A empresa que a Anthropic escolheu no Brasil

Author: NoCode StartUp · Watch on YouTube

Video authored by the channel shown, not by IANIA; all rights belong to the creator. Transcript and summary generated automatically; the original-language content is preserved.

Original language: English

Original video: A empresa que a Anthropic escolheu no Brasil · Channel: NoCode StartUp · 1:04:10 · Watch on YouTube

Topics: AI agents, Enterprise AI applications, Task and workflow automation, Claude Code, Software engineering with AI

Summary

Indicium founder Daniel Avancini discusses the company’s growth from a Florianópolis data consultancy into a global data and AI firm, and its work with Anthropic to support adoption among large Brazilian enterprises. He explains the difference between traditional machine learning and generative AI, and describes how AI agents are changing software development while making sound architecture, process mapping, and documentation more important. The conversation covers enterprise use cases, security and governance, the productivity gains and limits of AI, and how companies can introduce the tools through training and focused pilot projects. For individual professionals, his advice is to get hands-on with a capable AI tool, build a practical project, and combine new technical skills with existing domain experience.

Key points

Indicium is a global data and AI consultancy with offices in the Americas and Europe, close to 700 employees, and a role supporting Anthropic’s local operation in Brazil for large enterprises.
Traditional machine learning predicts outcomes from historical, labeled data; generative AI learns to predict the next token in text.
Companies need reliable data foundations and documented processes before they can effectively build AI systems or automate work.
AI agents can automate tasks such as reviewing inspection videos and checking company documents, as well as help workers with existing tasks.
AI coding tools have made software development much more accessible, but architecture, product thinking, debugging, and understanding fundamentals remain important.
Enterprise adoption brings challenges around sensitive data, security controls, governance, internal training, and complex legacy codebases.
AI productivity gains depend on well-understood processes; a potential 90% reduction is described as a trend or possibility, not a universal present-day result.
Companies should assume employees are already using AI, provide basic training, identify low-friction tasks, and improve processes through real pilot projects.
Running many agents in parallel can create a supervision and workload-management burden for the human operator.
Professionals should learn by using a capable tool on a practical personal project and combine AI skills with their existing industry experience.

Chapters

Start	Chapter	Summary
00:00	AI, scale, and the changing nature of work	The opening remarks frame AI as a force changing software development and making knowledge work more scalable.
01:03	Introducing Daniel and Indicium	Host Mateus Castelo welcomes Daniel Avancini, who introduces Indicium as a global data and AI consultancy.
03:14	How Indicium began and grew	Daniel recounts the company’s beginnings in Florianópolis, its consulting focus, and its expansion into international markets.
04:19	Traditional machine learning and generative AI	Daniel contrasts supervised prediction based on historical data with generative models that predict the next token.
07:01	From predictive models to data platforms	The discussion covers early sales and marketing models and why companies need strong data foundations before advanced AI.
09:14	Services, expansion, and AI implementation	Daniel explains Indicium’s service model, international growth, and end-to-end work from consulting through implementation and training.
13:02	Practical AI use cases and enterprise challenges	Examples include AI-assisted video inspections and document review, alongside the difficulties of deploying agents safely and mapping processes.
18:19	Why building with AI still takes fundamentals	The speakers argue that easy prototyping does not replace process design, software architecture, product thinking, and technical foundations.
23:04	Claude Code and agent-assisted development	Daniel describes how newer models and tools changed his development workflow while noting the constraints of enterprise codebases.
26:54	Security, productivity, and the 90% trend	They discuss data safeguards and training, then consider current productivity gains and the longer-term potential of AI.
31:47	AI, economies of scale, and professional value	Using the quartz watch as an analogy, Daniel argues that AI will make knowledge work more scalable and that people should focus on what differentiates them.
37:10	Competition and combining domain expertise with AI	The conversation considers AI-enabled startups and explains why existing business experience can be a professional advantage.
40:23	A practical adoption path for companies	Daniel recommends accepting that staff will use AI, training them, finding useful low-friction tasks, and improving real processes incrementally.
43:32	Managing parallel agents and project specifications	Daniel describes supervising multiple agents and using planning documents and specifications to guide development before generating code.
51:24	Standards, Anthropic’s Brazil work, and governance	The speakers discuss evolving internal standards, Indicium’s support for Anthropic adoption, and enterprise governance and training.
58:13	Advice for professionals and closing	Daniel advises individuals to subscribe to a capable AI tool, learn through a practical project, and combine practice with study before sharing company and career information.

Quotes

just because you can build it, doesn’t mean you should build it. — 18:19

The work of software architecture or product design or product management, it hasn’t disappeared. — 19:55

the trend always wins. — 31:16

maybe you’ll be one of the best logistics professionals who knows data. — 39:18

what’s the point of spending 15 minutes to do it wrong? — 49:19

Concepts and entities

Indicium AI, Anthropic, Claude, Opus 4.5, Claude Code, Databricks, OpenAI, machine learning, IA generativa, agentes de IA, mapeamento de processos, specs, Light House

Full transcript (original language)

00:00 And now we are one of Anthropic’s first global partners. We have been working closely with them in England, in the United States, and now also in Brazil. When Opus 4.5 was launched, things clicked in such a way that suddenly you had a model that was very intelligent and had the right tools. And that changed the way of developing software. Basically, today we no longer touch code. Like it or not, AI provides knowledge for development. It’s not just code, right? It gives it an industrial, engineering, factory-like feel. Knowledge is increasingly becoming subject to economies of scale. Everyone is going to kill themselves to reach 90%. That’s why Anthropic, OpenAI , everyone is fighting, a battle, >> spending trillions of dollars, because

00:32 they know it’s possible to get there. So, everyone who is against this today will be run over. When the quartz watch was launched in the 60s, right, back when there were mechanical watches that you had to wind, and they invented the quartz watch. The first ones cost tens of thousands of dollars. Rolex had a quartz watch. In less than 10 years, there were quartz watches for a dollar, because they knew it was possible to build them for a dollar. It’s just that there wasn’t scale yet. So you have to understand, this is the trend. It won’t get there in a year, it won’t get there in a month, but it might get there in 10 years. We used to use this ladder reference a lot, right? Trying to jump from the first step to the last step. There is a whole issue of foundation, I need to understand the past, then follow my present, understand my

01:03 present to later be able to predict the future. And there are companies, many companies that still can’t even see the past. Welcome to the No Code Startup podcast. My name is Mateus Castelo, I am the founder of the AI school for non-technical professionals. And today I am with Daniel, who is the founder of Indicio, one of the largest data consulting firms in Brazil and the world, and he will share a bit of his story here and how he is using AI projects to accelerate his projects within the company. Hey, Daniel, how are you doing? Tell us a bit about your history, introduce yourself to everyone

01:35 . >> Thanks, good afternoon. Good afternoon, Mateus. Uh, thank you very much for the invitation. Uh, I’m very happy to talk to you and talk to the No Code folks. Uh, we have been working with professionals of all levels. Mateus will also be able to share a bit of his experience with Indium, right? With Indium in the past. >> For those who don’t know, I was the 12th or 13th employee. And how is it today? Yeah, I think Nocode is probably just as big, right? But

02:10 >> a mere fraction, >> but we’ve always really enjoyed working with both technical and non-technical professionals, right? We’ve always done this work of developing career transitions for professionals who want to enter the field, which seems to be a big part of the audience here on the No Code Startup podcast, right? But to introduce myself quickly, I am the founder of Indicium AI. We are a global data and AI consultancy, right? Today, we have offices in Florianópolis, São

02:42 Paulo, New York, London, and Lisbon. So we’re basically in the Americas and Europe. >> And how many employees today? >> We’re close to 700 employees. 700 employees. That’s a lot of people. >> A lot of people. >> Yeah. And we are today among the largest partners of Databricks, right? Which I think you guys must have talked about quite a bit on the channel. And now we are one of the first global partners of Anthropic, the developer of Claude.

03:14 >> Cool. We’ve been working a lot with them in England, the United States, and now in Brazil as well; we’re going to help them start their local operation for Anthropic here in Brazil, mainly focused on large enterprises. >> Hey, I want to explore that topic more later, but tell me a little bit about how Indicium started and how it is today. Like, in summary. >> Great. Well, we started the company here in Florianópolis, right? As you mentioned, you were one of the first employees there.

03:46 >> Uh-huh. >> And in the beginning, it’s been basically almost 10 years since we started this journey, right? And our start was really thinking—we aren’t technical, at least I and our Americas lead, Mateus, aren’t technical, right? We aren’t from a software background, so to speak. So, our focus was really on consulting. And supporting companies in using data at the time—mostly data , although I have an economics

04:19 background, and let’s say what today is no longer called that, but back then it was called data science, right? >> Uh-huh. >> And it’s not as fashionable anymore, but I had a background in data science. >> I was already working with AI, so a long time ago. >> Already working with AI, right? Machine learning. Exactly. >> Today, people need to separate things. So we call it classic machine learning, >> uh, >> versus Generative AI, right? But yes, we had that, uh, that background. Just a point you mentioned, just explain to the folks what the difference is, very briefly, so people understand, because

04:52 sometimes folks get quite lost in these concepts. >> Good, good. Uh, well, basically when we talk about machine learning, or more traditional AI, we are talking about being able to interpret patterns from large volumes of data. Yes, >> right? And these patterns usually require you to do supervised learning, as we call it. So, in simpler terms, we have a series of past data that says, for example, when a person enters the

05:25 site and navigates a certain way, staying for more than 5 minutes. So, what is the probability of this person clicking a button or adding a product to the cart? Uh-huh. >> So, we take a series of historical data on how many people accessed the site, right? And we try to make the model predict what the probability is for the next people who haven’t accessed it, whom we don’t know, to click the same button, right? So, we have the past as a base, we have a >> a variable that we have already measured. So, the keyword would be trying to predict what will happen

05:58 based on the past. >> Predict and supervised, because we have the click data, we know what the click was. >> I see. Okay, >> right? Whereas in Generative AI, which is the basis of ChatGPT, language models, there is no such final click itself, right? >> Ah, yes. >> So, what we do in Generative AI is we take text, >> take books, historical texts, and we simulate this by doing what? We hide a part of the text. >> Uh-huh. >> And try to predict what the next text would be, in this case, the next token,

06:29 right, which is a >> it’s part of the word, really. So, it’s a different view. Yeah, first of all, it’s much more text than numbers. >> Uh-huh. And second, the way of doing this prediction is very different. >> But you were already working back then with machine learning and starting to offer consulting services in this area, so how did that evolve? >> Exactly. Exactly. We started by building models linked to marketing and sales for regional companies, let’s put it that way, right?

07:01 >> Uh-huh. Thinking a lot about sales forecasting, churn prediction, right? When you have a client who stops buying from your store or stops subscribing to your product, demand forecasting models , for example, for stores, store positioning in a mall, what is the best store to place in that mall or that location, or what are the best—it’s what we call spatial models, what is the best place for a new store, a new supermarket for chain X based on the

07:34 consumption profile of the region, right? Right, on the roads, on the streets, right? So, that’s what we did and still do, right, for many clients. But over time, we ended up moving much more toward the platform side, more of the technical side of the business. So, in the beginning, we wanted to generate more insights, maybe much of what the audience here on this podcast does, right? Like using data and tools within the business area. Yeah, but as we kept

08:09 working, we realized there was a lot of work to be done on the foundation. >> I see. >> You can’t just build these tools, because even though companies want to do machine learning and AI, often it’s the final stage of an entire journey, right? >> Exactly. >> People want to skip several. >> People want to skip. We used that stair analogy a lot, where people try to take that giant leap, trying to jump from the first step to the very last one. There is the whole issue of foundations , and now, data has become a bit more

08:43 hidden than AI because of the AI hype. But we always talked a lot, in the world we work in, about needing to understand the past, knowing my past, then tracking my present, understanding my present to then be able to predict the future. And there are companies, many companies, many departments in companies that still can’t even see the past. >> Uh-huh. >> Right? And so, for you to be able to see the past, especially in a very large company, you have to have a data platform, a well-built data structure.

09:14 And we ended up going a lot into this side of structuring as well. >> Did you go more towards the software side or the service side, delivering a platform, how was that business model? >> Service. Service, >> right? We work with software and with large data platform companies like Databricks, like the cloud providers, right? Right? So we decided not to develop our own platform. We use, uh, we help clients, especially large companies in the enterprise space, to use their platforms correctly, to bring

09:45 data into them, to configure security, governance, etc. >> Uh-huh. Cool. And so, how was that evolution? And then you started winning big clients, then you went outside of Brazil, which was a big leap in recent years, right? >> Yeah. Yeah, so my vision is that if you do good work, things start to appear, right? You dedicate yourself a lot to >> to looking at what is most advanced and so on. >> So, we always looked a lot abroad, always looked a lot to—I think you even participated in some things like

10:18 that in the past, right? Yes. We looked , we were very small, but we already looked, like, hey, what is being done in the United States, what is being done at >> at the main companies and so on, and we tried to bring that to Brazil as quickly as possible. And then over time we managed to start getting into some larger companies, right? We started regionally, we started getting into larger companies in São Paulo, some clients with famous logos, etc. Then we started expanding here and also saw the opportunity to work on projects in the

10:49 United States, doing the same thing there. Since we were already using the same technologies, it was somewhat easier, right? >> Then with the pandemic and the remote world, it was like the world became one single world, right? Everyone was working remotely, so it made no difference working from Brazil to the United States or from Florianópolis to São Paulo, from Recife to São Paulo. We have a lot of collaborators in the Northeast, for example. So we managed to go to the United States, we started working there, then we received an investment from an American fund.

11:21 >> Cool, >> right? We expanded our presence in the United States, and now at the end of the year, we merged with a company in England, and so we also entered the European market. >> Cool. Always with that focus on services, right? AI services, besides data, how is that going today? Are you offering Generative AI services? What kind of service is proving most interesting to these companies? Yes, we do, we try to do what we call end-to-end, right? From consulting,

11:54 advisory—a fancy name for consulting, but it’s because consulting can mean many things, but the idea is to really support people on what to do, right? So , what is it that I need to do? Where can you give me a scenario analysis of where I stand today? So, we have this consulting work, and then we also have the hands-on implementation work, right ? For example, data platforms >> and now a lot of AI. So, what is this in AI? How can I bring AI agents and

12:29 things like that into my company, into my operations? Uh, then doing it in practice, building agents, bringing, for example, cloud code to >> my software area, training the people. So, we implement and then we also help enable and train the people, the teams, the clients to use it, right? And just to make it clear for the people watching, are there some examples like, “oh, we made such-and-such agent that does this, we made such-and-such automation that changed a company”? >> There are, there are several, let’s

13:02 think of some cool ones here. Interesting ones, >> yeah, >> thinking about day-to-day, there are some really cool things. So, for example, there is a project we did—it wasn’t even in Brazil—where there is an energy company, let’s say, that needs to go to a house to do a to do an analysis of, for example, of the clients, of the consumers, for a number of legal reasons. This is filmed, and then they need to ensure it was done correctly.

13:33 >> Uh-huh. So we take the images and use AI to follow an entire checklist. The AI analyzes the image, compares it with the checklist, and says: “Look, this image, this visit was done well, this one was not done well.” So imagine that you would have to have a human looking at two hours of video for each one. Uh-huh. >> And all of this is done in 1 minute, 30 seconds by AI. That’s one example. >> Wow, cool. >> We have projects, for example, many projects related to documentation issues in Brazil. Large companies have a lot of trouble with, uh, reviewing

14:04 many contracts, for instance, that don’t match what’s in the system. This is extremely manual, extremely boring work. We can, we have projects where agents do this and ensure that things are, are working. >> This is still kind of back-office, right? But >> there are agents for building commercial proposals, there are uh, there’s the actual work itself, these are, let’s say, applications, right? >> Uh-huh. But much of what we’ve been focusing on now and supporting clients with is not necessarily building products with AI, but how I use AI to

14:38 do the work I was already doing before, >> right, which is what we call productivity. >> Which is where no-code comes in a lot, for example. And I imagine a lot of what you teach here at No-Code, right? >> Cool. Yes. Yeah, in our case, it’s usually for people who are either creating their first project, want to create a product, want to create a SaaS , there’s this more entrepreneurial crowd that’s just starting, but now a lot of companies are also appearing. Generally, our focus ends up being small and medium-sized businesses. And folks focus quite a lot on that, right, on productivity, on being able to

15:11 automate areas, automate processes. So we are exploring this area a bit more, which before was very much about entrepreneurship, right? And now it’s a bit more in the corporate environment. From what I understood of you guys, it’s a bit more for large companies already. >> That’s it, 100%large companies. Yeah, >> yes, 100%. >> Yeah, no, our focus is on enterprises, so in general, they are companies that have a revenue of at least 1 billion reais, let’s imagine that. Uh, >> cool. So, it ends up being larger companies. Not that we haven’t worked with smaller companies, we, you know, started with companies of all sizes,

15:41 >> but it turned out that over time we ended up, uh, targeting larger companies, right? But the problems aren’t that different, >> right? >> I would say that very large companies have much more bureaucracy, processes, political difficulties, mainly because >> sometimes you’re talking about a legal department of a company that has 700 people, 500 people. And what has been the pain point for these large companies? Do we want to be more productive in the legal department here to review contracts and automate this entire process? So it’s basically

16:13 process automation in departments. >> Yeah, I think back-office process automation is safer. >> Uh-huh. >> So, as good as it is today, putting it in a client’s hands, right? I mean, even ridiculous examples of people using a pizza chatbot. I won’t mention the client’s name here, the company’s name here, >> but people using a fast-food chain’s chatbot to solve their coding problems, because you send the prompt and the system returns the result. So people are like, “Oh, you’re out of tokens, just use the

16:45 >> chatbot from restaurant X, they’ll answer you,” you know? Because it’s really hard to control it 100%. And humans are very creative. >> Yes. And when you give a human a challenge, for example, “I want to break this system,” there are people who will spend 24 hours a day trying to break it, just to post on Twitter that they broke the system. >> They’re dedicated. >> So people are truly dedicated, right? So when we talk about the end customer, there are so many guardrails, there’s

17:15 so much, right? There are many complexities in putting it out to the end customer. When we talk internally, it’s easier, because whether you like it or not, there’s, you know, >> they won’t reach that, there won’t be that incentive, right, to break the system. But then another problem comes in. You mentioned a very important word : process mapping. It’s very difficult to find good process mapping. So >> I’m inside a huge client talking about legal, talking about CRM, talking about these systems, but there’s no well-defined process, like, so much

17:48 stuff is in people’s heads. >> Uh-huh. >> So, no matter how good the model is, no matter how good the technology you’re developing is, it won’t decipher what’s in people’s heads. So, we have to be able to get what’s in people’s heads into documentation, into a system. So that’s very much a process job, right , production engineering, those things, more engineering, >> than necessarily AI and technology. >> So that’s the challenge for big companies, but it’s also the challenge for smaller companies, even a family business. >> Yeah, I think it’s everyone’s challenge

18:19 , right? What I see most often is people wanting to skip several steps and create an AI agent to connect to WhatsApp right away. Skipping several steps and automating something that doesn’t even have a process for it. You’re automating before you have the process. What is the step-by-step? Make a checklist, then you can train and follow that process, right? >> Yeah. And there’s a phrase people use: just because you can build it, doesn’t mean you should build it. >> Uh-huh. >> Right? AI made it very easy to be able to build. You can build things faster and easier. >> Uh-huh. >> But you shouldn’t always build it. You have other things to do first.

18:51 >> And I also think an important point you should talk about is that it’s very easy to make a prototype. >> Uh-huh. >> It’s not that easy to make something that will work every day for all scenarios. >> Yeah. And that’s where the whole software development structure comes in , right? Not that you can’t do it alone and not that it isn’t easier today than it was 5 years ago. It’s much easier. >> Yes. Yeah. >> Right? But just because you started now doesn’t mean you’ll be able to do it in a week. You have to understand a bit

19:21 about how software is developed and so on. Maybe today you don’t need to learn programming like it was in the past, so hardcore. You need to know how >> to debug, find a bug, a problem within the Python code, for example. AI does that for you. >> Yes. >> But you knowing how you organize your work, how you organize a SaaS, for example, >> Uh-huh. how you organize the modules, the product features? >> Basically an architect who connects all the puzzle pieces, knows how to join

19:55 and orchestrate, right? >> The work of software architecture or product design or product management, it hasn’t disappeared. It’s even more important than it was before. >> Yes. No, you hit on a really cool point that has always been our mission at COD startup, that we wanted to democratize access and the creation of technology for non-technical professionals. >> And we had no-code tools, right, like Bubble and FlutterFlow, which still exist today, >> but we never managed to carry on with that mission, it was difficult, it

20:26 wasn’t the reality for companies or people; it was possible, but it didn’t democratize, and nowadays with AI, it has become much easier. So, basically, it’s saying that everyone is a builder now, right? Everyone can build now because of low-code and cloud tools. So , everyone can do it. Now, what sets people apart is understanding the fundamentals, process mapping, what each technology does, and how they interact to actually build projects that aren’t just MVPs, but are truly useful and scalable.

20:59 >> 100%, 100%. Anyway, I don’t want to discourage anyone, right? I don’t, especially because everything is possible to learn. I mean, I’m perhaps a practical example of that too, and you are as well, right? We aren’t trained programmers, we didn’t spend >> 5, 6, 7 years in programming, and we managed to work with this and build this, but it wasn’t in a month, it wasn’t in a week, and it wasn’t without effort. >> Uh-huh. >> Right? We have to study, we have to

21:30 look for, for example, if I want to develop software, how is software developed? What problems does it address, right? I’ll take a course, watch a YouTube video, watch one, buy a book on software development. >> Yes. >> Those books aren’t programming books. >> Uh-huh. >> If you pick up classic books on programming, on software development, they aren’t books with a bunch of— there are a lot of lines of code there, but there are >> concepts, there are, you know, best practices. Exactly. How you’re going to do anything, right? And when you think of architecture, for example,

22:00 engineering and so on, right? It’s rare today for an architect or an engineer to have to do calculations by hand. You’ll use software, some CAD program to do it, >> but you have to understand the basics of calculus so you can understand what it’s doing, right? >> Uh-huh. >> It’s the same thing. It’s easier to change things incrementally. So, when you already have software that’s somewhat solid and say, “Hey, I want to add a new button, right? I want to change the color of something.” That is very good, it’s very easy to do. But when you say, “No, I’m going to create a system, a SaaS, for example,” and

22:33 even more so for those starting out who say, “Man, to make money, I need 200,000 users.” Because that’s very common in SaaS, right? You start a SaaS there, but in the beginning it’s cheap, and so on. So you have to scale this tool sooner or later, and then you depend on a bunch of problems, not just infrastructure, >> Uh-huh. >> cloud, >> yeah. >> Data platform, databases, >> that if you do that very poorly, very wrong, it won’t work, it’ll start to

23:04 slow down, it’ll break. >> Uh-huh. >> And then you start getting mad at the AI. But it’s not the AI’s fault, right? It’s just that >> behind the computer there. >> It’s a bit about who is providing, who is writing the prompt for it. Exactly. >> Yes. Great, Daniel. Let’s talk about cloud code, since I’m curious because you mentioned you’re already using it in some projects within indicium, I don’t know if there are internal projects inside the company. I also wanted to know about external cases, how you see cloud code and similar tools within companies, the good things , the not-so-good things, the main challenges. >> Cool.

23:35 >> How is it going, >> man? It’s great. Great. Uh, so I’ll try this: before cloud code, right, you must have, I remember you did some courses on cursor and other tools before, right, Windsurf, anyway. >> Yes. >> So, this this this area, this idea of development with agents, it actually predates cloud code, right? Uh, the thing is that in the last six months, basically since November of last year, some things have happened, right? One

24:10 is that in the past, the idea before Cloud Code was very much that you had to develop these agents, they would use some very organized ways to identify what’s in the project, using rag, using these things that are a bit out of fashion today. And with cloud code, the idea was like, man, you don’t need to do anything too complex, you just need to let the agents search on their own. You give it tools, you tell it how to search for files, how to read files, how to execute things, and it will do it by itself. It already had some good

24:43 results at the beginning, but around November, when Opus 4.5 was launched, it was the new model from Anthropic, things clicked in such a way that suddenly you had a model that was very intelligent and had the right tools. >> Uh-huh. And this changed the way we develop software, like, >> night and day, >> night and day. So, we tried. For example, last year, we ran tests using these tools ourselves and such; it worked for some cases, but it wasn’t like, “wow, I’m going to replace the way I work,” or “I’m going to change my day-to-day workflow.” After cloud code,

25:18 Opos, etc., mainly starting in January of this year, let’s say, it’s very recent, it’s been 4 months, but it feels like an eternity, it changed a lot, just like at Anthropic when they talk about this, right? Basically, today we don’t touch code anymore, this is becoming widespread, I practically only use cloud code for almost all my development, because it actually gets things done. Now I know what I’m doing when I’m developing with it, right? This architect’s job. I said, it’s not just telling it to do it, you have to

25:52 talk to it, understand what it’s doing, right? So it is a great assistant that handles much of what was hard to do in programming, where you had to understand the code, it does it for you , but you have to guide it >> in the right way, right? >> Yes. And we did start using it both for internal development and in several projects where we can use it, right? Of course, it’s not simply replacing everything, let’s say, now we’re only

26:22 going to use cloud code for everything. >> Uh-huh. >> Especially in large companies, because there are many complexities, right? >> Bureaucracies too, right? There are bureaucracies to use it, there’s a series of internal things that make it much easier to start a new project than to work on a company’s codebase that sometimes has millions of lines, right? So >> it doesn’t hold the entire context of the software. You have to tell it to read such documentation, read this, read that. These are the restrictions for how to put this into production. So

26:54 >> there are some challenges. A question, when you go to do a project, for example, using cloud code within companies, >> how is this relationship with sensitive data there? What will the cloud code be able to access? If companies are open to this or if they are very strict, which for me is one of the main problems there. That is a big challenge . It always has been, right, in the world of consulting or services as a whole. >> Uh, there are ways for you to protect and ensure data security, both when using cloud code and when humans are working. So, there are, you can,

27:30 depending on the account, like with Anthropic, configure things so it doesn’t train on your data. You cannot pass keys and secrets to it. You have to do it >> using this credential reading the right way, right? I don’t want to get into anything too technical here, but >> you don’t want to pass a bank password to cloud code, even though it’s unlikely to appear anywhere, but for security, you don’t want to, right? Uh, but look, there are ways you have to create within a very large company as

28:04 if they were, it’s like a prison, I don’t know if it’s a good example, but you have security perimeters >> controlled. security perimeters. You have the immediate perimeter, let’s say , the cell for a specific piece of data . Then you have the block, then you have the wall. So you keep creating several perimeters so that in case there’s a problem with something, you have a security layer, just like it’s done in almost everything in a large company regarding network systems and infrastructure, right? >> And training, right? teaching people how to use it, saying what they can’t

28:35 do, and how they can. But I feel that large companies today, just like most of the people listening to this podcast , know there is so much value in using these tools, right? I’m always asked how much I generate? There are projects where you can get 30%, 40%productivity like that, without much effort. >> Or work hours, >> imagine that I used to spend 10 hours to do something, now I’m spending 7 hours, 6 hours to do something. >> This is without much effort. What do I mean by that? I’m not automating

29:06 everything yet. I still have someone pushing the button, someone looking at what it’s doing and reading it. If you can create a very well-rounded work process, which isn’t easy and not just anyone can do, and it doesn’t happen in a week, because you understand it as you use it. There are people who joke and say: “The first week using cloud code, it’s like, oh my God, it’s over, it’s over, I’m going to get fired, my life is over, I have no more work.” >> Uh-huh. After three or four weeks you say: “This thing is so dumb, I’m all

29:37 set, my job is safe.” And after two or three months, you say: “Hmm, maybe I should be a little careful, because” >> as you keep using it, you go through waves of illusion and disillusionment, right? Illusion of “wow, it’s amazing,” then “wow, it’s so dumb.” But it’s neither of those, right? It is very intelligent, very powerful; you just have to know how to use it. >> Uh-huh. But if you know how to use it within your world, so, for example, you say: “Ah, there are many people here who aren’t devs and work in a business area or a company with a very streamlined process that you understand

30:11 deeply,” >> sometimes you can reduce 90%of your work. >> Uh-huh. Have there been cases like this in companies, would you say, more personally >> that it already accelerated >> that I reduced 90%of the work. >> Uh-huh. You know, in a company, with all the complexity of adoption, it’s much harder to reach 90%, but it is possible to get there. And so, one thing I say at the company is that there is reality, and there is the trend. >> Uh-huh. >> When we, so, when we work in a lab, right, which is what we try to do

30:41 inside, at least I try to do a lot, in what I manage within the company, I say : “Man, let’s stress test this to the limit, to see how far we can go, right? ““And sometimes, depending on what it is, you can reach 90%.” >> Uh-huh. >> You can even reach more than that. So, what does that mean? It might take 10 years to reach 90%because there’s a series of problems that need to be solved, but if it can be reached and there is value in reaching it, it will be reached. >> Uh-huh. Right? There will be a path where companies, processes,

31:16 universities, everyone will adjust because there is a lot of value in reaching 90%. Today, perhaps the reality >> for large companies is 10%. >> Yes, I understand. >> In 3 or 4 months, maybe it will be 20%. >> Uh-huh. >> But that is the trend. So, why do I say this? Because those who aren’t in this world yet, who aren’t learning, who don’t know it, and don’t care much about it, have to understand that the trend always wins.

31:47 >> I see. >> If you don’t prepare yourself, you’ll be left behind by it, right? Because it is possible to reach 90, someone is going to get there. Let me give a very, I like to use this same example, but, eh, it has nothing to do with AI. But, to understand the problem, AI is, like it or not, it applies to knowledge, to development. It’s not just code, right? To knowledge, everything that involves words, that we use words to solve.

32:19 >> It adds a touch of industry, of engineering, of manufacturing to it. In other words, knowledge is increasingly becoming subject to what we call economies of scale. I am an economist, for those who don’t know. So, >> yes. >> And here is an example. When the quartz watch was launched in the 60s, which is the watch we have, everyone has one, since there were mechanical watches that you had to wind, and they invented the quartz watch. The first ones cost tens of thousands of dollars. Rolex had a quartz watch. >> Uh-huh. In less than 10 years there

32:52 were quartz watches for a dollar, because it was known that it could be built for a dollar, it’s just that the scale wasn’t there yet, but the trend was so strong that after 10 years they were worth nothing. So much so that today any watch, you can buy a very, very cheap watch anywhere for just cents. >> Uh-huh. >> And that is somewhat the problem with AI here, you see? When you say, it is possible to reach 90%, even if it’s in a lab, everyone is going to kill themselves to reach 90%. That’s why Anthropic, OpenAI, everyone is fighting

33:24 , >> the battle, >> spending trillions of dollars, because they know it is possible to get there. So, everyone who is against this today will be, will be run over. So, you have to understand, that is the trend. It won’t happen in a year, it won’t happen in a month, but it might happen in 10 years. So I need to adapt to that and understand where >> the value is, where the value is in what I do, what I know how to do, and what I need to learn so I don’t have to do it anymore. >> Uh-huh. You say that applies both to the employee, the professional, and

33:54 also to the company, right? Whether companies are adopting AI or not. >> Both. Exactly. Professionally, individually, me, Daniel, me, Mateus, >> I need to understand what I do, what makes me different, and what buttons I press. Anything that is just pressing a button, I’ll likely have a problem, because these tools press buttons very well. >> Uh-huh. And not just pressing a button like in the past, like no-code, where sometimes you’d do something, but it took work, and anything that wasn’t

34:25 within those rules we defined would break. >> But these tools go way beyond the obvious, they go much further than those other tools. So it’s a much more complex button that they press. >> Uh-huh. >> So we have to understand, man, I’m a product guy, what do I know about product? What kind of input can I provide? What can I do differently? How can I study the product side better than what I know? How can I study my market better? >> Uh-huh. >> And leave the part, for example, of making the best slide in the world to AI, because AI makes slides very well, it’s already learning to make slides, >> it’s already good.

34:57 >> So if my business is making slides, I have to understand the trend is that this is worth less and less. die. >> If my business is instructing it to make the best slide, storytelling won’t die. Storytelling won’t die. I mean, making a bunch of slides will die. Now, making three slides that say everything >> is an art. >> Yes. >> So that’s what we have to understand as individuals. >> And as a company, it’s something similar, understanding that if today I have 100 people whose main job is making slides, how much of that is an

35:30 art? Is it something that generates a lot of value for my company, and how much of that is simply things that have always been here, processes that exist just because they have to, >> yes, >> and not something that really generates value. This, again, will have…it might not be you, it might not be your biggest competitor today, but there will be someone, it could be a startup, it could be someone watching this video here, this podcast, someone who is starting out, who wants to say: “Man, I’m going to open a…for example, when I look at an insurance company, right, there are 5,000 people at the insurance

36:04 company who are reading documents and that is obviously part of the cost of the insurance, it’s in the insurance premium.” >> Yes. If I were to build an insurance company that has 100 people instead of 5,000, I could probably charge lower insurance premiums and win the market. So, if you are an insurer with 5,000 people or a lot of processes, you ask: “How do I use these 5,000 people for more useful things and not let this startup come in, as has happened in other markets before AI, and in other

36:37 areas, right?” Banks, for example, where people came in saying: “Hey, I don’t have branches, so I’m much more efficient, I can offer cheaper rates, and I’m going to steal the market,” and it happened. >> Uh-huh. So today with AI, if I’m a company, I have to look not only at what I and my competitors are doing, but that a new company could come along , a startup could appear. So people say this will be the biggest explosion of entrepreneurship in history, because anyone can look at a process or a business and say: “Man, 90%of this, in theory, I can change,

37:10 >> so I’ll start a new company and win this market.” >> But then there will be a lot of people starting new companies too, right? Yeah , competition will go up, there’s that point. >> Exactly. So, >> there are two problems. A, it’s the timing; whoever starts first will have a better chance. >> Yes. >> And there’s the difficulty too, because it will be a super competitive market. Because, and there’s another point I think is cool, just going back to what I said about processes and such, which is, if you think using the cloud or using an equivalent from OpenAI or

37:41 Google, etc., is a competitive advantage, it’s not. Because anyone can subscribe, >> but it ties into what we said, right? Anyone is a builder now, anyone goes out there and builds. But the difference, I think, is even in these fundamentals, these slightly more technical concepts, right, to know how to connect them. And for me, for example, a product manager role, the person who knows product processes, understands the theme, can unite all the intelligence of creating the product, of launching the product,

38:11 combined with technical fundamentals, but the low-code is unstoppable. >> Yes. Eh, and that’s a point even going back to the world of data, right? And and over at Indicium, uh, we also have our own academy, right? We’ve trained a lot of people in data, with courses related to data and BI, dashboards, analytics, etc. What I used to tell people who wanted to make a career transition, for example—a lot of people want to, and do want to

38:45 transition into the data field. >> Uh-huh. Uh, I would always say: don’t throw away your experience, right? Sometimes a person has 15 years in marketing, 15 years in logistics, 15 years in the financial sector. >> That is their unique differentiator, right? >> Exactly. I’d say: “Man, no, you aren’t going to be, you aren’t going to be the best data professional in a month, with the experience of a one-month course you’re about to take .” >> Uh-huh. >> But maybe you’ll be one of the best logistics professionals who knows data.

39:18 Exactly. Now you’ll know both. So, bringing it all together. >> Exactly. So, if you can combine your business experience, what you know how to do, your life—right, your life experience—with this technology and know how to bridge the two, then you have a competitive edge. And many people forget that. People forget what they had in their background >> and think, “No, I’m going to dive headfirst into this new tech and I’ll know everything about it and I’ll become a millionaire with AI,” right? It might be; some succeed, but you have to think that just as you have three

39:51 months of coding, there are thousands of other people who will have three months of coding. But, right, how much you’ll really be able to differentiate yourself >> is is harder. >> Cool. And Daniel, man, a tip you would give, perhaps a bit more focused on small and medium-sized businesses— which I think is most of the crowd watching—to start inserting code into their routines, into their processes, to gain productivity. What would you say, like, phases or steps, like: first thing, I’ll map the processes. Now I’ll

40:23 start using code, more individual and simple automations. Then I’ll use a whole AI operating system here; what tip would you give to these folks? I think you have to parallelize things a bit, right? >> At the very least, you have to realize one thing: everyone is already using it . You have to assume that everyone is going to use it. First, there’s no point in trying to block it and saying: “In my company, it can’t be used.” People are going to use it, >> they are already using it. >> There’s no way to control it. Because if you say you can’t use ChatGPT, you

40:53 can’t use Claude, the person can just take their phone, take a picture of the screen, and use it. >> Uh-huh. They can talk to their phone, or make a video of the phone showing the screen. So it’s practically impossible for IT, for example, the classic approach, to say: “It is forbidden to use tool X in my company.” So, first, assume that everyone is using it. So, the first thing is, let’s train everyone, let’s hire a course from you guys, hire a basic course, at least so the staff has some notion of what they are doing and doesn’t, right? Not put confidential information there

41:23 into the system, etc., not share a wrong prompt there, and so on. So, the first thing is to reach a minimum level of team qualification, then find the low-hanging fruit. There’s a lot of low-hanging fruit, as we say, right? There are certainly areas that people don’t like to work in, >> manual tasks, like, >> manual, tasks that were necessary, are necessary, but that nobody likes to do, and if we eliminated this process, automated this process, it would be a gain for everyone. Everyone would be

41:56 happy. Uh-huh. >> So, I think starting with the “happy” stuff is always good, right? Like, “I need to fill out this form here that everyone spends a good amount of time on.” Man, filling out forms with AI is a marvel, it fills them out super well. So let’s fill out forms, right? And so, in this work of mapping processes, right? You can hire an external consultant or train a company employee >> to be, let’s say, a champion inside there who will >> radiate knowledge. That is very, it works very well. There are many people

42:27 wanting to enter this world. So, sometimes, hey, I’ll take this person here who likes technology, they are already doing this in their, in their personal life. Let’s train them a little more, take them out of their day-to-day a bit and say: “Look, your job now is to try to adopt these technologies in here and teach other people how to do it.” This works very well. And then do this process work, etc., but with real cases. It’s no use just staying, “I’m going to map every process in my company.” That doesn’t work. >> Uh-huh. It’s slow and it’s a drag. So I

43:01 have this process that’s a real pain, nobody likes doing it, I’ll take someone, train that person or hire an external consultant, and I’ll fix this one first properly. >> And little by little, right? Look at what’s most critical and focus only on that one, because sometimes I think people get paralyzed, you know? Since it’s all cloud code, so much to automate, it feels like you’re way behind, sometimes just focus on one single thing and keep testing and going . There’s a problem people have been talking about a lot, which is that supposedly AI reduces our workload. We were just saying here that it can be reduced by 90%.

43:32 >> Yes. >> But if you do 100 more things, you get just as overloaded anyway. >> Yeah. And everyone is doing extra things, right? In theory. >> A lot. I personally had a time a while back where I had 12, 15 agents running at the same time. >> Uh-huh. >> You know? So I was like looking at 15 things at the same time, but there comes a point where your head explodes because we don’t have the capacity to supervise them. It’s the same problem as…Whether you like it or not, like, at the company, in my area there must be about 350 people.

44:03 >> That I lead, not directly, there are leaders, managers, etc. But there must be about 350 employees in my, in my department, right? That’s, it’s hard. You have to have managers, coordinators , you have to have a promotion process. Uh-huh. >> Right? Performance reviews, all these HR things, let’s say, >> it’s the same thing with the agents. >> So, many people are starting to, the developer or the person is starting to suffer from the same problems of managing people. >> Uh-huh. >> Because you have eight people working.

44:35 Are they doing what’s needed? Am I giving the right instructions? They’re idle, and you say: “Oh my God, I’m leaving this agent idle here, I’m being , >> I’m being a bad manager because it’s not doing something.” So you have to be a little careful not to get overloaded yourself. >> Yes, >> because now that you can do too much, you end up doing even more than you can handle. So that’s also a challenge. >> Out of curiosity, what were these agents doing while you were managing them? >> So, various things. Sometimes I have, for example, an agent doing research, so let’s imagine I’m developing something, software, a product.

45:06 >> Uh-huh. So, sometimes I have one agent developing one part of the product, another agent developing another part, another feature. I have another agent that is, for example, fixing a bug or looking or testing things, right, to see if they are working. And sometimes there is a >> all in parallel, >> all in parallel. And sometimes there is even another agent that is, for example , doing research. So, >> for example, uh, I have a project there that is, let’s, let’s say, an agent orchestrator, which is what we do in our work there. So, how do I make the

45:37 agents work the same way so that my team can develop in a standard way and so on? There are several orchestrators. So, sometimes I’m there doing things, I say: “Dude, research what others are doing new, because there’s a lot being released.” So, sometimes, every two weeks, one week I say: “Now I want you to look at everything that’s being released.” Is there any good idea? Is there anything we could be using? Since in the open-source world things are open? >> Yes. So there’s a series of things you can do in parallel, which in a software team ends up replicating how a standard

46:09 software team would work. So we would have the product people who are doing this research or the designers. >> I have the developers who are developing the features that I already developed. I have the QA and testing personnel who are testing and debugging and fixing what is going wrong. >> Uh-huh. >> Sometimes there is already someone who is visual. So, for example, if it’s an app that has a front end, you can have another agent that is just looking at the UI, at those things. >> So, it’s not, there isn’t a single way to do it, but this was the way that I

46:39 >> that I I want each, let’s say, each agent to stay focused on one type of work to >> objective >> for me to organize my head of what I need to do. >> Cool. Because it can, each agent can do everything, but if I let them do everything, the problem isn’t the agent , the problem is the human, the operator who goes crazy. >> Yeah, lots of tokens there, right? >> So many tokens. So many tokens. Yeah. >> And do the agents talk to each other or do you manage them? >> I manage them in general, they communicate through the artifacts they build, through the code they make, and so on, right? Everyone will be >> cool,

47:11 >> let’s say, uh, every once in a while you have to sync, right? Uh-huh. >> to make sure everyone is looking at the same thing, >> but I don’t, I personally never saw much advantage in making these models where one agent talks to another and so on. >> Yeah. And a question regarding standardization, for example, do you have a standard for working with your folders, your structure? >> How are you doing that at Indício, I don’t know, for projects to maintain this standard and for other builders to be able to use the same standard you are using or work on the same project as you, which might get a bit more

47:42 technical, right? But if you could give a general view, >> so there are some standards, uh, it’s a very very volatile area, >> right? Very dynamic. >> Yeah. >> Uh, what I’ve been seeing and liking is the idea of using specs, right? To use them as if, you know, you describe what you want to do before you start doing it. So, you don’t just say: “Make a page for me, make a, make a SaaS for me , >> right? Not a CRM SaaS for civil

48:13 construction, for example. I wanted, I want to make a new CRM for civil construction. No, you say like, look, you first start defining with, with the agent using another one, I’m going to brainstorm. Man, I wanted to build a SaaS for civil construction. What does a civil construction SaaS need to have? Ah, it needs to have such and such modules. Okay, cool? Where do I start? Which module do I have to make first? And you go on doing this brainstorm and saving it in documents. >> Uh-huh. >> Right? In files where you save this research, this documentation, this architecture of what you want to build, as if it were a business project. And

48:47 then you can use some some uh packages, plugins, for example, there’s one called specit, uh, Superpower, it’s one from cloud that’s famous, that people call, >> that helps you organize this thinking, >> right? And up to the moment you say:” No, now I’ve more or less designed what I want to do, I’ve already stress-tested it with the AI itself, if it makes sense, for how many users I want, then I’ll send it to be really made. “ >> So you have a big process beforehand, right? >> Exactly. Many. Sometimes I get >> more important, right?

49:19 >> Sometimes I want to do something, and I spend the whole week just talking to the team, just talking about what I want to do. >> Uh-huh. >> And that’s because doing the actual work often takes 15 minutes. Writing the code itself, building the screen, but what’s the point of spending 15 minutes to do it wrong? >> You’ll have to redo it over and over again. >> So that’s one of the points regarding the problem with AI that people talk about. It’s very fast at doing something you don’t need. >> Because you’re the one who knows what you need, not the AI.

49:49 >> Uh-huh. >> Right? And there at Indicium, I’ve been working a lot lately on building, for example, even an open-source plugin —I won’t mention it yet because it’s still in development—but developing my own version of what I find useful in these plugins. >> Uh-huh. >> A sort of” super-power, “but geared toward what we do. Yeah. >> Because then there are our processes that I’m…not just ours, but what I see as the most useful processes that I’ve been, I don’t know, two, three months building and testing on other

50:21 things, and saying:” Man, this doesn’t work very well, that doesn’t work very well. “ >> Uh-huh. >> Because, even using these plugins, for example, you start after a while…in the first week it’s brilliant, you think:” Wow, it’s so fast! “But later , you want to change something, you want to do something more advanced that depends on a bunch of things you already did, it starts to get harder. Sometimes it’s even better to start over, right? >> Sometimes it’s even better to start over. >> And in theory, what I think—I don’t know if it’ll work, I don’t even know if it will ever work—but I think that if you can develop this initial

50:51 part of architecture and specs well, you could throw away all the code and tell it to rebuild. >> Uh-huh. >> And it should work. Then you tell the AI:” Look, it’s all already mapped out. “So, that’s the vision of what I’m building. I’m building something where, as I build the code, I’m also building the reasoning behind why that code exists. >> Yeah. >> So that in the future, who knows, I can throw away all the code, give the reasoning back to the AI and say:” Here’s the reason, you build it for me. “ >> Uh-huh. Actually, Anthropic launched

51:24 yesterday, or the day before, a new feature for Claude called” outcome, “ where instead of telling it what you want—instead of telling it the” how, “like” build me a screen “—you say:” I want you to make a SaaS, “and it tries to get there without you explaining much of the how; it figures out the how itself. >> It does the >> optimizing for the result you want. >> Uh-huh. >> Which seems easy, but it’s not easy. It’s hard to do. >> A huge challenge. >> It’s quite difficult. Another question here: if there’s a dev or a builder who wants to create a project,

51:54 do they have to follow an Indicium standard, or do you already have a pattern for structure, processes, and folders, just like you mentioned? Do you spend a week talking, or is there a replicable structure for the team members or not? And how do you do that? We have a few, because we’ve been making versions over time. As technology changes, you have to redo everything. That’s one point. There are things we did last year for this that were very important, which this year are no longer useful >> because the technology has changed. So, we’re reinventing this all the time

52:25 and building plugins, agents, and sub-agents. And today, the main thing we’ve been working on, which I’ve been working on personally, is this: building this standard, just like there were other packages for the data field before AI. Even so that when I go into a project, I know where to start, I know what I need to understand, and it’s not just a blank screen. Because with AI, the blank screen is a human problem; there are so many possibilities that you don’t know where to start. So, how do we organize this? >> Uh-huh. >> Yes. And so I’ve been building what,

52:57 like it or not, is somewhat similar to what OpenCla does. So you have memory there, you have tools that can be used, and it’s geared toward the enterprise world, the software and data world that we work in, right? >> Which isn’t necessarily the same as what a smaller company needs; sometimes there’s more bureaucracy than in a smaller company, >> so I have to have more >> guardrails and protections in this world. Yes, we have, >> because it’s something I see as quite useful, usually for small and medium-sized companies, you know, since I have more of that perspective here, for them to have their standard >> and be able to have other employees

53:30 follow that standard to have their own environment there for cloud code >> and deliver good results, you know, so that it doesn’t become a situation where everyone does it their own way. It’s like we talked about a little earlier, right, uh, Castelo, uh, it’s good that we start in one place and scale up, because if we have everyone working at the same time >> Uh-huh. >> it starts to become a mess, right? Everyone starts doing it in a different way, then you did it one way, and when I go to work—and suddenly in companies, people change, right? People

54:03 leave. So you have a project there that was done by someone that has no documentation at all, it was the Cloud that did it. Uh-huh. >> It’s all in that person’s head. It’s difficult, isn’t it? So it’s good to have these standards, but it’s also no use starting just by creating standards , because again, you never reach any results. So the cool thing is to build standards based on some real projects the company has and scale that in parts . That’s what we do at the company, right? Indíci, >> right? we take a project, we say:” Let’s take a project, and with the

54:35 project happening, with the reality of this project, we build our own standard , >> understanding what works, what doesn’t work for us, which isn’t necessarily what works for other companies. >> And it varies a lot, standards vary a lot, right? >> And something new pops up all the time, right? So it’s crazy. Yeah, that’s one of the biggest difficulties. Even people say you have to, uh, slow down, they say, you have to hold back a bit. >> It’s because, in fact, it’s too fast. Uh, and sometimes it’s even bad for adoption, you know? People get lost. >> Uh-huh.

55:05 >> Because it’s very fast. Uh, there are things that don’t work that well. So, like, several tools we use still have a lot of bugs, there’s a lot because it’s very, very fast. So, like, and, and, what, but what I think is: people need to take a, stop a little and think about what they already have today or, including a lot of what is being released, I don’t even use, because what we already had last month and I already know how to use is already very good. >> Uh-huh. So I say: “Man, first let’s

55:37 just use what we have.” I learn, unless there’s really something revolutionary, but every day 10 new skills come out, 10. No, you don’t need to use all of that, you know? >> Yeah. Learn what works, what works for you, and use it. Give it some time to mature. You say: “Man, I already know how it works, it’s working, it’s cool, let me test something new, let me try testing this other function, this other tool.” There’s no point in always wanting to test the next tool, because then we never get anywhere.

56:08 >> Uh-huh. Cool. And are you partners with Claude, with Anthropic here in Brazil? >> Yes, yes. Yeah, we’re supporting them, Anthropic now in adoption at large companies, right, workshops, uh, >> internal training, right, because there’s a lot of interest, like, >> even very, very large companies that >> uh, even surprise me a little, they were mostly more conservative about using these tools, but everyone wants to use them, everyone is after it, so

56:41 as a consultancy, we have to support these companies to do it right. >> Uh-huh. >> A lot of what we discussed here, >> is usually on the data side, right? >> It’s data and also software. >> Software. >> Software too. >> Uh-huh. >> There are also processes in the area of automation processes. >> Uh-huh. >> But code generation and well-documented processes are still easier. >> Uh-huh. >> So companies are still starting a lot with code, but there is a trend, right? For example, Anthropic is launching

57:11 Claude for the financial sector, Claude for the healthcare sector, >> yes. >> Which then already has a series of things for these sectors that are very specific to them and not focused on code, focused on day-to-day, for example, creating >> uh, documents, you know, linked to PowerPoint, those things, which is, for example, the world of the financial sector. >> Yes, >> right? Uh, we also support that, but our focus has been more on helping, let’s say, the governance part of the

57:42 tool, how to monitor what is happening, ensure there are no leaks, right? Train professionals on how to use it. >> Seems too challenging. >> It is quite challenging, >> yes. And one of the last questions, Daniel, man, what advice would you give to a professional, an employee who’s at a company, wants to start using it, but is so lost with a bunch of tools, a bunch of launches, so many new things, and doesn’t know where to start? >> We talked a bit about the company, right, but now for the professional, what would be their step-by-step in

58:13 your view? >> Good, great. It’s a good one, a good question, okay? But I’ve been saying this internally, I’ve never said it like this, I’ll try to put it this way. I think people need to consider now that having access to a good, paid one —because it’s different than saying, “Oh, I have ChatGPT free; it’s not the same thing.” >> Uh-huh. >> But having access to a good one and knowing how to use it is the same thing as knowing how to use the internet. You have to consider it in your personal monthly budget. Don’t wait for the

58:45 company to give it to you. Subscribe on your own. Get a Claude license. Go online, put in your credit card. “Oh, it’s 100 reais a month, that’s the cheapest one.” Man, how much does the internet cost? 200 reais a month. You have to spend it. >> Uh-huh. >> You have to do this. You have to know how to do it. It takes two, three weeks for you to start understanding why the thing is good. First, there’s that. It’s not on day one, it takes two, three weeks. That’s why it boomed in January, because people said that over the year-end holidays, they took the time off to use it and said: “Wow, it’s really good,” because you need time,

59:17 right? And you need to do something cool. Make an app; there’s a professional at the company who you even know who likes fishing. He decided to make a fishing app, he said: “To identify in the competition who’s catching the best fish,” he put one in that identifies the size of the fish by the photo, and it’s completely a hobby. >> Uh-huh. So I think that people have to take it, if the company doesn’t have it or still has a lot of limitations, do it personally, do a personal project, make an app for, I don’t know, I like

59:49 soccer, I’ll make an app that does something with soccer, right? Learn to use it and treat it as if it were an everyday tool that you can’t escape from. >> Uh-huh. >> So I think that’s the first tip. You have to jump in, enter this world >> and do a practical project, >> a practical project, know how to use it . >> And if there are limitations at work, do it personally. >> Uh-huh. >> Don’t wait, because those who wait too long will be left behind. >> It would also be good to learn the

1:00:22 fundamentals, the basics, right? >> Absolutely. And again, I think if you want to go further, everyone has their own way, right? It’s just like college, some people like to write everything down, some people don’t take any notes. >> Uh-huh. >> In college, right? So yeah, people learn in different ways. So, if you’re a hands-on person, just get started, play around, just do it. Oh, no, I really like to know, to understand better. Take courses, buy a course. >> Uh-huh. >> Either way, you’ll have to study more. It could be by taking your course, a

1:00:54 YouTube course, the official training from a platform, but when to do it, I think that varies. Some people prefer to start with a course, others prefer to start with practice. >> Uh-huh. >> Either way, you have to do both. If you want to be a complete professional, >> the cool thing is that people often get very lost, right, with so much stuff, but just make the payment, use it for the basics, sometimes just really use the cloud, maybe move to Copilot, which is a bit of a middle ground, then move on to coding. So just taking those steps like that >> really helps a lot. But alright, Daniel , we’re wrapping up our podcast here. I

1:01:28 think the talk was very good. I’d love to stay for another two hours to explore what’s on your mind regarding agents, cloud code, schedule another chat, thank you for coming, and I wanted you to mention your social media or the Indícium social media, whatever you’d like to share with the folks. >> Sounds good. I’m the one who should thank you for the invitation to, uh, have a chat with you guys, right? And I’m very happy about the success, right ? Uh, of seeing the community growing. Uh, I’m also very interested in participating in other conversations. And so about social media, right?

1:02:01 There’s the company’s official LinkedIn , which is Indícium AI. Uh, there’s mine too, I think it’s Daniel Avancini, I think. >> Uh-huh. >> I’m not exactly sure, but you can search for Daniel Avancini, you’ll find me. And there’s also the Daniel Avancini Instagram, those are my socials, my professional Instagram, let’s say. So, those are the places to find me and, of course, feel free to send a message, chat, or connect, I’d be more than happy to exchange ideas. Uh, there’s also the company YouTube channel, which has some things we post

1:02:33 about AI and data. A lot of the content is in English, so that’s an important point. Yeah, see there, everything in English now. So fancy. >> Yeah, a lot of things are in English. So, uh, but also there’s Indício Academy, our learning platform, where things are generally in Portuguese, so there’s plenty of content there, including cloud training for data. >> Uh, we have some cool stuff there too. Uh, and of course, I can’t forget to mention this, or I’d be killing myself, right? We have several open positions for those who want to work in the field , even those in a career transition, as

1:03:06 you mentioned, that could be a large part of the audience. Uh-huh. >> So we have openings in data engineering , data science, AI, project management, a series of open roles, and it’s growing a lot, both for working in Brazil and in some cases abroad. >> Uh-huh. So, oh, I want to work in Europe, if you have, for example, a European passport, that really helps. So, we >> have been hiring people to work on projects abroad. >> I think you guys are one of the biggest trainers of data and data science professionals, right? >> We’ve trained a lot, we’ve trained over 2,000, I think over 2,000 people, maybe 3,000, something like that too. And now

1:03:40 in June, I think June, we have another selection for our Light House cohort, which is our training program, that is very focused on career transition, right? And we always run two cohorts per year. >> The last one had 8,000 applicants. >> Wow, that’s competitive, isn’t it? Competitive. >> But it’s very, very good, like, many people have gone through the program, a good part of the company came from there. So, anyone interested, keep an eye on our social media. Yeah, I think

1:04:10 there’s even a pre-registration page for the program there. >> Cool. We’ll leave the link here in the comments. >> Well, thank you very much. Thanks to everyone watching us. If you have any questions, leave them in the comments below and we’ll answer them for you. Follow Daniel there and I’ll see you in the next video. See you later, everyone . Great.