How I Forgot How to Code

How It Started

In August 2025, I started using LLMs to code. A bit of ChatGPT here and there, then Cursor more aggressively, and then Claude Code. It was an incredible feeling. I could build an app from scratch over the weekend and have it live on the internet by Sunday. The speed was unlike anything I had experienced before. I built DanbingAI, DecodeThisText, and a few other apps. Even at my full-time job I was shipping fast. At a certain point I was mostly testing to see if things worked, glancing at the code to make sure there were no security issues and that the AI had not accidentally committed environment variables to GitHub. I knew something felt different but I did not think much of it at the time.

The Setup

The company I was working with was not particularly stable. I had previously worked with the co-founder the year prior and that company had shut down suddenly with no warning. He started a new company, got funding, and did a mass hiring early on which felt premature for a company with no solid roadmap or product yet. I brought it up but moved on. Either way, I kept working there while knowing in the back of my mind that things could go sideways again.

December 2025 came around, the same time of year things had gone wrong before, and the co-founder was acting strange. A couple of us who had been at the previous company started updating our resumes just in case. Then after the holiday break, instead of returning to work, we got a Slack message in #general saying we had all worked hard and deserved another two weeks off. Shortly after, the co-founder messaged us privately that the company was closing and we were being let go effective immediately.

Rust, Not Amnesia

Here is where I want to be clear about something. When I went to interview at another company and hit a live coding question, I struggled. I could reason through the problem conceptually but translating that into actual syntax felt slow and awkward. It was frustrating, but it was not some permanent condition. It was rust.

This happens to developers all the time. Take a year off, switch to a management role, or spend six months doing nothing but prompting AI tools, and your typing-code-from-memory muscle will weaken. That is just how skills work. The same way someone who took a year off the gym will feel weak their first week back but gets back to their previous level much faster than a beginner would. Coding muscle memory comes back quickly once you start writing code again. It is not gone, it is just dormant.

What made my situation harder was the timing. I was job hunting immediately after getting laid off, without a runway of time to ease back into writing code manually before being put on the spot in interviews.

10xing with LLMs

For about six months I had been prompting Claude Code and Cursor to build feature after feature, reviewing the output, and shipping. I got genuinely good at it. I could run multiple agents in parallel, use markdown files to manage tasks and context the way I used to manage developer work as a product manager and scrum master, and move faster than I ever had as a solo developer. I experimented with spec-driven development, started using OpenClaw before it started trending, and even threw an OpenClaw party at my place where developers in my city came to set things up and build together.

The tradeoff is that during those six months I was not the one writing the code. I was directing it. And when interview time came, the interviewers were not asking me to direct anything.

The Tradeoff

A lot of founders and engineering managers right now will say things like "I will fire any developer who is not using AI to be more productive." Companies are hiring fewer engineers because they believe AI can absorb some of the workload, and they are probably right to some extent. The pressure to use AI tools is there.

But if you let AI write the majority of your code for an extended period, your hands-on coding fluency will fade. Not permanently, but enough to matter when you are suddenly in an interview writing code in a shared editor with no AI assist. Companies push you to use AI, then test you in environments where you cannot.

Using AI less means potentially moving slower than your peers. Using it too heavily means your raw coding skills quietly atrophy in the background. The key is balance and being intentional about which parts of the work you keep doing yourself.

The Practical Takeaway

Do not stop writing code yourself. Use AI for repetitive tasks, boilerplate, documentation, and as a faster alternative to Googling. But keep solving problems on your own regularly. Work on side projects where you are the one writing the code. Keep enough savings to give yourself a buffer if a job transition happens, because the market for developers is competitive right now and interviews still heavily test fundamentals.

The skills come back fast once you start using them again. That part is not complicated. The harder part is not letting yourself drift too far from them in the first place.

Where I Am Now

I am still applying for roles while continuing to build on my own projects. DecodeThisText has some paying users, which is a signal worth paying attention to. Whether I double down on it or land a new role first, I have started writing code manually again every day and the rust is coming off faster than expected. That part at least is reassuring.