Tags: ai, career, productivity, discuss
Last month I met a developer at a Shenzhen tech meetup who told me he was "waiting to see where the AI thing goes" before deciding whether to learn it. He's been waiting for two years. In that same two years, I've watched someone with zero coding experience ship a working SaaS product using Cursor and Claude.
The gap between these two realities is what keeps me up at night.
Let me be clear about something: I don't believe AI is going to replace software developers. Not in the way the Twitter panic threads suggest. The idea that a CEO is going to type "build me Facebook but for dogs" into ChatGPT and fire their entire engineering team — that's fantasy. The technology isn't there. The real threat is subtler and, honestly, more interesting.
AI won't replace you. But someone who knows how to wield AI effectively absolutely will.
Here's what I mean. I spent six years in corporate tech before going solo. I've worked with brilliant engineers who could debug memory leaks in their sleep but refused to touch any AI tool because "it writes bad code." They're not wrong — AI does write bad code sometimes. But they're missing the point the same way developers in 2005 missed the point about Stack Overflow. "You shouldn't copy-paste code you don't understand!" True. Also, irrelevant. The developers who learned when to trust Stack Overflow and when to verify shipped faster than the purists. The same dynamic is playing out now, just accelerated tenfold.
I use AI for the boring parts. Writing test cases. Generating boilerplate. Converting data formats. Drafting documentation that I'll then edit heavily. These are tasks that would take me hours and drain my mental energy for the real work: architecture decisions, user experience, figuring out what to build in the first place.
The junior developers who embrace this approach are terrifyingly productive now. I've talked to bootcamp graduates who are shipping full-stack applications in weeks. Not toy projects — real, deployed, revenue-generating products. They don't understand every line of code, but they understand enough to make it work, and they learn the rest as they go. That used to be impossible. Now it's the default if you're curious and persistent.
So where does that leave the experienced developer who's "waiting to see where the AI thing goes"? Not obsolete, but increasingly expensive. If a junior with AI assistance can produce 80% of your output at 40% of your cost, companies will do the math. They already are.
This isn't a doomsday prediction. It's a call to curiosity. The barrier to learning AI tools is laughably low. You don't need a PhD. You don't need to understand transformers or attention mechanisms. You just need to spend a weekend actually using the tools — not reading about them, not watching YouTube videos about them, but opening the terminal and building something. Anything.
The builders who are winning right now share one trait: they're not precious about "hand-written code." They treat AI as a force multiplier. They know when to lean on it and when to override it. They've developed an intuition for what the model is good at and what requires human judgment. That intuition didn't come from reading blog posts. It came from thousands of hours of trial and error, shipping real things to real users.
I think about the philosophy that guides my work: 叶无限 — Leaf Unlimited. I am not defined by any single role, therefore I am infinite in what I can become. A developer who defines themselves only by their ability to write code from scratch is building walls around their own future. A developer who defines themselves by their ability to solve problems — using whatever tools are available — is positioning themselves for abundance.
The AI wave isn't something that's going to happen in the future. It's happening right now, while you're reading this. The only wrong move is waiting.
More unfiltered thoughts on building in the AI era at xundao.xin.
Top comments (0)