The AI Transformation Company for Latin America

A conversation between Alex Tran and Primero co-founder Jose Murillo.

In the early 1990s, SAP became the worldwide ERP standard. By the end of the decade, Salesforce had arrived as a global software force. A company in Monterrey and a company in Munich bought the same CRM and adapted themselves to fit it. Latin America was a market for software, never a builder of it.

AI breaks the one-size-fits-all approach, opening the door for regional champions. Deploying AI inside an enterprise isn't a purchase, it's an implementation. Now, the best forward-deployed engineers can walk into a company, map a tech stack nobody has fully documented, and make opinionated calls about what intelligence to build on top of it. That work is local, and it rewards taste.

Jose Murillo thinks that's the opening. Primero, which he co-founded with Santi Buenahora and Andrés Rosales, is building a system of intelligence for Latin America's largest enterprises: a layer that reads and writes across their systems, holds the context of how the company actually operates, and deploys agents on top. Between them, they have experience at Meta, Goldman Sachs, Robinhood, Y Combinator, McKinsey, and more.

GC's Alex Tran called Murillo to talk about why enterprise AI is the first software category that can be won regionally, what it takes to sell outcomes instead of productivity, and why a company's brain is the one thing it won't hand away.

The Conversation

Alex Tran: What led you to entrepreneurship in Latin America?

Jose Murillo: I grew up in Mexico City in a family of academics and government officials, so I grew up in a context where there was a lot of patriotism for Latin America. I believe that building a transformative technology company is one of the ways that I can have the biggest impact in this region. At Primero, we're building the AI transformation company for Latin America with world-class talent.

AT: Tell me about the talent. How did you meet your co-founders?

JM: Santi is the sort of engineer who has always been very hacky. His friends at Penn joke about how he earned his pocket money building bots that arbitraged trades on eBay. He loves building.

He's also someone who, a bit like me, grew up in a family who really cared about Latin America. So after working at Robinhood, he went to LatAm to start a company as a solo founder via Y Combinator. And then Santi and I met each other, both living in Colombia, became really close friends, and we decided to start our prior company, Samsam, which was a low-cost e-commerce marketplace, and learned a lot in the process. That's when I first partnered with General Catalyst.

My other co-founder, Andrés, grew up in Guadalajara. We became close friends as undergraduates at Harvard, where I was studying economics and computer science and Andrés was studying math and physics. Andrés had founded a company before, and he wanted to do it again with Primero.

AT: Why does Latin America need a regional champion in AI?

JM: It seems like a very contrarian bet, right? Because enterprise software has never been built regionally. SAP was the winner everywhere. Salesforce was the winner everywhere. So in the CRM and the data warehouse era, software was global. It was never regional.

AI changes that. In the age of AI, implementing AI at an enterprise requires a good chunk of forward-deployed engineering (FDE). The sort of companies that we work with have very complex needs and technology stacks.

One of our customers, for example, is one of Mexico's biggest insurance companies. They have a technology stack that includes over 300 different systems that touch any given process. And this is so complex that the company spends over $200 million every year just maintaining the software. Any time they want to build a new product, either internally or connecting something external, they need to deal with this messy tech stack. And what an FDE motion enables is a team that has the best taste to go into the company, figure out how to work with their tech stack, and build solutions that are custom to their workflows.

AT: Can you tell me about your hiring philosophy?

JM: We're very proud of the engineering team that we're building. We're really trying to build the best technical team in Latin America. Two of the engineers on our team hold the highest scores in Colombia's history at the International Collegiate Programming Contest (ICPC). They're very ambitious. They want to build something big.

Three quarters of the team at Primero either studied at a top US school—Harvard, Stanford, Penn, Chicago—or worked at companies like Uber and McKinsey. But I truly think this is the most talent-dense team in all of Latin America.

AT: What's the difference between a system of intelligence and a context layer? Can you dive deeper into the different layers of your product?

JM: In the past, if you were buying a CRM from Salesforce, that was that. You bought the CRM the way it was. What's different about AI is that every company is a bit different in the way software gets implemented, and for AI to work well, it needs access to the company's tech stack.

I haven't said context layer because I think of the context layer as one part of the system of intelligence. The context layer is exclusively the part that operates as a system of record, collecting context on the company's workflows and decision traces. The product Primero is building centralizes that context, but it also lets you deploy agents on top of it. Primero is a tool that enables enterprises to deploy agents across their organization with context of how the enterprise operates. This product today has three layers.

Layer one is helping companies deal with their very disparate software. These are enterprises like the insurance company I mentioned, and they can't deploy AI because those systems don't talk to each other. So the first thing that we do is help them connect those systems and have a layer that lets companies read and write into their underlying software.

Layer two, once we centralize that data, is writing down and defining the context that the AI needs to do work at the enterprise. This is essentially the company brain, and it would be insufficient to only have data from the systems of record. So much of the context inside these enterprises lives as unwritten rules and business logic.

And then layer three, which is really easy to understand: we deploy agents of all kinds across the organization using this context.

Each of those three layers requires taste, because there are many different ways to build them. And the product has this very compounding nature—it keeps gathering context and also capabilities as it connects to more of the stack, and both make it a lot cheaper and a lot more impactful to launch every next agent in the organization. If you're a big bank in Mexico, would you rather do that with us, or build a system of intelligence with a random dev shop?

AT: Most enterprise AI today is sold as productivity. You've talked about selling outcomes, not productivity. What does that mean to you, both globally and in the region?

JM: We really try to focus on workflows that are high ROI, where if the AI is doing a better job, the value comes from the fact that it avoids value leakage that wouldn't be possible for a human to find.

I can give you an example from one of our finance operations customers. We work with a company that processes more than 100,000 supplier invoices every month, and it's so easy for a human to make a mistake around those payments. It's way more likely that an AI doesn't make a mistake. And for this sort of company, saving 10 basis points on mistakes translates to millions of dollars in the bottom line.

In the past, when software as a service was sold to enterprises in Latin America, you would go in and say, "Hey, you should use this SaaS product because it's going to make your team 10% more effective." And I think in Latin America many companies have less budget for innovation, and that made them way less excited about purchasing software on a leap of faith that productivity was going to be valuable.

What's super different about AI is that it enables you to take a workflow end-to-end and automate it. And when you do that, there's a lot of value for the enterprise, whether that's through labor efficiency or taking a workflow and entirely doing it better than a human. But the fact that you can take something end-to-end is what enables you to sell an outcome and not just sell productivity gains.

There was a CFO at one of the CPG companies that we work with who said, "Oh, why would I work with you guys if I'm already paying Microsoft and I want to use Copilot? It probably doesn't make sense." And I told him, "Well, is Copilot helpful? Can you do actual work with Copilot? Does Copilot understand your ERP? Does it understand your accounts payable process?" He's like, "No, obviously not." And I said, "Well, that's where we come in."

AT: If you're right about where AI is headed, what does the next five to ten years look like—for the region, and for Primero?

JM: Open-weight and open-source are super important, and enterprises need to be model-agnostic. It would be a mistake for an enterprise to be married to a single frontier model for every workflow. The enterprises that we work with in Latin America are cost-conscious, and they care about these workflows being efficient on price. The customer that processes more than 100,000 invoices a month, or the bank we're starting a pilot with that processes over a million pages every month, what they care about is the unit cost of page processing.

The other thing is, a good chunk of these enterprises do care about running on-premise. Banking in Mexico is the obvious example, where regulation forces them to run on-prem. And a really good chunk of enterprises care broadly about operating within their own perimeter. Being able to use open-source models to do that is a huge advantage in terms of enterprise security preferences.

This is also why data sovereignty will continue to be hugely important. When you have a virtual company brain, that's your alpha. You don't want to hand that to a third-party model provider.

We want to be the company brain. We want to be the ecosystem where every module you deploy leads to a compounding benefit on the platform. And if we can be the trusted partner that runs this within their sovereignty expectations, that's a huge benefit of working with a company like us. So my short answer is that enterprises in LatAm are going to hugely benefit from implementing AI, and I think Primero will be their partner to carry out this transformation.