Avoid AI atrophy – new tool promises to reverse vibe coding skills decay
The Register
A dev built a CLI tool called Atrophy that drills your coding skills so AI agents don't quietly turn you into a helpless button-pusher. It scores you like chess Elo, separately for AI-assisted and solo work, so you can watch your unaided skills erode in real time.
Based on reporting by The Register — read the original for the full story.
Summary, retelling and take written by AI under human oversight; images are AI-generated illustrations. How we work · Report an error
Ashutosh Rath, a developer in Bengaluru, noticed something a lot of us have felt but few have measured: the more you let an AI agent write your code, the rustier your own hands get. So he built a fix, or at least a diagnostic. It's called Atrophy, a command-line tool that treats your coding ability the way chess treats a player's rating, and forces you to keep drilling so the number doesn't quietly slide.
The mechanics are straightforward. You take a baseline exam covering five skill areas — syntax recall, debugging, code reading, API memory, and decomposition — which takes about 25 minutes and gives you five independent starting scores, each pinned at 1200. From there, Rath recommends short 5-to-10-minute drills two or three times a week, with Atrophy automatically picking whichever skill has gone neglected the longest. There's no hard floor or ceiling on the score, so if your muscles really have withered, you can watch the number sink well below where you started.
What makes this more than a gimmick is the separate tracking loop for AI-assisted drills. Once a month, Rath suggests doing an exercise with AI help, scored on its own axis, specifically to measure the widening gap between what you can do with an assistant and what you can still pull off solo. That gap, not the raw score, is the actual point. Rath told The Register he built the tool for himself, to see how much rust was forming while he leaned on agents day to day.
This isn't just anecdote-chasing, either. MIT researchers last year found that people writing essays with chatbot assistance showed measurably less brain activity than those working unaided, plus worse fact retention and weaker recall of their own writing — what the researchers called shallow encoding. Swap essays for code and the mechanism looks awfully similar: outsource the thinking long enough, and the thinking part gets quieter.
Rath is careful to frame Atrophy as anti-atrophy, not anti-AI. The drills are a proxy, not a certification, and he's upfront that the numbers matter more as a trend line than as an absolute score. But the timing is pointed. As agentic coding tools get pitched as productivity multipliers, a tool built purely to check whether your own skills are still there when the wifi drops feels like a pretty reasonable answer to a discomfort a lot of engineers have been quietly sitting with.
My take — AI-written commentary, not fact-checked reporting
I like this way more than another AI benchmark nobody asked for, because it's aimed at the human instead of the model. We spent two years hyping agents that write code for us and approximately zero time asking what happens to the person holding the keyboard once the agent leaves. Rath's tool won't save anyone from a bad outage, but at least it gives you a number to be honest with yourself about — and honesty about skill decay is in short supply in an industry that keeps selling autonomy as pure upside.
Read more about this at: The Register
