From Algorithms to Agents: Lessons from Building Copilot

​What Generative AI Has Changed and Which Fundamentals Still Matter - Vadim Smolyakov

​Vadim’s path spans from algorithmic ML and MIT research to building Copilot at Microsoft, providing him with a front-row view of how practice has shifted from hand-built algorithms to agentic systems.

​In this conversation, he reflects on what genuinely changed with generative AI and what fundamentals still matter.

​We’ll discuss assistants as complements to human thinking, why embodiment (“physical AI”) may be the next step, and how goal-setting and sharing knowledge shape work that lasts.

​We plan to cover:

• ​What Copilot taught about scope, trade-offs, and evaluation

• ​Agents as assistants: avoiding over-reliance while gaining real leverage

• ​Fine-tuning, synthetic data, and how the craft of ML has evolved

• ​Physical AI: what to build in vs. what to learn from the environment

• ​Purpose, habits, and leaving durable digital/physical artifacts

​

​About the speaker

​Vadim Smolyakov is a Machine Learning Engineer at Microsoft, working on Copilot AI, an author of Machine Learning Algorithms in Depth, and a former MIT PhD student. He focuses on generative AI (agents, RL), self-development, knowledge sharing, and the societal impact of intelligent systems.

​

​​​​​​​DataTalks.Club is the place to talk about data. Join our slack community!