Tables are the ground truth of the enterprise, yet for years AI’s progress on text and images dominated the headlines. Probabl’s State of Tabular AI Report 2026 argues that AI for tabular data deserves recognition as a category in its own right. The report maps the innovations that shaped it this year based on quantitative analysis of research and industry trends and interviews with researchers, open source developers, and labs.
This session begins with a 10-minute presentation by Gaël Varoquaux, who shares the report’s main findings. First, open source and open science are driving innovation in Tabular AI, from the scikit-learn ecosystem to tabular foundation models. Second, new tools like tabular foundation models and agentic frameworks are expanding the data scientist’s toolbox rather than replacing it. Third, as data scientists increasingly adopt agents in their workflows, the bottleneck is shifting from pipeline building to validation.
Then, Gaël is joined by Madelon Hulsebos for a 20-minute discussion about open questions in Tabular AI research and innovation: where tabular foundation models still fall short on real-world data, how to evaluate data science agents, and how to support the open science flywheel of Tabular AI.