Body:
Yesterday, a news headline nearly made me spit out my coffee—a complete coding newbie earned 170,000 yuan in five days using Vibe Coding, while I, a decade-long veteran, was still stuck fixing product managers’ demands. This thing was hyped by Karpathy last year as a “programming paradigm revolution,” but now Silicon Valley is already dismissing “vibe programmers” as amateurs. Nope, I had to see these AI interns for myself.

I opened the hottest AI coding tool and spoke into the mic: “Make a Chrome extension that automatically fixes bugs.” Three seconds later, a pop-up appeared: “Suggested rename: ‘Code Nurse,’ with a pink heart icon generated.” Staring at the UI that looked like it was ripped from a Barbie game, I silently turned off voice input.

But what really broke me was the SQL injection test. I asked the AI to build a user login system—it handled encrypted transmission just fine but left a little “1=1” Easter egg in the password field. It’s like hiring a chef who can prepare a royal banquet but secretly slips laxatives into the dishes. That security firm’s report, Claude Code Enables Silent Hacker Intrusions, wasn’t exaggerating.

The most surreal part? A course at Beijing Normal University. A friend showed me student projects built with Vibe Coding—one sophomore girl used AI tools to map Dream of the Red Chamber character relationships into a 3D galaxy. Her GitHub commit history was 80% “Felt wrong, starting over.” In my company, she’d have been fired eight times over. But her professor said, “That’s exactly the kind of intuitive iteration we want.”

Now I get why Jeff Dean compared AI programmers to interns. Last week, I had five different tools write the same shopping cart feature: the first stored cookies on a blockchain, the second defaulted prices to Vietnamese dong, and the third—the piĂšce de rĂ©sistance—linked the checkout button directly to the author’s PayPal. I had to admit: managing 50 AIs requires an architect who can keep them in line, like a kindergarten teacher wrangling toddlers.

But there were legit “this slaps” moments too. At 3 AM, when ops woke me up to tweak a promo page, I mumbled to the AI, “Make the button bigger and add glitter animation.” Three minutes later, it delivered. The code was spaghetti, but when the product manager dropped a “yyds” in the group chat, I finally understood the hype—this thing isn’t for writing rocket control systems; it’s digital ibuprofen for the modern worker.

Silicon Valley is already pivoting to “elite battle-tested engineers”—basically, hardcore pros who can tame AI. The most impressive guru I’ve seen wrote prompts like legal contracts: “Use React 18 + TypeScript 4, ban any types, every comment must include the author’s last four ID digits.” The resulting code was cleaner than what our juniors wrote.

Now, whenever I see headlines like “Vibe Coding Will Disrupt Traditional Development,” I laugh. Last December did bring a qualitative shift, as Karpathy said—but the change wasn’t in the AI. It was in us. Learning to sift gold from AI-generated garbage is the new programmer’s core skill. Like my grandma said: even the best washing machine needs someone to sort the laundry—unless you want silk shirts churned with jeans.