Anant Dole is an AI product builder and competitive chess player who became Head of AI at Take Take Take, the chess company co-founded by Magnus Carlsen. His work combines specialized chess engines with language models to explain not only which moves succeed, but why human players should understand them.
Dole represented South Africa internationally in chess for a decade; archived tournament games document his competitive background. He subsequently spent seven years at Uber across operations, product, and engineering, then joined Writer as a founding AI architect. There, he helped develop enterprise applications including medical-affairs workflows and tool-using agents for financial analysis. After two years at Writer, he joined Take Take Take.
Building a chess coach people can trust
With Asbjørn Steinskog, Dole developed an AI chess coach organized around several concrete engineering decisions:
- Stockfish-grounded chess coaching: Stockfish evaluates each position, dedicated detectors identify tactical and positional themes, and human-move prediction estimates which ideas players can realistically find. A language model converts those signals into explanations without being trusted to calculate the chess itself.
- Human-centered evaluation: Scenario-based tests examine blunders, tactical patterns, and hallucinations, while experienced players determine whether explanations are both accurate and genuinely instructive.
- Human-supervised agent workflows: Negative feedback can trigger a Slack-connected coding agent to investigate commentary, revise prompts or detectors, and regenerate results, with people reviewing proposed changes.
- Product-aware latency: Postgame analysis must arrive while players remain engaged. Dole weighs model accuracy against response speed and predictability, reserving slower reasoning for interactions where waiting is acceptable.
His experience playing Leela Chess Zero sharpened his interest in translating initially opaque computer moves into strategic understanding. He also advises We Are Agentic on practical AI-agent workflows.
