Only half awake, I’m mesmerized by the shadow of el cerro Tatamá, the highest peak of the western mountain range, located in Tatamá National Natural Park, which rises above the fog creeping over the Cordillera Occidental. I can’t distinguish the morning fog from the steam rising off the pocillo of hot, rich, black coffee in my hand. By the third sip, I’m fully present, and my teammates in New York are starting their day too, with Austin just an hour behind.
Currently working from Colombia, I share most of their working hours. That overlap, along with bilingual talent, is what Latin America offers a US team looking south to build an AI team. Still, the distance on the map is only part of the story. We can share a working day and still struggle to share the work itself: the code, the models, and the environment that lets a colleague pick up where we left off. When packages, models, and environments don’t match from one country to the next, one team quickly becomes three. Here’s what a distributed AI team needs to share to work as one.
Look no further than south
Are you building in the US and looking to ramp up your team? Will you look across the world, or in your own time zone? Pick any South American capital: it’s at most two hours from New York. English proficiency is strong in the region too: Argentina ranks first in Latin America and 26th in the world on the EF English Proficiency Index. Bilingual, highly skilled talent is available.
Imagine your team with engineers in Mexico City, Bogotá, Lima, Santiago de Chile, Buenos Aires, São Paulo, and New York, a few time zones apart, working on the same product, with work calls relatively easy to schedule.
In 2012, David Vélez, a Colombian, couldn’t open a bank account in Brazil; in 2013 he founded Nubank with Cristina Junqueira, a Brazilian, and Edward Wible, an American. Thirteen years later, it has more than 140 million customers. Its engineering network now spans at least eight countries across the Americas and Europe, from Uruguay and Argentina to the United States and Germany, and since late 2025 it has been hiring engineers in Toronto. It also calls Python “a core part of our stack, particularly for building AI agentic systems.”
Nubank has begun looking north: in January 2026, it received conditional approval to operate as a national bank in the United States, and in September, Nu launched in the United States, with Junqueira as CEO of Nu US. Nubank shows what Latin American talent can ship.
The gap
You could drive from Anchorage to Buenos Aires on the Pan-American Highway, if it weren’t for the Darién Gap, about 106 kilometers (66 miles) of tropical jungle where the road stops. The jungle behind my morning fog runs north to that gap, on the border between Colombia and Panamá.
El Tapón del Darién (the Darién Gap) is a fitting metaphor for the lack of bridges in tech, between our countries and between North, Central, and South America. The goal is to build strong teams that work together across borders.
When code has to cross borders
Geographical proximity is only the starting point. The engineering bridge is a shared way to choose dependencies and models, control agent access, and reproduce a teammate’s work. One engineer builds an AI app in a notebook in Bogotá, another wires a model in Rio de Janeiro, and a third gets it ready for production in San Francisco. Which package versions did they use? Where did that model come from, and what does its license allow? Who checked what the agent can reach? Can anyone rebuild the environment next month?
Consider four illustrative failures in that shared project:
Package versions: The Bogotá notebook runs, but a dependency resolves to a different version in San Francisco. The environment builds, yet the application behaves differently. The team needs a shared record of the versions and configuration that produced the tested result. AI Artifacts provides packages built from source and signed, each shipped with a software bill of materials (SBOM).
Model provenance and licensing: The engineer in Rio integrates an open-weight model with promising output, but its source and permitted commercial uses have not been checked. The team needs to identify the exact model and review its license before relying on it. AI Artifacts also includes models with documented provenance and licensing.
Agent access: A coding agent can reach repositories and tools beyond the task it was assigned. Even working code can conceal excessive permissions. The team needs explicit access boundaries and a review of what the agent can read, change, or execute. AI Security & Guardrails, with Enkrypt AI by Anaconda, red-teams models, agents, and Model Context Protocol (MCP) connections before release and applies guardrails at runtime.
Reproducibility and handoff: A month later, the production team cannot reconstruct the tested environment or identify which model revision was used. A maintainable handoff needs the environment, model, dependencies, and changes recorded together, with validation on the target infrastructure. AI Workspaces give builders approved models, agents, and reproducible environments in the tools they already use, Kilo included, and AI Orchestration turns projects into repeatable, traceable workflows that move from development to production.
How a team handles these four decides whether a promising demo becomes a system another team can maintain. Together, these capabilities make Anaconda Platform the trusted foundation for the AI Dev Factory your team is already building.
Back in Bogotá, that engineer could open the shared project in Kilo Desktop, in the same environment her teammates in Rio and San Francisco use, choose a model from the team’s approved options, and ask an agent to help write a change or its tests. Before passing it to San Francisco, she reviews the code and runs it in the project’s shared environment. Kilo helps her work with models and coding agents; the environment and controls described above help her colleague reproduce and review what she hands over. The policies, access controls, and budgets her company set apply to her work too. Having more than one model ready matters too, as I wrote in “What Colombian Coffee Growers Can Teach Dev Teams About Model Lock-In.” If you want to see what your builders would get, explore AI Workspaces.
Tools can take care of the code, but the teamwork making the dream work? That’s a whole different story. My team is fully remote, a blessing and a curse. I miss the water cooler and the kitchen talks, the trust you build working side by side every day. Tools keep our work connected, but they can’t replace trust. It takes looking each other in the eye, a warm smile, a cup of coffee brought to a colleague’s desk when you know they’re in dire need of one.
Alas, the last sip of coffee. The fog over Tatamá is thinning, and its peak looks somehow closer now than it did a while ago. A yellow butterfly flutters by, its wings resplendent in the morning sunlight. I can’t help thinking about Gabriel García Márquez and his 1982 Nobel Prize lecture. Gabo hoped that “the races condemned to one hundred years of solitude will have, at last and forever, a second opportunity on earth.” He was referring particularly to us, Latin Americans.
The nutty-caramel taste still lingers as my hope grows stronger while I think about AI teams that span the whole continent. The opportunity is already here. The road may never cross the Darién, but our teams can.
When you think about where to build next, will you look south? If you’re building and deploying AI with teams across borders, request a demo of Anaconda Platform.
