Over the course of the last year, it seems like everyone in the software industry has flipped from AI curious to AI by default. “Manual” or “Handwritten” code is already legacy and prompt-driven development is the way of the future. Perfecting prompts, skills, workflows, and AGENTS.md files is a daily process. Automate everything and ship faster than ever before. Keep up or you’ll get left behind! This is the reality, or rather this is what appears to be reality if you spend too much time on X.

I started playing around with LLMs for real in the summer of 2025. Previous attempts were disappointing but then I saw a shift and these tools suddenly became really powerful. I signed up for a Claude Pro account so that I could get access to Claude Code and after a week I was hooked. It clicked and like thousands of other developers around the world, I experimented and tested the limits of what they could do. It was a mixture of highs and lows. One-shot dopamine hits and frustrating misfires. I got better and better at prompting but the results were still a mixed bag. Still, I was convinced these tools were the wave of the future and I had better get comfortable using them.

At Paytient I was part of our AI Guild. I was one of maybe 10 developers who were particularly bullish on this new tech and we shared our tips, tricks, and insights with each other. It was exciting and new. At my new job at MaintainX, AI is almost a default. It is expected that you use these tools and they have basically removed any token limits with a very high spend account per developer. (No one has even come close to that spend limit).

The upside of this is that we are encouraged to experiment with AI workflows, invest in tooling and basically see whatever works. Last month my team spent an afternoon sharing skills, best practices and just talking shop about how we’re using AI. I gave an informal presentation on how I’m using git worktrees to make it easier to run agents in parallel.

It’s all very exciting but I already worry about “AI brain rot”. If I only outsource my development to machines, I’m not just concerned I’ll lose my ability to code by hand — I’m already convinced it’s happening and will continue to.

While preparing for interviews in November and December of 2025, I wrote tons of code by hand. I even turned off LSP support in vim so that I wasn’t getting autocomplete suggestions. I wrote code in online browsers to get comfortable working in environments I wasn’t used to. The idea was to force myself to think, remember and be conscious of every character I typed into the keyboard.

In Live coding horror I described freezing during live coding interviews because of the strange phenomenon of forgetting basic programming skills when faced with an audience and an unfamiliar coding environment. I overcame this by removing assistance, intentionally getting uncomfortable and doing everything “by hand”.

“Agentic coding,” as they are calling it, is the opposite of all this. You issue commands and outsource the actual work. Your role changes from implementer to reviewer. There’s nothing wrong with this (and it’s an incredibly powerful ability) but reading code and writing code are not the same. It’s no different from reading and writing prose. If you only read, your ability to write decays.

I’m riding the AI wave like everyone else in the industry and I’m amazed by the things I can do now that were simply impossible only a couple years ago. But I remain sceptical that I can or should stop writing code by hand. AI brain rot is real and I want this new technology to enhance my skills, not replace them.

I don’t know what this looks like yet. Perhaps it means taking an AI-free week every once in a while. Maybe it means having a side project that you only write by hand. Whatever it looks like, it has to be an intentional decision to keep your mind sharp.

Ross Noble is a software developer, ultrarunner, podcaster and former van-dweller with a passion for the outdoors. He writes about running, cinema and anything else that interest him.

Montreal, QC