A post hit the front page of Reddit this week. A developer with nearly a decade of experience claimed he’s completely stopped writing code.
Not reduced. Stopped.
I get it. I really do.
Claude Code deploys stuff to my homelab in minutes. It writes Kubernetes manifests, Terraform configs, and bash scripts faster than I ever could.
But here’s what that Reddit post leaves out.
The Part Nobody Wants to Talk About
My AI workflows only work because of two things:
- A well-designed architecture
- Plenty of hand-written documentation
The AI isn’t conjuring infrastructure from thin air. It’s reading my existing patterns, understanding my conventions, and extending what I’ve already built.
I’ve spent years documenting how my systems work. Where the config files live. Why I made certain decisions. What the dependencies are.
Without that foundation, the AI would be guessing. And guessing in infrastructure means downtime.
I’ve Had to Stop It From Breaking Everything
Let me be clear about something.
I’ve had to intervene and stop Claude from making disastrous mistakes. Multiple times.
It wanted to run destructive commands without understanding the blast radius. It suggested changes that would have broken authentication across my entire cluster. It confidently proposed solutions that violated basic security principles.
The AI doesn’t know what it doesn’t know.
I do. Because I’ve made those mistakes myself over the past five years. I’ve broken production. I’ve debugged at 3am. I’ve learned why certain patterns exist.
The Experience Prerequisite
“I don’t write code anymore” only applies if you’ve done it for years.
You need to know:
- What good architecture looks like
- When the AI is confidently wrong
- Which mistakes will cascade into disasters
- How to debug when the AI-generated code fails
A junior engineer using AI to write infrastructure code is like giving someone who’s never driven a car the keys to a Tesla with autopilot. Sure, the car can drive itself. But when something unexpected happens, they won’t know how to take over.
The Real Skill Shift
The Reddit post asks how young people will develop software engineering skills in this environment.
It’s a valid question.
My answer: the fundamentals matter more than ever, not less.
Understanding networking, security, systems design, debugging—these skills become more valuable when AI handles the typing. You need to evaluate what it produces. You need to catch its mistakes. You need to architect systems it can understand.
The skill isn’t writing code anymore.
The skill is knowing what good code looks like.
What This Means For You
If you’re experienced: embrace the tools. Document your systems obsessively. Let AI handle the implementation while you focus on architecture and review.
If you’re learning: don’t skip the fundamentals. Build things the hard way first. Break things and fix them. Develop the intuition that lets you catch AI mistakes.
The engineers who thrive will be the ones who can direct AI effectively—and that requires deep understanding of what you’re directing it to do.
AI is a force multiplier.
But you can’t multiply by zero.