AI assistants are increasingly writing code for developers, but with convenience there is less noticeable risk: the usual skills can weaken without obvious signs. Indian developer Ashutos Rat proposed to check such degradation with the help of Atrophy, a new team line utility for regular training of programmers.
Atrophy evaluates the skills on the principle of Elo chess rating and shows which sides of the independent programming begin to subside. The tool is designed for developers who actively use AI agents and want to understand how confidently they cope with the code without prompting automatic assistants.
The utility checks five directions. The developer is asked to remember the syntax and write a small function according to a given description, find a hidden error in a code fragment, read the code and predict the result of the execution, fill out the pass in the call of a standard library or decomposing the task into architectural steps. So far, the Atrophy supports Python and JavaScript, and the exercises are available in three levels of complexity. Tasks are generated in new variants, so the training does not boil down to a repetition of the same set.
Before starting work, the user passes a basic exam of five tasks, one for each direction. According to the author, the initial check takes about 25 minutes. Then Rat advises to undergo short workouts for five to ten minutes two to three times a week. The Atrophy itself chooses a skill that has been left without practice longer than others, and sets a mild time limit. Delay does not fail the exercise, but reduces the growth of the rating.
Each direction receives a separate score and starts at 1200 points. There is no strict minimum or maximum, so the rating may fall below the entry level if the developer does not train a specific skill for a long time. Rat notes that the model is only inspired by Elo, but does not copy the chess rating completely. The lack of classes does not reduce points directly, but only reduces the system’s confidence in accuracy. At the same time, the Atrophy is still running manually and does not force the user to train on schedule.
Rat emphasizes that the Atrophy does not oppose AI. The developer created a utility to see the gap between working with AI and self-programming. For such a check, Atrophy offers a separate training session with an AI assistant once a month. The results of such tasks are stored separately and show whether the dependence on automatic support is growing.
The author warns that the rating should not be perceived as an accurate measurement of the professional level. Exercises serve only as an approximate model of real tasks, and the main value of the Atrophy is revealed in the dynamics: graphs help to notice which skills are gradually weakening due to constant support for AI.
Fears around this dependency are confirmed by research. Previously, MIT specialists studied students who wrote essays with chatbots, and recorded lower brain activity, the worst memorization of facts and problems with retelling their own texts compared to participants who worked without language models. The researchers linked frequent support for AI to superficial assimilation of material and a decrease in independence. A similar risk is indicated by the MIT Sloan material about how the AI constant help can gradually delay the user to dependence on automatic hints.
The Atrophy is trying to close just such a blind zone in programming. If the AI assistant completes functions every day, corrects errors and prompts the decision structure, the developer may notice the shape loss too late: at an interview, during an accident or a day without access to the network. The utility makes it possible to fix the starting point in advance and see what skills require training.
Atrophy evaluates the skills on the principle of Elo chess rating and shows which sides of the independent programming begin to subside. The tool is designed for developers who actively use AI agents and want to understand how confidently they cope with the code without prompting automatic assistants.
The utility checks five directions. The developer is asked to remember the syntax and write a small function according to a given description, find a hidden error in a code fragment, read the code and predict the result of the execution, fill out the pass in the call of a standard library or decomposing the task into architectural steps. So far, the Atrophy supports Python and JavaScript, and the exercises are available in three levels of complexity. Tasks are generated in new variants, so the training does not boil down to a repetition of the same set.
Before starting work, the user passes a basic exam of five tasks, one for each direction. According to the author, the initial check takes about 25 minutes. Then Rat advises to undergo short workouts for five to ten minutes two to three times a week. The Atrophy itself chooses a skill that has been left without practice longer than others, and sets a mild time limit. Delay does not fail the exercise, but reduces the growth of the rating.
Each direction receives a separate score and starts at 1200 points. There is no strict minimum or maximum, so the rating may fall below the entry level if the developer does not train a specific skill for a long time. Rat notes that the model is only inspired by Elo, but does not copy the chess rating completely. The lack of classes does not reduce points directly, but only reduces the system’s confidence in accuracy. At the same time, the Atrophy is still running manually and does not force the user to train on schedule.
Rat emphasizes that the Atrophy does not oppose AI. The developer created a utility to see the gap between working with AI and self-programming. For such a check, Atrophy offers a separate training session with an AI assistant once a month. The results of such tasks are stored separately and show whether the dependence on automatic support is growing.
The author warns that the rating should not be perceived as an accurate measurement of the professional level. Exercises serve only as an approximate model of real tasks, and the main value of the Atrophy is revealed in the dynamics: graphs help to notice which skills are gradually weakening due to constant support for AI.
Fears around this dependency are confirmed by research. Previously, MIT specialists studied students who wrote essays with chatbots, and recorded lower brain activity, the worst memorization of facts and problems with retelling their own texts compared to participants who worked without language models. The researchers linked frequent support for AI to superficial assimilation of material and a decrease in independence. A similar risk is indicated by the MIT Sloan material about how the AI constant help can gradually delay the user to dependence on automatic hints.
The Atrophy is trying to close just such a blind zone in programming. If the AI assistant completes functions every day, corrects errors and prompts the decision structure, the developer may notice the shape loss too late: at an interview, during an accident or a day without access to the network. The utility makes it possible to fix the starting point in advance and see what skills require training.
