For decades, chess players have had access to something most athletes could only dream of: a machine that can tell them, with brutal precision, exactly where they went wrong.
The problem is that chess engines have traditionally been much better at being right than being helpful.
Stockfish can look at a position, condemn your perfectly reasonable-looking move and offer a line that feels as though it was calculated by an alien. Technically, that is useful. For a club player staring at a minus-3 evaluation and wondering what on earth just happened, it can also be deeply unsatisfying.
That gap — between analysis and teaching — is where a growing crop of AI chess coaches is trying to make its move.
The engine was never really the coach
Chess engines are extraordinarily strong. Stockfish 19, released this month, improved again over its predecessor. But raw playing strength is not necessarily what a learner needs.
Imagine a 1000-rated player hangs a bishop because they were so focused on attacking the king that they stopped checking what their opponent threatened. An engine can identify the blunder instantly. A good coach wants to know something else: Why does this player keep making this kind of mistake?
That is the more interesting promise behind today’s AI coaching tools. Instead of simply asking computers to play better chess, developers are increasingly asking them to understand the way humans play chess.
AI is learning to play more like us
One of the most interesting projects in this area is Maia, developed through University of Toronto research. Unlike a conventional engine whose job is to find the strongest move, Maia is designed to predict what a human player at a particular skill level is likely to play.
That sounds like a subtle distinction. It isn’t.
If Stockfish tells you the best move was something almost nobody at your rating would find, Maia can provide a different perspective: what do players like you normally do here? Which mistakes are typical at your level? And what choices are realistically within reach?
Newer apps are building that idea into consumer products. Chessdrive, for example, combines Stockfish analysis with Maia3, allowing players to compare the engine’s objective view with a model of human decision-making. It can import games from Chess.com or Lichess, identify critical moments and offer explanations in conversational language.
Chessy takes another route: it connects to a player’s Chess.com or Lichess history and tries to turn their own mistakes into a training programme, including personalised puzzles and weekly coaching reports.
Meanwhile, Chess.com itself has pushed coaching further into the mainstream. Its Play Coach mode gives feedback while a player faces a virtual opponent, adjusting the experience to the player’s chosen strength.
The clever part isn’t finding the best move
This is where AI chess coaching starts to become genuinely interesting.
The computer already won the argument about who can calculate better. Humans lost that contest years ago. The next challenge is whether software can become good at the very human business of teaching.
A useful chess coach does more than point at mistakes. They notice habits. They know when a student needs tactics rather than another opening video. They recognise when someone understands a concept intellectually but repeatedly fails to apply it with three minutes left on the clock.
AI systems now have an obvious advantage here: scale. Give an app hundreds of your games and it can search for patterns no human coach would reasonably spend hours cataloguing by hand. If you repeatedly mishandle rook endings, lose after pushing too many pawns around your king or collapse whenever opponents play a particular opening, software can potentially spot that pattern and build training around it.
That’s far more compelling than another button labelled “analyse with engine.”
But AI can still sound confident when it shouldn’t
There is a catch. Turning an engine evaluation into natural language does not automatically turn software into a great teacher.
Large language models are very good at producing explanations that sound coherent. Chess is unforgiving of explanations that merely sound right. A coach that confidently invents a strategic justification for a move is arguably worse than an engine that simply gives you a number and a variation.
There is also a difference between diagnosing a weakness and understanding a person. A human coach can see frustration, change the lesson when a student is overwhelmed, challenge lazy thinking and know when the real problem is confidence rather than calculation. That relationship is difficult to reproduce with a dashboard.
So, can AI replace a chess coach?
For many casual and improving players, the better question may be whether they need it to.
A player who has never paid for coaching suddenly having an inexpensive assistant that reviews every game, remembers recurring mistakes and creates targeted exercises is a meaningful improvement. AI does not have to replace grandmasters giving private lessons to change how millions of people learn chess.
It may instead occupy the enormous space between watching a YouTube lesson and hiring a human coach.
And that may be the real breakthrough. The most useful chess AI of the next few years probably won’t be the machine that finds an even stronger move. We already have machines that are frighteningly good at that.
It will be the one that can look at your bad move and explain, in a way you actually remember, why you’re likely to make it again.
Why this matters beyond chess
Chess is also a useful laboratory for a much bigger AI question. If artificial intelligence is going to become a personal tutor for languages, mathematics, coding, music or professional skills, being smarter than the student isn’t enough. It has to understand how people learn.
Chess gives developers an unusually measurable place to test that idea: every decision is recorded, mistakes can be analysed and progress has a rating attached to it.
Perhaps that is why the current wave of AI chess coaches matters more than it first appears. The board hasn’t changed. What is changing is the relationship between the player and the machine sitting beside it.
