited to to see what OutSystems continues to do in the coming years as you as you build upon this mission of enabling people all over the world to build. Where can folks who want to follow that mission follow you or follow the work that you're doing in OutSystems? So you can visit our website, www.outsystems.com.

Speaker 38:20 You have there the ability also to to try the platform and and to try Mentor on your own, so you can see it working and build your own application. And and if you wanna follow me, of course, I'm available on LinkedIn. Just just connect with me. Amazing. Rodrigo, thank you so much for coming on the show today. This has been a ton of fun.

Conor Bronsdon 38:37 Thank you so much. It was a pleasure. Loved the conversation. Absolutely. And, Atin, thanks for chiming in with all these fantastic insights from your end as well. Always great to chat with you.

Speaker 38:46 Likewise. It's it's very fascinating to talk to industry leaders Yes. Founders who are, you know, at the cusp of this incredible moment. It's quite amazing to be a founder in this this era because you get to have a front row seat to all the innovation that's happening. And, yes, people are figuring it out, but that's, I think, part of the fun part of the story. Absolutely agreed. And I I think this conversation really parallels

Conor Bronsdon 39:13 one we had last week with Charity Majors, CTO and co founder at Honeycomb, talking about how you have to change, how you build teams for this AI era, and how you have to think differently about like where the bottlenecks are, the same things that Rodrigo is alluding to here, think, around The way you put it, Rodrigo, with like ideation is now this bottleneck, and Charity's saying about like the life cycle of code. To me, these really mesh well together with the workout systems are doing. So if folks are listening and they haven't checked out last week's episode, I think it parallels really well for this as a great follow-up to kind of broaden your thinking through this. And why don't you just leave us a quick rating and review while you're here, while you're listening? Those matter a ton to us. Everyone's laughing at me because I say this every week pretty much. But, seriously,

Conor Bronsdon 39:56 we we appreciate y'all listening immensely. We appreciate everything you do, to share the show. And more than that, we appreciate our guests. Rodrigo, thank you so much for joining us today. It's been fantastic. Thank you so much. Awesome. Thanks, y'all. And we'll see you next week.