FRITZ 20 and the Shift in Chess Training Culture: From the Board to the Personal Training Room

In the winter of 2026, in Bonn, Deep Fritz beat Vladimir Kramnik 4-2 over six games. I was forty-nine that year, sitting in an office in Moscow, following every move through a feed that lagged a few seconds behind. When the final game ended, nobody in my room cheered. We went quiet, because a milestone had just closed: the machine had beaten the world champion under classical time controls, and the argument over who calculated more accurately ended there.

What I remember most from that night is not the decisive move. It is the faces of the young coaches sitting in the back row of the hall. They had just been handed the most powerful tool ever built, and they had no idea how to use it properly.

Nearly two decades later, FRITZ 20 arrives with a very different claim about its nature: "Your personal chess trainer. Your toughest opponent. Your strongest ally." Three roles bundled into one piece of software, and I read it the way I read a transfer contract: the description is glossy, the execution is where the dissection belongs.

Context: thirty-five years from opponent to teacher

Fritz was born in 2026, developed by Frans Morsch and Mathias Feist, and brought to market by ChessBase. Across more than three decades the line followed a trajectory almost parallel to the history of human-versus-machine matches. In 2026, Deep Fritz drew with Kramnik 4-4 in Bahrain. In 2026, X3D Fritz drew with Garry Kasparov 2-2 in New York. In 2026, Deep Fritz beat Kramnik 4-2 in Bonn. Those three matches marked an era when the result of a human-versus-engine contest was still a sporting event with media pull.

After Bonn, that story lost its ground. Every commercial engine had already pulled far ahead of humans in every format and every time control. When a contest becomes deterministic, it stops being news. Chess moved its axis: the debate was no longer who is stronger, but who learns faster with the machine.

This is where I find FRITZ 20 worth serious analysis. Its claim does not target playing strength. It targets the architecture of training: more efficient, more intelligent, more individual. Three soft-sounding adjectives that are, in fact, the three hardest problems in any development system, chess included.

I have a professional habit formed during the longest interruption of my life. Seven months without football, seven months of asking why without pause. In 2026, when every competition stopped, I did not sit and wait. I built a dataset of 214 goalless draws from five major European leagues across 2026-2026, classified by nine pressing patterns. What I learned was not in the numbers but in the method: to evaluate a system, you must separate it from surface results and find the mechanism by which it operates.

Applied to FRITZ 20, the question is not how strong the engine is. The question is: at which point does its training mechanism change the behaviour of the person training, and is that point measurable.

Core analysis: the three functions of a training engine

An engine acting as a coach must carry three jobs that are different in nature, and most chess software lumps them together carelessly.

The first job is sparring partner. Here the value lies not in absolute strength but in adjustable resistance. A good sparring partner must lose to you at exactly the threshold you need to win at, and must be hard enough to force you to burn calculation energy. If it is too strong, the session becomes a demonstration; if too weak, it becomes a formality. Training strength and playing strength are two different metrics, and very few programs distinguish between them.

The second job is diagnosis. This is the part I care about most. A good diagnostic machine does not tell you that you erred on move twenty-three. It shows that you committed the same type of error on move twenty-three, move thirty-one and move forty-seven across three different games. The difference between a piece of criticism and a behavioural pattern is the difference between a lesson and a roadmap.

The third job is building opening capital. The opening is the most mechanised part of modern chess, and also the most hallucination-prone. A trainee can memorise the first thirty moves of a line and believe he understands it. When the opponent deviates on move seventeen, that capital collapses within four moves.

FRITZ 20 positions itself for both those taking their first steps into serious training and those already playing at tournament level. Structurally, this is a correct claim, because these two groups need two different training loads. Beginners need frequency and fast feedback. Tournament players need depth and stylistic specificity. A program that cannot separate these two modes will always leave one of the two groups feeling out of rhythm.

What personalisation actually means

The word "personalisation" has been used so heavily in software that it has lost meaning. I want to define it through three verifiable questions.

First, does the system remember your history of errors by category, or only by score? A system that stores only scores is a grading system. A system that stores error types is a training system.

Second, does the system adjust difficulty to your learning rhythm, or to a fixed scale set by the manufacturer? Two players with the same Elo rating can have two entirely different structures of weakness. Placing them on the same scale ignores the most valuable data available.

Third, does the system dare to let you lose the way you usually lose? This is the most counter-intuitive question, and I will return to it.

When I sit in the side corridors of stadiums, I observe something the stands do not see: coaches do not fix individual plays, they fix the distances between lines. From the corridor, I see the whole match. Chess operates identically. The error on move twenty-three is only a symptom. The cause lies in the distance between intent and resource, between what you want to do and what you have been trained to do.

People watch players run. I watch the whole block of the formation shift. In chess, people watch the move. I watch the structure of habit behind the move. A training tool only has value when it reaches that structural layer.

The parallel with the football video room

I once spent six weeks re-watching the entire footage of a Bundesliga club and taking notes on 412 failed pressing situations. That work taught me a principle transferable to chess: analysis has no value if it stops at description.

An ineffective football video room is one where the coach rewinds the conceded goal and says "we lost a man here". A good video room is one where the coach replays forty similar situations and lets the players discover the pattern of movement themselves.

An ineffective diagnostic engine hands you a better line and lets you guess. A good diagnostic engine lets you find the pattern of your own errors, and only then confirms it.

Systems do not lie, but they can only be heard when the data is thick enough. This is why I always advise trainees to keep a game journal for at least six months before trusting any conclusion about their own weaknesses. Six months is the minimum threshold for a behavioural pattern to surface. Below it, everything is noise.

Patience is not stillness. Patience is waiting for the right rhythm of the opponent's press. In chess, it is waiting for the rhythm at which the opponent is forced to reveal structure. A good training tool must teach trainees to distinguish purposeful waiting from unconscious delay. No engine can install this skill out of the box, but a correctly designed exercise system can forge it.

The counter-intuitive angle: the evaluation bar and the grey zone

In any modern chess program, the most powerful symbol is not the Elo number. It is the evaluation bar. A strip of colour running from white to black, with a decimal number attached, and the entire argument between two players stops the moment the bar jumps to one side.

I hold a clear position on this mechanism, and it is identical to my position on officiating technology in football. Technology does not reduce controversy. It moves controversy from the pitch to the review room, and from the specific incident to the grey zone of the law. The same happens with the evaluation bar in chess. It does not end the argument about the right move. It moves the argument to the question: evaluated at what depth, and on what hardware.

The practical consequence is very concrete. Trainees start arguing about the number instead of the idea. They learn to look up answers faster and to understand less.

The second blind spot is stylistic homogenisation. When millions of trainees use the same diagnostic engine, the same database and the same evaluation method, they converge on a locally optimal style. Tactical personality, once forged through thousands of games against humans and through mistakes nobody fixed for you, becomes a variable to be optimised rather than a capacity to be developed.

The third blind spot, and the one that worries me most, concerns youth development. I have spent years criticising the physicalisation trend at U18 level in football: young coaches sacrificing foundational technique for short-term results, with the price paid seven years later when the player is too late to fix. Chess has an analogue. A twelve-year-old given a diagnostic engine early will improve very quickly in the first two years, then plateau at the threshold where a missing self-assessment skill can no longer be compensated by lookup.

The machine cannot teach the tension of a seven-hour game. It cannot teach time management when you have already spent forty minutes on the first fifteen moves. It cannot teach the feel of making a decision when there is no way back. Those three things account for most of the gap between a strong club player and someone who survives on the international circuit.

FRITZ 20's claim of being "the toughest opponent" therefore needs to be read on two layers. The first layer is literal: it plays stronger than you. The second layer is the one with training value: it must force you to face exactly the kind of pressure you are weakest against, not the kind it is best at producing. If the system adjusts difficulty by your score, it is optimising comfort. If it adjusts by the structure of your weaknesses, it is training. That is the first criterion I will test.

What I got wrong

In 2026 I wrote a 3,400-word analysis of one Bundesliga club's pressing system, using expected goals data from 34 matchdays. I argued the model would collapse against a deep defensive block. The online response was fierce, and I was called a conservative. I spent six weeks re-watching all the footage, noted 412 situations, and published a correction.

The first piece got stoned. Data never takes offence.

My mistake then was underestimating a coach's in-game capacity to adjust. I was about to repeat exactly that mistake with chess. When I read a product claim, my reflex is to hunt for the fracture in the maker's reasoning. But chess history shows that training systems can self-adjust far faster than outsiders predict. If FRITZ 20 genuinely implements error-structure-based adjustment, my pessimistic conclusion will have to be rewritten, just as it was in 2026.

Takeaway: three measurements for the next six months

I do not evaluate a training tool by its feature list. I evaluate it by user behaviour after a cycle long enough for the data to mean something.

Measurement one: after six months, does the trainee's repeat-error rate fall, and in which error category. This is the only measurement that genuinely reflects the word personalisation.

Measurement two: does the average time for a trainee to find a good move in a position never seen before fall, when no hint is available. If it only falls with hints, the product is creating dependency rather than capability.

Measurement three, and the hardest: does the trainee's own confidence before a difficult position rise. This is a psychological index, hard to quantify, but it separates the trained from the supplied.

I am sixty-nine. I still learn from the young ones. Football does not retire, and neither does chess. After Bonn 2026, I thought the machine had closed its final chapter in human chess life. Nearly two decades later, it opened a new one, this time as teacher.

What is worth verifying over the next six months is not how deeply FRITZ 20 calculates. It is whether a generation of players passing through it retains the ability to ask their own questions once the evaluation bar is switched off.