Marc-Alexandre Côté

Marc-Alexandre Côté is a Principal Research Manager at Microsoft Research Montréal who leads the TextWorld project, a sandbox where reinforcement-learning agents learn language by playing text-based games. In 2017 he finished a PhD at the University of Sherbrooke under Hugo Larochelle and Maxime Descoteaux, coauthored a deep-learning tractography paper at MICCAI, and joined Microsoft. He was first author of the 2018 TextWorld paper and later coauthored Jericho for interactive fiction in 2019 and ALFWorld in 2020. More recently, he coauthored debug-gym for LLM-based coding agents in 2025 and the FrogNano coding agent in 2026. He is also an associate industry member at Mila.

Now at: Microsoft Research and Mila

Timeline

2017

A Sherbrooke PhD and tractography

He completes a PhD at the University of Sherbrooke in 2017, with Hugo Larochelle and Maxime Descoteaux as doctoral advisors. At MICCAI 2017, he coauthors "Learn to track: Deep learning for tractography" with Philippe Poulin, Larochelle, Descoteaux and others.
2017

Joining Microsoft

He joins Microsoft in 2017. His research at Microsoft Research Montréal focuses on reinforcement learning and natural language understanding.
2018–2019

The TextWorld sandbox

In 2018, he is first author of TextWorld, a Python sandbox for training and evaluating reinforcement-learning agents on text-based games, presented at the Computer Games Workshop at IJCAI 2018 in Stockholm. In 2019, he gives a seminar on TextWorld for the Game AI Group at Queen Mary University of London, while leading the project.
2018–2020

Organizing the Wordplay workshops

He sits on the organizing committee for the Wordplay: When Language Meets Games workshops at NeurIPS in 2018 and 2020.
2019–2020

Jericho for interactive fiction

In 2019, he coauthors "Interactive Fiction Games: A Colossal Adventure" with Matthew Hausknecht, Prithviraj Ammanabrolu and Xingdi Yuan; the paper introduces Jericho, a learning environment for man-made interactive fiction games. In a Microsoft Research blog post of January 2020, the authors explain that giving agents action templates and vocabulary cuts the action space from 240 billion possible actions to 98 million.
2020–2021

The ALFWorld simulator

In 2020, he coauthors ALFWorld, a simulator in which agents learn abstract text-based policies in TextWorld and then carry out goals from the ALFRED benchmark in a visual environment. The paper appears at ICLR 2021.
2025–2026

debug-gym and FrogNano

In March 2025, he coauthors debug-gym, a textual environment with tools such as the Python debugger pdb for developing LLM-based agents in interactive coding settings. In September 2026, he is a coauthor of FrogNano, a 4B coding agent post-trained with reinforcement learning on around 1,500 software-engineering environments with synthetic tasks.
As of October 2026

Research manager at Microsoft

He is a Principal Research Manager at Microsoft Research Montréal, where his research since 2017 focuses on reinforcement learning and natural language understanding. Mila lists him as an associate industry member, with deep learning, LLM agent reasoning and reinforcement learning among his research topics.