Asbjørn Steinskog is an AI developer and competitive chess player who builds chess-coaching systems that transform engine calculations into practical advice for human players. His work combines computational linguistics, online chess communities, and consumer AI engineering.
He studied at the Norwegian University of Science and Technology and coauthored 2017 research on modeling topics in short social-media posts, examining how grouping tweets by author or hashtag improves the context available for language analysis. He later created the Offerspill Chess League, organized around regular Lichess tournaments, served as a Lichess moderator, and joined the board of Offerspill Sjakklubb.
At Take Take Take, Steinskog and Anant Dole developed an AI-powered chess coach that separates reliable chess analysis from natural-language explanation. Their system uses Stockfish to evaluate games, identifies tactical and positional patterns, and supplies that structured context to a language model that generates accessible commentary.
- Stockfish-grounded chess explanations: Specialized detectors surface threats, candidate moves, forks, pins, pawn weaknesses, and strategic plans, keeping commentary anchored to verifiable features of the position.
- Human difficulty as coaching context: Maia estimates which moves players at different rating levels are likely to find. Combined with Stockfish evaluations, those probabilities distinguish the objectively strongest continuation from an unusually difficult or instructive decision.
- Human-supervised feedback loops: Reports of poor commentary enter Slack and a Claude Code workflow that investigates positions, adjusts prompts or pattern detectors, and regenerates explanations. Steinskog reviews proposed changes before they become product updates.
- Consumer-ready chess evaluation: Real-game scenarios test tactical recognition and hallucinations while experienced players assess whether explanations make sense. The resulting system must balance accuracy with the fast responses players expect immediately after a game.
His chess-coaching work with Dole applies a clear engineering principle: domain-specific systems establish what is true, while language models translate those findings into guidance people can use.
