How AI Agents Are Changing Data Science Work

Summary

Data scientist Robin Linacre explains how large language models and AI agents are moving data science from hands-on coding toward task design, feedback loops, and verification. Examples include a 25% Splink speedup, documentation upgrades, code porting, game balancing through self-play, and semi-autonomous machine-learning research. He argues that agents work best when they generate new evidence and verify their progress, while human judgment remains necessary for quality they cannot assess reliably. Linacre expects longer-running cloud agents, stronger memory, and teams of agents to become common, increasing the importance of architecture, creativity, judgment, and customer feedback.