Why this exists

I went all in.

I'd been writing software for a long time. Long enough that the basics had become muscle memory: reading stack traces, holding a few dozen regexes in my head, knowing the shape of common errors without having to look them up. When the models got good, I handed all of it over. Why wouldn't I? It was faster, and for a while it was great.

The first time I noticed something was off, I was looking at a stack trace. Five frames, nothing exotic. I read it once, didn't fully understand it, and pasted it into the chat tab. The model explained, suggested a fix, I moved on. Twenty minutes later it hit me that I hadn't really tried to read it. I'd skimmed it, felt the small friction of not knowing right away, and reached for the chat tab the way I'd reach for coffee, before I'd really decided to.

It kept happening. The model would name a function for me. It would write a one-liner I could have written in my sleep two years ago. One day I caught myself asking it for a grep flag I'd been using for over a decade. Each of these was small enough to shrug off on its own. Together they were harder to shrug off.

I'd known about use-it-or-lose-it forever — everybody knows it about French and piano. I'd just never thought to apply it to my own field, where I was supposedly an expert.

Once I started looking, I saw it everywhere. Engineers I respected, people who used to write thick, careful posts about systems they understood deeply, were admitting the same thing — never in the post itself, always in the replies. "I can't write without it anymore." "I've forgotten things I knew for ten years." "I feel slower when I'm not using it, which is the opposite of the promise." You could tell they weren't really joking when they laughed about it.

So I started running a small experiment on myself. Each morning I'd take one problem and work it without help. Read the docs instead of prompting. Read the stack trace before doing anything else. After a while I started calling these reclamations, because that's what they felt like — taking back a small piece of thinking I'd quietly handed over. None of it was anti-tool. I still used the models all day. I just stopped using them for the reps I needed to keep.

Later I added a weekly version: write one real thing — a doc, a post, a note to my team — with no help at all. The first few were rough, and slow in a way that embarrassed me. The third or fourth were less rough.

After a few months, some of it was coming back. Unevenly — the regexes returned long before the design instincts did, and my memory for what I read is still worse than I'd like. But stack traces felt natural again. I'd catch myself typing out a pattern before it occurred to me to ask. Things I used to be able to do, returning, at the speed of any other practice.

It made me think other people might want a version of this.

That's what quitting.ai is: the experiment, packaged the way I wish someone had handed it to me. Not a guilt trip — I use the models every day and will tomorrow. A diagnostic that names which capacities are quietly atrophying, and a 30-day plan to put the reps back in. The goal is to use the tool on purpose instead of by default.

There's a bigger hope underneath it, and it's mostly about company. A lot of people — engineers, writers, anyone whose work used to require an inner argument — are noticing what I noticed, and noticing it alone, in replies and side comments and the bad kind of laugh. I wanted to find the others.

That's who this is for.