GenAI and the Evolution of Software Development with Paulo Rosado of Outsystems

Thank you.

Speaker 2 

What you’ve done to kind of shape the company to where it is and from where it start? 

Speaker 1 

Well, I think that my major contribution has been. 

Speaker 

Oh. 

Speaker 1 

On culture, really imposing a culture of, of innovation, of failure is acceptable, and that you really, when someone tells you that something is something, is a is a urban myth or something is something that. That you cannot change really grow and challenge steps. And that started with the at the end of the 90s, there was a perception that software. Late projects were effective life. So you talked with a lot of CIO’s, a lot of it a. Lot. Of software developers, and they usually blame the business because they couldn’t get the requirements right, so everything was always late. There were missed the misconceptions about what was built and the when we looked into the. Problem is that why is that everyone accepts that projects are late projects take long. You cannot predict them and so we kind of almost look at the problem and the reverse engineer until we got into the Asus platform and then add those. Of challenging the status quo has always been part of the culture. But not only with me. Most of the innovations do not come really from my head anymore, OK, so. 

Speaker 2 

You. Driven your company to a point now where? Can you? You’ve developed? Low code so that it is now accessible to. Everybody. But now you’ve you’ve brought in generative AI. To become a little. Bit more accessible and more effective for for enterprise applications. So can you talk about some of those use cases and scenarios that you’ve done using the combination of these two technologies? 

Speaker 1 

Yeah. What what we’re facing now, we’ve been. So we’ve been using AI and Gen. AI especially. In order to enhance our products, so there’s a lot of agents and generic technology inside our platform, but what interests also is is. I’ll do a business applications change with inclusion of AI agents in the middle, because today the components are things like portals, workflows, logic, business rules, repositories of data. And where is the rule? Where can you put an agent? Where can you put the AI so that you change the way digital systems are built? And what’s interesting is that we building a new agent. Is, is, is, is, is is one step. That needs to be surrounded by a bunch of logic by a bunch of software until it becomes really effective. And so we’ve we’ve been having a lot of experience with our customers as they deployed agents and Gen. AI. That a lot. Of the work, and that’s by eventually becoming a mixed. Of the typical. Additional software like a portal on top of an agent or a collection of policy rules. With that control, an agent access to data, so a lot of these things create are are are necessary so that we can create a system that’s actually usable and adoptable. 

Speaker 2 

1. So you’re you’re talking about, when I heard the word policy. So it seems like. They’re trying to inject almost rules and processes in a more automated way in the system, through an agent that allow it to be what we what do we say trustworthy?

Speaker 1 

Yeah. For instance, we have a. We have a big customer rush, OK? Or decided to create. In this first instance ChatGPT, but the safe ChatGPT for the whole company. 

Speaker 2 

Is it The thing is a safe ChatGPT? 

Speaker 1 

Yeah, because what happens is that that particular ChatGPT can access internal data from rash. So what they did is that they they they deployed that in the multi set up depart. And for instance, when they reach HR, you can ask any questions about your benefits, about a bunch of stuff that otherwise you’d have to call somebody in HR. But one of the things you can ask is. How much does my boss make? Are paid in terms of salary and so one of the things that they had to do was whenever you accessing and emerging these agents with access to internal data, they put a series of policies that have to do with the what is the data that you actually can access and mesh. That with the bonds, all of that was done with the low code and it was very iterative. We have solutions that can be deployed in one week, but in the meantime there are probably 50 to 60 versions and so being able to iterate. Test them and see if they are. If they really are respecting policy access rules. If they don’t, go into too much hallucinations, you know all of these things are crucial to build a highly adaptable, adaptable. 

Speaker 2 

System. OK, I want to change a little bit. And talk about. The. The ability for you to actually tackle impossible projects, and when I say impossible, let’s talk about the legacy systems because we know a lot of companies have it. They’ve invested millions and millions of dollars on it, but in the past it’s been difficult to integrate transformation. Transformative solutions against legacy systems. Can you talk about how that’s? 

Speaker 1 

Changing. Yeah. What what we find today? A lot of organizations have a large number of legacy systems. That are fundamentally end of life. There are 1520 years. We’re just talking with the custom that has a system that was built in 70. 4. And so you can imagine the language. Yeah, it was a global system mobile RPG. 

Speaker 2 

Mobile. 

Speaker 1 

But all developers have disappeared. There is no manuals, there’s nothing. And so a lot of these systems are chain balls that actually are attached to the to the feeds of the the business. They cannot evolve, they cannot. They cannot build new systems on top of it. And what? The the potential. Of this disruption of of compressing the time that you develop is that instead of looking into a two to four year development project to replace something. You can now do it a project of three to four years. You can do it. In seven months, right? And when you do it in seven months with the technology like for instance out systems where the cost of change is so far. Best you can actually not only compress the time, but any type of surprise that you might find like requirements that you software a particular way kept on changing. You can incorporate them in real time and so suddenly you go from a two year plus project that you didn’t know if. You’re going to deliver. The seven months project that you deliver on time and on budget and that has made a lot of these legal. Projects that were considered impossible because they have no way of being transformed. Suddenly it becomes possible to rewrite them. 

Speaker 2 

So from your perspective, does that mean over time, even with the use of your systems and a couple? Regenerative AI that. The cost of development will come down significantly. 

Speaker 1 

The will and the. For instance, what we’ve seen is we have so many, so much experience with this that we’ve been compressing projects of about four years into between seven months and four. And what we see is that a lot of the Genii capabilities in a lot of steps today done manually can actually further compress at least two ex and so reduce something that would otherwise take three years to about 3 months. And so we’re seeing orders of magnitude of compression. That’s certainly for a lot of organisms. That are looking into. I want to build this. I want to build a new project or a new system because I want to release it three to four months and it’s pretty complex. Certainly it’s possible to do this with a relatively small number of people and a lot of tooling and a lot of help. Assistant school, right? 

Speaker 2 

Now Jen AI is still very nascent. It still has its issues. So what are the current risks that you see right now that you’re trying to solve for? 

Speaker 1 

Yeah, one. One of the the ways training is being used for these these type of platforms, the software development platforms. Is really this this concept of copilot where you have a companion to developer that helps you generate pills. And what we’re seeing already is that the code that’s generated is 5050%. Longer than what you actually need if you write. It by hand. So you’re right, you’re already creating a large chunk of technical debt in the code that’s being generated, and we were expecting this because we faced that back in 2004 when we started deploying our first versions of the platform. And so one of the issues that we realize is that as you build these systems that you generate more code, the code becomes more opaque. It’s more difficult to understand. What it does? To a point where after a while you start to have a bot that explains you. What the code does? And so you have a bot that generates the third another bot that explains that the code becomes less and less important. Is it knowledge transfer mechanism and So what we find is that it’s extremely important to be able to explain how did a Genii model created the particular piece of code that why? It what was the rationale behind it and that we believe that that that is a a fundamental function of research in the next year. 

Speaker 2 

Umm. 

Speaker 

Well. 

Speaker 1 

Because otherwise it’s going. To be very difficult to apply. 

Speaker 2 

This. Yeah, absolutely. 

Speaker 1 

It’s a it’s a problem we solve with the combination of techniques that involve local the Genii, but it’s a real problem for very large systems. 

Speaker 2 

It’s been a problem even in narrow AI, especially when it comes from an IP perspective. Start to see you say you start to see companies who want to protect what’s in that. Black box, right? But if there is a lot of organizations are going to trust these systems, there needs to be a level of transparency that you’re talking about for it. 

Speaker 1 

Right. And you can do it with tests. But it. But we prefer in the software development arena. One of the things about software developers is they want to understand why. And when they write it, they understand why, when it gets generated, they don’t understand, they don’t really understand why. That’s why it’s very important. The systems need to kind of reverse engineer and explain what they’ve done. If they become very, very automated. So there is an explainable ality factor. 

Speaker 2 

Perfect. 

Speaker 1 

That’s a crucial aspect of this system that’s very difficult. 

Speaker 2 

So from your perspective, the significance of generative AI in in the application of of software development in the future it it is going to be a catalyst to change. So so you you believe that the speed, the power, the agility. 

Speaker 1 

It’s massive. 

Speaker 2 

Expected to see is that going to be a standard? For our future. 

Speaker 1 

I think it’s going to be a yeah. Because if you think about. What we did from. From 20 years ago was really a modernization and and remove automation of the craft of software developing not only software development but also managing of the cycle of change and deployment. And there’s a lot of oil. There’s a lot of bad stuff. Stuff that you don’t really need to do. And Jenny I produces the capability of automating more of the tasks, more of the functions of that cycle. And so as the as the potential to elevate the role to a much more strategic, much more gratifying role of developers. If I miss it so our comma. Which is that 10s of thousands of developers today, 50% are proud developers. They are computer scientists, a lot of them from grade school, and one of the things they like is the fact that tools like these allow them to elevate to a point where they become more architects. They become designers of. Very large systems. They absolutely compress the time it takes to deliver business value. They like that they’re like, they’re like that, the pregnant is getting stuff done weekly. 

Speaker 2 

Right. 

Speaker 1 

And at the same time being. Able to design at the much higher level than marks the marks of. 

Speaker 2 

It it’s a significant shift if you think about it. If you think about the common developer today, they don’t necessarily go to school to aspire to the role of a. 

Speaker 

So. 

Speaker 2 

A development arc. They love the idea of coding and figuring out and challenging themselves on on solving problems. So what? Do. You. How do you speak to those that are coming out of school or that have been doing this for many years and they’re quite satisfied with have they been? Developing and letting them know, by the way your role was going to change, right? 

Speaker 1 

I’m a computer scientist, right? Like came out of time, I. Went to several schools. I’ve always felt software development as a mixture. Of artistry with engineering. There’s a lot of creativity going on. And what’s going to happen is there’s going to be more needs for software developers. But software developers that can really add value from a strategic point. 

Speaker 

Of view. 

Speaker 1 

In terms of high level design. Time. Thinking about what is the right construction of the system to provide a particular type of outcome, a real understanding of how some disruptions can be integrated to get the maximum benefit. And if you think about it today, if you. A lot of our. Communities made out of people after 10 years that have 10 years of experience, that have become tired of becoming support engineers. And being waiting up at 4:00 AM to go and fix the big attached to a piece of software for 10 years because there’s no one else who can maintain it. They don’t like that. That’s style. That’s not creative. And so. The future of this is going to elevate the profession where we think it’s going to be more strategic in one way and in some areas where you need very precision. Still you need to code, you need the. To go down a little bit like like what happened with the Elon Musk when the over engineer that tells the fact. 

Speaker 2 

Right. 

Speaker 1 

That he thought everything could be done with bots and then a lot of screws and a lot of precision work actually had to be complemented by humans. And we see that also in this new generation of the way software development is. Going to evolve. 

Speaker 2 

OK, perfect. I could talk to you forever, but I think we’re running out of time, so. Thank you so much for joining.