Max Peters, CEO of Adapta, on building AI agents for Brazilian SMBs

We've previously covered the multi-model AI workspace category through our coverage of Glean ($208M ARR, up 89% YoY) and Langdock ($25M ARR, up 925% YoY).

To learn more about how this category is taking shape outside the US, we reached out to Max Peters, co-founder and CEO of multi-model Brazilian AI workspace startup Adapta.

Key points from our conversation via Sacra AI:

To start, let's talk about what Adapta is.

The way we frame it, Adapta is the largest generative AI Latin American application. We're serving more than 100,000 small and medium businesses, and we are a multimodal AI workspace. The way to think about it is a place to centralize the use of AI for your entire company. Instead of having a bunch of different applications and having your context in a bunch of different places, multiplied by how many employees you have, you center everything in just one application, and we are it.

What was the original insight that led you to start it? Was it about educating people, or a worry that Brazilian professionals would get left behind as AI adoption moved so quickly?

It was actually all of the above. If you think about our name, Adapta, it's not by chance. Since the start, our mission was to help companies adapt and prosper in the age of AI. That's the long-term mission, the vision, the thing that doesn't change. How we do that is via our products, which are an evolving system that we started.

But the insight was pretty obvious: people and companies, especially SMBs, will need help becoming AI-native, having the tools and the knowledge and the capacity to adapt to this new age of technology.

You have an interesting background, copywriting, digital marketing, and a consumer products company right before Adapta. Can you talk about how that informed the way you approached AI and shaped the company?

Sure. My older brother and I have always been entrepreneurs. We have had a bunch of different businesses. We started with a tech business that required VC money to function. It was a fantasy game. Ultimately, the business model was not the greatest, as everybody now knows about fantasy games, but that led us to become really disciplined. For the next ventures we made, we said, "Okay, I don't want to depend on someone else giving me money for my business to function." So we decided to always do our businesses bootstrapped, without outside funding.

We did a bunch of different things. We sold two companies. One was a pet supplement company, which was really interesting to build. It was the largest DTC pet supplement company in Brazil. We sold it to Emerge Ventures. We also had different businesses in courses and marketing. And we were always fascinated by the concept of intelligence. That's why I've been writing a book about human intelligence since 2018. So of course I was always interested in AI too.

My brother as well. An interesting thing is that he was a professional poker player from around 2010 to 2015, and he decided to stop playing because he realized that AI was going to change everything in poker. He was one of the best poker players in Brazil, always winning, and he realized AI was going to eventually change poker forever. He said, "Okay, I don't want to do this anymore." Also, poker is kind of an ungrateful profession. You can be really good and still lose. So that also made him want to start doing businesses.

We were always thinking about artificial intelligence, and we know a lot about SMBs because we had a bunch of them. Probably larger SMBs, but still, it's interesting because we have backgrounds in many different kinds of businesses. So it was really obvious when AI was coming along. When ChatGPT came out, I took my team. We had been working together already for five years and were already selling the other companies. We said, "Okay, we need to focus 100% on AI because this is becoming interesting way faster than we predicted." So in the beginning of 2023, we decided to start Adapta.

And from the beginning, was it an AI workspace concept that you were pursuing?

The mission was the one I mentioned: to help companies adapt and prosper in the age of AI. The insight was, there is the API and we can use that API to create whatever product we want. So the first product was thinking about how I would make a better ChatGPT. What would it need to have for my profession? Because ChatGPT always had to be everything for everyone. It had to beat Google at being Google 2.0, 3.0. We were always thinking about work. From the start, I was thinking about work. I wanted to have an application for work. That was the first insight, because ChatGPT wasn't for work. Now it's better for work, of course, and now there's Claude, there's everything. But if you think about it, when ChatGPT came out, it wasn't for work. It was more fun.

We were always thinking about work. That's why we never had a free plan, for example. There's no free tier. We were always paid. So from the start, we were what some VCs would call prosumer; but never for consumers. That was the insight.

Has the core customer evolved over time, or has it remained the same?

It's kind of the same, but one thing that's clear is that the core customer, the person using it for work, is our top of funnel now. Where we're growing even faster is what we would call actual B2B. Think about it: you're an entrepreneur, you buy for yourself, this is cool, I don't have to have four different AI tools, I can concentrate my use in just one, Adapta, because Adapta has all the models. That's a very solid, easy-to-understand value prop. And then you say, "Okay, I'm going to do the same for everybody that works in my company." It's really straightforward, and we're growing a lot.

We grew more than 100% last year, and we keep growing 100% this year. But the thing that's growing the fastest is actual B2B, multiple people using inside just one account.

I noticed on your website that you have 4,200 different roles represented across your user base. So it sounds like it's extremely horizontal, not mostly salespeople or mostly marketers.

It is. But I'll say definitely the person buying it is usually the owner, the entrepreneur. He's buying it for himself, then getting it for everyone inside his company. Business-wise, it's clear to us that it's mostly services businesses. But still, that's a lot, right? I'm talking about marketing agencies, law firms, clinics, architects, real estate. There's a bunch of things, but usually services.

I noticed you can have a course-only subscription. Is that a user acquisition thing, or is it core to the user experience?

It's mostly just user acquisition. I'd say not even 1% of revenue is from that course-only subscription.

In terms of what people actually do, what user patterns do you see, copywriting, customer support, document analysis, image generation?

We see three layers of usage. **Amplification**; people doing their existing work better and faster, which is where most value lives today. And here lies everything you just mentioned: from copywriting to legal documents, presentations and most knowledge work.

**Systematization**; building internal tools, CRMs, dashboards inside Adapta using Skip, the company we bought.

**And automation**; agents and workflows running in the background. We tell customers: amplify, systematize, then automate. We've turned that into a methodology.

Doing more research, I wanted to ask about gross margins, a topic that's interesting to a lot of companies who use APIs in the background. How do you manage gross margins effectively? Is it like the gym membership model where some low-usage customers subsidize high-usage ones? Is it the routing models you develop internally?

That's a great question, and it's more the latter. Of course, there are always going to be people that don't use the gym, like you said, but when it comes to AI, that number is really low because AI is so useful and each day it's getting more useful. The percentage of people that buy and forget, the gym ghosts, doesn't happen anymore. So we have to be really disciplined and really good at the architecture of our software. At how we can provide the service as well as they would find on Claude or Gemini or ChatGPT, but being profitable and not burning cash. This is really hard to do, and it's one of our secret sauces. That's what we test every day.

We want the user to use our own model. If you think about Cursor or Perplexity, they do the same thing. Everybody that's doing a great job at being this agnostic application is doing a really great job at that, and we're included. Our UX, everything we do, is trying to get people to choose our own model, making our own model the best one. That's what makes it work.

You have some open source models too, which I assume are cheaper to serve. Is that part of the mix?

It is. It's everything, but the main thing is making people have a better experience choosing our own models rather than other models. It's the same approach the best applications are doing.

Let's get into the market. People might not know that Brazil has a massive internet population and a massive SMB economy. Would you consider AI adoption to be early in Brazil, and what are your takeaways that other markets might think about?

If you think about it on a global scale, Brazil is for sure ahead of the average. It's no secret. If you go to any big AI application, you'll see that Brazil is for sure top five on 90% of them. Of course, being a big country, I think we're the eighth or ninth, this is going to happen. But people in Brazil use the internet a lot, they use WhatsApp a lot, they use their mobile phones a lot. So it's not weird that AI has such high usage here. I wouldn't say it's early. I'd say it's right behind the US. Even on a per capita basis, the adoption in Brazil is probably even bigger than the US.

But it still has some peculiar things unique to Brazil. One of the things we focus on is having localized use cases, examples of usage, integrations, and everything that we can to help the companies here use AI better. That's the biggest thing. Besides that, the problems businesses have in Brazil compared to other countries are similar.

In terms of business model, it's very common in Brazil to pay in installments. Is that important to your business model, adoption, and monetization?

That's a great question. You did your homework. We sell just annually. We don't have monthly subscriptions. Why? Because it's an interesting thing in Brazil that we can sell the annual subscription as if it were a monthly one. So we charge 12 installments of, for example, R$99, and the user feels like it's a monthly installment, but we actually have the annual subscription. That's something singular to Brazil.

Our annual retention is really good. We talked to VCs and they were flabbergasted. We have really amazing numbers. Of course, this makes it harder to sell, because if you're selling an annual subscription, it's way harder to sell. But since we have such a known brand here in Brazil, since we do so much education, marketing, and content, this helps. It's not something easy to pull off, but we can do it.

You have the event series as well, right?

We have the largest AI event for businesses in Brazil. If you look closely, it's really similar to the playbook that HubSpot ran. We're doing that for AI and SMBs.

In terms of category, when you're talking to US VCs, what swim lane is it in? Enterprise AI workspace? Multi-model AI aggregator? AI education or productivity platform?

The way we frame it now, it's an AI operating system for SMBs, because that's what we're building toward. For example, we just did our first acquisition in January. We acquired a really interesting company that was the Lovable of Brazil with a really strong product and tech team. That fits well into our vision.

We want to build a system where the owner and each employee of the company can do 80% of their work in just one platform. That's Adapta. That's what we're building. It's also common that we use the word workspace here, because one of our projects is called Adapta Workspace. But in the end, what we see is this operating system for SMBs. That's the way we frame it.

Tell me more about the forward vision. I assume the Skip acquisition adds website creation. I saw you have Gamma on the platform too, which is deck creation. What does the vision look like moving forward?

We always want to be our own example. For instance, at Adapta, our own video team created their own kind of ClickUp. They created something that works for them with specific features only they need. The marketing team did the same thing. We have consultants that help people actually adopt AI and do other things inside our platform. They also have their own client portal that they created with Skip.

Now you can create your internal systems, CRMs, client portals, you can create landing pages, sites, everything. Then you can have the chat, the agent builder, the automations builder. You have everything. You can create everything inside just one platform. It's what we already do for chat. You don't have to have all the chats. You can trust me. You don't have to keep losing your mind switching different apps all the time. You can focus on me and I've got you. Now we're doing the same for everything, or at least 80%, being done inside Adapta. That is powerful, because then you have all the context from your company in one place. It's easier to grasp, easier to do everything you want to do.

That's what we're building toward, and it's already happening. It's not like this is going to be in five years. No, it's already happening. We're doing that inside Adapta itself, and we're helping companies do that.

That is really interesting. You mentioned context. If you're building with that vision, then you get enterprise search for free, so to speak. Because if people are doing everything already within Adapta, you don't have to have a thousand connectors for them to get context from other apps. As things get more agentic, how do you think about that?

Exactly. That's the point. Of course, we do have integrations. We have integrations with Brazilian software, which helps. But the vision we're going toward is that maybe you don't even need a bunch of these integrations, because everything's already inside, and you've created something specific for you. We're helping you do all of that, making sure that it actually sticks, that it's actually helpful for you. It's not just AI slop. It's everything together.

Glean is a company we've looked at a lot. Enterprise search was its first wedge. Notion is doing this now too. Is connecting to company knowledge across different apps something you had to deal with? Is that not as core to your product? Is it a direction you're going to go into more, despite what we just said?

It's all connected if you think about it. We are already doing some of that, but it's not the wedge that we needed because we're SMB-first. We're not enterprise-first. Although we do have enterprise customers, it's not the focus right now. In the future, even Notion now has a portion of its sales coming from enterprise, but it was an SMB play. HubSpot is the same. Even for the big kings in SMB, it's usual to first become the king in SMB and then start developing your enterprise sales. We'll probably do the same thing.

There are a bunch of really interesting things about being SMB. The feedback cycle is so fast. We learn so much, we can improve our product so much and so fast, that it's something really important for us right now. We're really good at doing that, so why focus on enterprise now?

But of course, we're already doing it in some ways. You can use Adapta Workspace as a sort of Glean. Even last week, for example, we do internal trainings all the time, so people are always trying to push our platform to its limits, because we don't even know everything they can do with it. People were showing how they had already created something like that for Adapta itself, for us, using our platform. So there are a bunch of things you can do already. It is already in the product in some ways. Like I said, if you're creating all the Skips, all the internal systems, it will become even more important for you to easily find things. It's something we're also looking into.

I assume if you create a Skip project, you can say, "Hey, connect to my Google Drive," and that system can connect to Google Drive? Or do you have a native Google Drive connector?

We have all the major integrations that you probably need.

I noticed Meta made it so that agents can't index WhatsApp. A lot of SMBs in Brazil run on WhatsApp. Is Meta making WhatsApp relatively closed to external AI systems something that's caused issues for you?

We haven't seen that yet. At least for now, I don't think it is here in Brazil. The way I see it, from Meta's perspective, I don't think they will close it. They might start charging. But why would they close, right? People might be flocking to Telegram. So I don't think they will ever close it, but maybe they'll try to monetize it better. They probably will, but not close.

Another example we thought of was Langdock in Europe. It seems the main wedge for Langdock is GDPR compliance, more auditing, data residency. Is that a trend in Brazil as well? Is there a version of that problem for Adapta where local procurement rules, trust, regulation, and local billing are a big part of your wedge?

That is not as important as it is for Langdock and Europe. I think it's more of a question there. For sure, we do have government clients and some Brazilian enterprises, and for them it is a bigger deal. We offer that for them. But it's not our biggest value prop or the thing that we lean on, at least for now. If we figure out that this is something really important that we need to deploy more resources on, then we will. But for now, it's a small percentage of our revenue and usage to focus on that value prop.

In terms of competition, Gemini, ChatGPT, Claude, Microsoft Copilot, and I assume a lot of local AI consultancies that work with SMBs. How do you think about the market that Adapta is competing against and how do you position Adapta vis-à-vis those competitors?

If you think about it at the macro view, there's always going to be harsh competition on the consumer side. The big players there would be Gemini with Google, ChatGPT, and Grok, probably. Then all the way to enterprise, the biggest competitors would be Microsoft with Copilot, where it has the distribution, the resources, everything. And you have Claude.

So you do have this middle ground, which I think we can be the best at, specifically in Brazil and then later Latin America, because we're focused on it. Of course, all of the players that compete on either side, consumer and enterprise, can get into the SMB market too. But we're focused on it, we're speaking that language, we're creating features for that type of business. So we have a good right to win in this specific market.

The big tech players are always going to compete for the best models. So that race is going to make sense, at least for the foreseeable future. The value prop of being agnostic on models, saying to the customer, "You don't have to think about which model is best now, because in two months there's a new model that's beating everyone, and we say you don't have to think about it. You have access to all the best models inside just one place,". This will keep making sense. And if it keeps making sense, we keep making sense, because we're agnostic to it. There's also this good value prop of being the center of everything you use inside Adapta, so you don't lose context across a bunch of different applications. You have everything under one roof. That's our right to win and our right to play.

In terms of pricing, I believe ChatGPT Go in Brazil has some price discrimination, trying to be more accessible. Google Gemini is heavily subsidized. I assume Android and Google Workspace are very popular in Brazil, so there's a platform advantage. Is there some combination of bundling and pricing that would make it very hard for Adapta to compete? Even thinking about ChatGPT having ads, an SMB running on ChatGPT for free because they put up with ads. Once model companies really start to go after the Brazilian market in the prosumer segment, would that make it very difficult for you to compete?

Not really. If you think about it, Google already has a killer offer. Google's offer is amazing. You have Workspace, Gemini, NotebookLM. It's so good. It has so many things that people don't even know about. The PMs at Google don't even know what they should be working on. It's so much that people don't know. The technology is just one thing, but you have to make people use it. You have to make people create value from it.

Think about Excel. Excel and Google Sheets are the biggest competitors to HubSpot, but why is HubSpot a multi-billion dollar company? Because they created a solution. It's not just about the tech. It's about the implementation, the go-to-market, teaching people how to actually provide value from it. We're on the same playbook. The model is not just about the application itself. It's about actually using it and getting value from it. Most companies can't do that just on the enterprise side. Then you have probably Claude doing some partnership with Accenture or BCG or whatever. That's for enterprise. But we created a way to do that at scale for SMBs, and we're the best at doing that.

Our results and how far we've come in so little time shows that, and this is going to make sense forever. The product from each of the big techs is already great, but the product itself is not enough. That's why we'll always have this space for us.

What does adoption look like? You mentioned it's usually the entrepreneur, the company owner, who gets Adapta and then maybe buys it for the rest of the team. So that implies an owner-led adoption at SMBs. Maybe consultancies play a big role. Or is it a lot of individual usage that eventually gets rolled up?

The pattern we see the most is the owner champion. The owner is the one buying it. He knows he doesn't want to get left behind. That's the wording most of them even use. They don't want to get left behind on this AI wave. They've seen other waves before, the internet wave, the cloud wave, the mobile wave, and they don't want to lose the AI wave. They know it's important. So they're the ones testing it and then buying our B2B workspace and making sure that people actually adapt and adopt.

This is important, the owner being the one doing it, because he also needs to make sure. Of course people are always going to be scared, like, "Okay, you want me to use this AI thing, but are you going to replace me with it?" So they have to do the teaching. They have to reduce the fears, the objections, and make sure people actually use it and test it out. Then we have all sorts of frameworks, trainings, and consultancy to actually help them do that.

What about consultancies and agencies? Are the folks on your team customer support, or are they like forward-deployed engineers? Does this consultancy aspect mean you're competing with AI consultancies and agencies, or are they also channel partners?

I wouldn't say competing. We have our own, but we are also creating a channel, kind of like HubSpot did. I wouldn't say they're customer support or anything. These are people that had their own consultancies and agencies. They were doing that by themselves, and they joined Adapta and started doing it with much more scale. They're really curious, young people, tech-savvy, who also understand businesses themselves, because they have to. It's a really interesting model. It's kind of like a forward-deployed engineer, but it's not actually deployed. It's more high-level, but it does get into details, into the workflows, into which systems the entrepreneur wants to build. We help them do that.

Besides that, we are creating this entire partnership ecosystem, but this is still in its infancy.

So the consultants are a post-sale type of engagement for maybe larger deals. Is that fair?

That's fair.

In terms of expansion, there's a pretty impressive history in Latin America of companies started in Brazil, Colombia, or Argentina expanding to neighboring markets, especially from Brazil. Is that something you're looking at closely? Why hasn't that happened yet?

There's no bigger case than Nubank, and Nubank did the right thing. They were really disciplined about when they would get the right to move to another country. We're doing great in Brazil. I was at a panel with one of the founders of Nubank, and he was telling the story of how they wanted to move to Mexico in, I think, the fifth year. One of the board members, who was one of the founders from Mar, said, "What are you talking about, man? You haven't conquered Brazil yet. You're at, I don't know, 20 or 30 million accounts, something like that. You haven't conquered Brazil yet. You don't have the right to go to Mexico yet." And he said how that was the right way to look at it, and how they went to Mexico just three years later.

That's the same for Adapta. The Brazilian market is way too big for us to think about Latin America. We have to be doing at least $300 million before thinking about moving elsewhere, and that should happen in two years or so. So we're not looking into it.

That sounds like a thoughtful approach. In theory, do you think this is a model that could work in a lot of markets? Or do you think there's something idiosyncratic about Brazil that makes Adapta make more sense for Brazil, and most other markets just take whatever is offered to them off the shelf?

That's a great question. This model can be replicated in whatever country you think about, but the thing is, can it be a big company in every country? That's the question. The model makes sense in a bunch of countries, but I don't know if it can become a big company doing that in most countries. Brazil is a big country, people love new things, they love AI, so it's really specific.

Of course, the model, people looking from the outside think it's easy, because you just grab the API keys and vibe code something. It's not that. If you think that, you're going to have a hard time. We've seen a bunch of players, even successful entrepreneurs, try to copy us here in Brazil, but they can't. Why? Because it's really hard. You have to be really good at lots of things at the same time. You have to be good at tech, product, marketing, sales, recruiting. You have to do all of these things in a really specific way. And we're doing all of that, and we haven't raised. We're profitable. So it's not easy. It's a bunch of work. In a way, we're competing with the biggest companies in the world, with the biggest talent pool in the world. So it's not easy.

It can be replicated in a bunch of countries, but we do have a really specific and interesting business here.

It sounds like in Brazil, there's a combination of relatively high usage and demand and a pretty large market locally. Those two are kind of prerequisites.

Exactly, large usage, large market. You can do it in different places, but it might not be that big.

How do you see adoption evolving in Brazil with the agentic turn? It feels like a fundamentally different paradigm from how we've used technology in the past. Do you think Brazil will leapfrog, as it has in the past, with agents?

That's a great question. I'll separate it into two things. The first is what we do most here at Adapta. We get a lot of clients saying, "I want to automate everything. I want to create 10,000 agents. I want to fire everyone." We say, "Okay, this is not how it actually goes." Actually, the vast majority of the value you get from AI is by amplification, not automation. That's a big thing we tell customers here at Adapta, and it's the truth. You'll get way more value doing things that you never could.

For example, right now, before sending any agreement to a lawyer, I will do my personal review first with the help of AI. So I'll get a way more informed conversation with my lawyer. Multiply that by everything. Now you have a zero-to-one in any field. So you're amplified. The things you used to do before, you can do better. The things you had no idea how to do, you can now do. So amplification is the thing, at least for now. It probably will be in the near future too. Automation is getting way better, agents are getting better, but the majority of the value is still on amplification.

The second thing, which is interesting, is that SMBs will be able to actually use agents before enterprise. It makes sense if you think about it, because in enterprise, you have to have lots of guardrails because you have more to lose. A data breach in an enterprise might be a huge loss. In an SMB, not so much. So SMBs are willing to take more risks, willing to have more speed in adopting things, willing to take more tries. So agentic AI will actually be faster in SMBs than in enterprise. Since Brazil has high usage and high willingness to try things, we will see lots of them doing and trying. And we're already seeing that. I'm not talking about the future. This is already happening with our customers.

If you think ahead five years, what does Adapta look like if everything goes right, and how are Brazil, the SMBs you work with, and the world different as a result?

How I frame our category right now is an AI operating system for SMBs. That is the vision we're building toward. In five years, we can definitely say we are that. For now, I'd say we're building toward it. We have most of the things you would need for that vision to be true now. But in five years, we are doing that. We are the first thing the business owner and his employees open on Monday. Everything is inside Adapta. The work that his agents did, the work that his employees did, his dashboards. Most of the things, 80% of the work, is in one place that has context for everything. That's the way I see it. And that means we have a business doing more than a billion dollars in revenue. That's the goal.

To close, in terms of the models you're building internally, which as you said is really important to your business model and supporting this vision, can you tell me more about how that works? Are you taking open source models and post-training them, or developing your own models from the ground up?

What we do a lot of is trying to find the best model for each prompt. We're optimizing that for quality and price. We're testing and creating our own ways of testing at scale, which answer is the best at a given cost for a given prompt. That's how we're doing it. And it's working. Our own model is the model that people love the most.

There's a bunch of things we're doing behind the surface. The goal is really simple: have our own model and the best answer for everything. How do you do that? Through a bunch of different processes and tests.

It sounds like a big piece of it isn't just having your own model. It's also the orchestration, meaning routing the prompts to whatever you define is the best.

Exactly. We're already trying a bunch of different things with open source and what not. Does it make sense for us to create small models? There are a bunch of different avenues we've tried, but as of now, the main way we go about it is actually orchestrating and having a system to do that in the best way possible.

This transcript is for information purposes only and does not constitute advice of any type or trade recommendation and should not form the basis of any investment decision. Sacra accepts no liability for the transcript or for any errors, omissions or inaccuracies in respect of it. The views of the experts expressed in the transcript are those of the experts and they are not endorsed by, nor do they represent the opinion of Sacra. Sacra reserves all copyright, intellectual property rights in the transcript. Any modification, copying, displaying, distributing, transmitting, publishing, licensing, creating derivative works from, or selling any transcript is strictly prohibited.