Your Edge Isn’t Writing Code
The other day, Naval posted something about the future of software engineers. I almost scrolled past it — not because the topic didn’t interest me, but because I’d been stewing on the same question for the past couple of weeks already.
The question everyone’s dancing around: what’s the value of a software engineer now that anyone can ship an app from a text prompt?
You’ve seen it. Vibe coding is real. Apps are cheap to make. The job market is slower than it used to be. And the takes online are either “your job is safe” or “you’re toast” — with nothing useful in between.
Here’s where I’ve landed.
The people panicking are focused on the wrong thing. They’re worried about whether they’ll still be writing code. That’s not the moat. It never was.
Your edge isn’t writing code. It’s knowing what good code looks like.
AI is a pattern replicator. Feed it a well-structured codebase — clear types, consistent conventions, good module boundaries — and it extends those patterns. The better your foundation, the better AI performs on top of it.
Vibe coders don’t have this. They can’t tell the difference between a codebase that’ll scale and one that’ll quietly collapse under its own weight. As their apps grow, entropy compounds, and AI starts making it worse instead of better. You know the difference. That knowledge is now a force multiplier.
There’s a second edge that’s less obvious: engineers have tooling leverage that vibe coders don’t even know exists.
Take linter rules. Write one rule enforcing a particular pattern, and from that point on, every line AI generates gets checked against it. Automatic feedback. Zero overhead. The pattern holds whether you’re in the loop or not. That’s not a shortcut — it’s infrastructure. And most people shipping with AI right now have no idea it’s an option.
The apps built without this will show. Yours don’t have to.
Worth your time
- Naval on the future of software engineers — More uplifting than the doom threads. Naval cuts through the anxiety with a cleaner frame for where the value actually sits in a world where code is cheap. Worth reading if you’ve been sitting with this question.
From me lately
On this same line of thought I wrote From Better Prompts to Better Structure but also how to use Codex to help ship faster using git worktrees.