Everywhere you look, there seems to be an underlying trend bubbling up “If you are not using AI you’re falling behind.”

It becomes even more difficult to avoid when there’s a million AI coding tools coming out daily: Claude Code, Codex, Cursor..etc. Don’t forget about AI agents. If you’re not spending your day setting up AI to do random tasks that you probably won’t repeat often, you’re falling behind. Even prominent engineers, like Linus Torvald, are seemingly joining the vibe coding movement.

However, I’m here to propose a seemingly counter intuitive point. The more you avoid using AI coding assistants the better you are at them.

I myself fell into the headspace of falling behind, so I decided to turn to the dark side and give a few coding assistants a try. Shockingly, not really, they were really good…with a strong caveat…they were really good when I knew the code base.

To truly utilize AI coding assistants for long term meaningful projects, I needed to understand the code base, the language, and the tech stack. I would then act as an approver for the AI. Whenever I started to lose track of the project, the AI assistant would magically get worse and worse and the vision I had would deteriorate.

The difficulty does not come from understanding these tools. Frankly for most developers, I estimate, it would only take a week to get up to speed with any of these AI coding assistant, some even less. These tools are simple and intuitive…as you can see by the amount of software slop being produced. The true difficulty lies with understanding and building a mental model of your code base. This is where the more you rely on AI the worse you get at this skill.

I predict that the true future of coding and being a software engineer are mental code models (MCMs): your understanding of your code base. The only way to get better at that skill is to turn off the AI assistants and practice.