Alan Descoins

Alan Descoins keeps an unfashionable belief in a fashionable industry: that artificial intelligence is a piece, not the point. He runs Tryolabs from Montevideo, where the office faces the Rambla and the work goes out to Sephora's pricing team, Hyundai's engineers, and a UNICEF dashboard that maps children sweltering through heatwaves. The list is wide on purpose. He prefers the word "tailor" to "vendor."

The TIME100 AI list arrived in 2025. The neighbors on the page included Sam Altman, Elon Musk, Mark Zuckerberg. Descoins's reaction was to describe it as a crown on fifteen years of work by many hands. He meant it. The collective phrasing is not modesty as PR move. Tryolabs began with four people in 2009, when the words "machine learning" still required a definition before a contract.

He picked up programming at twelve. There was no online course, no bootcamp, no influencer telling him to learn Python. He read, broke things, fixed them, and kept going. By university he was studying Computer Engineering at the Universidad de la Republica, the kind of public-school path that does not normally end at TIME magazine. It did anyway.

Tryolabs's first decade was a long argument that prediction was useful. Clients wanted dashboards; the team wanted to build the model behind the dashboard. They published. They open-sourced Luminoth in 2017, a computer vision toolkit that ended up in talks at PyImageConf and on hobbyist GitHub stars charts. They wrote blog posts the way other firms wrote case studies. The blog turned into pipeline.

Then came the hackathons. In 2019 the team flew to Hong Kong for the Allianz Global Investors AI hackathon and came home with first place. The story of the win sits on the Tryolabs blog like a yearbook page. Descoins's name is in the credits, not the headline. That is the pattern.

Today Tryolabs is over a hundred people. The mix is roughly half enterprise work - Nvidia partnerships, Hyundai engineering, retail price optimization for Ulta Beauty - and the rest is the kind of work that is hard to bill at corporate rates. The UNICEF heat-exposure tool. Lion tracking for The Nature Conservancy. A model that finds rural schools in satellite images so they can actually be funded. He has been clear that the nonprofit work is core to the business, not a charity line item.

Descoins's pitch on AI strategy is steadier than the average LinkedIn carousel. Think of the model as a component. Optimize for the system, not the demo. The hard part is integration, not the algorithm. He has watched the conversation shift from "what is machine learning" in 2010 to "how do we scale responsibly" in 2026, and his answers have not shifted as much as the question.

The most quoted line in his Spanish-language press is sharper. "The one who is going to take your job," he told Montevideo Portal, "is not ChatGPT. It is someone else who uses it better." It is a line you can imagine on a co-working space wall in Pocitos. It is also, you suspect, the line he tells his junior engineers.

He posts on X under the handle dekked_. The output is mostly technical, occasionally dry, never thirsty. He shares Tryolabs's weekly AI roundup. He talks about model quantization, edge deployment, the gap between paper benchmarks and production. The follower count is not the point.

Inside Tryolabs the work has a specific texture. The team does deep learning when deep learning is right. They will also tell a client to use a simpler model and save the GPU budget. The "tailor" metaphor turns up in nearly every interview Descoins gives, and it is doing real work as a metaphor: a suit is sized to a body, not stamped from a die.

The Uruguayan context matters too. Montevideo is not the obvious place to lead an AI shop. Power is cheap, talent is fierce, time zones overlap with the East Coast, and there is none of the burn-rate theatre of Silicon Valley. Tryolabs has used all of it. The firm describes itself as a global consultancy with a Latin American base. Descoins has used his TIME profile to argue that more should follow.

What he says he wants next is unflashy. Be the AI partner clients keep through every shift. Hire well. Keep the nonprofit work alive. Ship things that survive past the slide deck. The plan reads like a fifteen-year-old company that still thinks about year twenty.