Max Mergenthaler Canseco and the Trouble With Tomorrow

In 2016, Max Mergenthaler Canseco wrote about a machine winning a board game. AlphaGo had defeated Lee Sedol, and Canseco was interested in what happened after the applause. A capability people had associated with human intuition had become a piece of software. What would that do to the culture that had built it?

His essay approached the match through a longer history of people and their inventions. Machines had already become stronger, faster and better at certain calculations. This latest encounter invited a different kind of reckoning. Canseco’s attention settled on the change in human life that followed a change in technical possibility.

A decade later, he runs a company that asks machines to anticipate the next number. As co-founder and CEO of Nixtla, he works on forecasting tools for data recorded over time. The destination seems far from a cultural essay about Go. The question underneath has stayed close: once a machine can do something useful, how should people work with it?

An essay before an algorithm

There is a small literary shelf in Canseco’s public record. In 2015, he wrote about Richard Rorty and the purpose of philosophy. In March 2016, he wrote about William Blake, opening with a scene from Jim Jarmusch’s Dead Man. Poetry, cinema and questions about machines sit comfortably beside one another. It is a more varied reading list than the usual startup biography permits.

That Blake opening is especially pleasing. An essay about a poet begins with a film character who shares his name. Canseco lets one cultural form lead into another. Readers expecting a tidy résumé encounter someone willing to take the scenic route.

The Rorty essay had examined divisions within philosophy and the question of what the discipline was for. His AlphaGo essay later moved toward what a new technical capacity meant for ordinary human culture. These pieces supply an early view of his interests without requiring anyone to turn a poem into a business plan. Blake can remain Blake. There are limits to what even a founder should monetize.

A philosopher learns another language

Canseco studied philosophy at Freie Universität Berlin. His academic interests included logic, the philosophy of science and the philosophy of mind. The formal side drew him toward programming; he taught himself Python and R online.

The passage from philosophy to code has a certain practical neatness. Both ask for definitions. Both punish a muddled assumption, though a computer tends to deliver the punishment sooner. An argument can survive an afternoon seminar on charm. A program usually wants its variables settled.

His educational record also includes corporate finance study at ITAM and an entrepreneurship specialization through Wharton. Those additions make the route easier to understand. There are questions about knowledge, questions about writing a working program, and questions about paying for the organization that maintains it. A founder eventually gets examined in all three.

Canseco’s subsequent work included research at the Max Planck Institute for Human Development and leadership of research and data teams in the Mexican government. He later moved into startups. The sequence puts institutions as well as algorithms in his background: places where an answer needs to travel through a team before anyone can act on it.

“Time series are in one aspect, the DNA of the world.”

Max Mergenthaler Canseco

How people learn to choose

In 2018, Canseco, Dirk U. Wulff and Ralph Hertwig published a meta-analytic review in Psychological Bulletin on the description-experience gap. Its subject was how people learn about the possible consequences of a choice.

The distinction is easy to recognize. One person reads a description of the odds. Another learns by encountering outcomes. Those routes can produce different choices. Having information on paper and acquiring information through experience are separate events in a person’s life, even when they concern the same decision.

For a forecasting founder, that earlier research makes an interesting companion. A prediction will eventually land with someone who brings previous experience to the screen. There may be a familiar supplier, an unreliable delivery day or a demand pattern the team has learned to distrust. A number enters an existing working life.

The connection is a useful way to read his career, rather than a claim that one paper caused a company. Decision research examines the person choosing. Forecasting software tries to improve the information available before the choice. Between them sits the troublesome business of acting while the outcome remains unknown.

Tomorrow, with an installation guide

Nixtla was preceded by other ventures. Canseco co-founded Orax, a supply-chain software company, and served as CTO and co-founder of Levo. Y Combinator lists Levo in its Summer 2021 batch: a treasury-management platform for startups and smaller businesses in Latin America.

Supply chains and cash management provide practical settings for a recurring problem. A decision gets made today, while its consequences depend on what happens later. Stock occupies space. Money has a job to do. Forecasting belongs close to both conversations.

At Nixtla, which he co-founded in 2021, the work took a public form through open-source software. By November 2022, he was discussing its forecasting libraries with Federico Garza on the Data Exchange podcast. StatsForecast, MLForecast and NeuralForecast offered developers different families of methods. They made the technical choices accessible through software people could install and inspect.

HierarchicalForecast addressed another unglamorous requirement: forecasts at different levels should agree. A retailer might predict sales by product, shop and region. The separate answers need to add up. Otherwise the forecast produces an argument about arithmetic before it produces a decision about stock.

His public product announcements can be measured, too. When introducing an experimental neural approach in 2024, he asked: Should you start using KANs now for all your forecasting needs? Probably not. He encouraged exploration while stopping short of prescribing the new method for every job. That qualification belongs beside the release announcement.

Canseco is a co-author of the research behind that framework. Its focus on coherent forecasts gives the work a pleasingly concrete quality. Grand conversations about artificial intelligence eventually reach a spreadsheet where two totals disagree. Somebody still has to fix the totals.

A model that brings its own education

In October 2023, the original TimeGPT-1 preprint appeared under Azul Garza and Max Mergenthaler-Canseco’s names. The proposal was to train a model in advance and have it make predictions for new time series. The paper tested that approach against statistical, machine-learning and deep-learning methods.

The practical attraction was a shorter journey between historical data and a forecast. Instead of beginning each application by building and training a model, a developer could call on a model that had already learned from other data. Zero-shot forecasting is the technical term. A head start is the ordinary one.

TimeGPT also gave Nixtla a commercial product alongside its public libraries. Its software client is open source; the forecasting model itself is proprietary. Developers access it through an API, and the company offers deployment on customers’ own infrastructure. The distinction matters to anyone deciding what will run where.

For Canseco, this puts two jobs beside each other. He helps build tools that a public community can use and improve. He also leads a company selling a service. The installation instructions cannot answer every business question, however elegant the installation instructions may be.

The customer has a calendar too

In February 2026, Energize Capital announced that it was leading a $16 million Series A in Nixtla. GreatPoint Ventures and True Ventures participated. Energize partner Juan Muldoon would join the board, with Agustina Soriano Sergi joining as an observer.

The investor’s account described a San Francisco headquarters and a distributed team of 17 engineers and researchers at the time. The next stage involved expanding the team and developing the enterprise platform. That is an organizational assignment as much as a research assignment: customers need a product they can keep using after the pilot ends.

By June, Canseco was discussing adoption at an industrial AI gathering in Chicago, alongside leaders from Nexxa.ai and Siemens. His remarks focused on technical buyers who understand both a business problem and the technology being offered to address it. He emphasized showing customers practical improvements in cost, time and resources.

Here, the early interest in human decisions returns to view. An organization already has a process, even if the process is awkward. Changing it means persuading people who will live with the result. The forecast has to earn its place in their day.

The company’s public customer list includes Nestlé, Unilever, Zalando and Decathlon. These are settings where predicting demand leads to concrete choices. The CEO’s conversation has to reach beyond model behavior to the person responsible for the next order, the next deployment or the next budget.

The poetry of making things agree

A September 30, 2026 preprint, The Nixtlaverse: An Open-Source Ecosystem for Forecasting, includes Canseco among its authors. It examines how different forecasting model families can share data formats and outputs while keeping their specialized implementations.

The paper tests the approach on public M5 competition data, including reconciliation across a hierarchy of 42,840 series. The premise is practical: a developer should be able to work across methods without rebuilding every step around a different package.

It is an apt recent chapter for Canseco. His career has moved among disciplines; this work helps different technical approaches meet through common interfaces. Agreement is doing useful work at several levels, from a dataset’s shape to the totals in a forecast.

His ambitions also reach smaller organizations. He has described a future in which a coffee shop or hot-dog stand could use forecasting to plan sales and reduce waste, alongside much larger businesses. It is a useful image to keep beside the enterprise contracts. A forecast becomes valuable at the scale of the decision someone actually needs to make.

The cultural essayist and the software executive can coexist without a dramatic reconciliation scene. One asks what changes when machines gain new abilities. The other has to help people use those abilities on a Tuesday. Tomorrow will still arrive with inconvenient timing. Canseco’s work concerns the tools people have ready when it does.