This is quite interesting.
Recently, I came across an article titled The Problem with Vibe Coding, which popped a big bubble in the current AI-powered coding tools.

What is Vibe Coding?
Simply put, it’s “when the vibe is right, the code writes itself”—using AI to generate code without needing to understand technical details, just by describing requirements. Sounds great, right? But the article drops a hard truth: Even if AI can write code for you, you still need to know what you’re actually building.

This hits way too close to home.
I’ve seen plenty of teams go wild with GPT, only to realize the final code isn’t what they wanted. Why? Because no matter how advanced AI gets, it can’t answer questions like “Why are we building this feature?” or “What’s the user’s real pain point?”

Technical barriers drop, but cognitive barriers remain
Today’s tools do let non-coders “build things,” but here’s the kicker: Most people still can’t clearly define what that “thing” should even look like. It’s like being handed a magic brush but having no idea what to paint—you’ll just end up with a messy doodle.

The article introduces a concept called Solutioning (solution design), which I think is key. It’s not about coding skills but the ability to turn vague requirements into concrete solutions. Ironically, this has become the biggest bottleneck—because AI hasn’t learned it yet.

PMs in the AI era might need to “regress”
Here’s a harsh reality: Product managers may have to relearn forgotten fundamentals.
Not wireframing, not writing PRDs, but truly understanding business logic and dissecting complex problems. Because when code is no longer the barrier, the person who can define the problem clearly will wield AI like a scalpel.

It’s actually pretty ironic.
For the past decade, we’ve heard “everyone can be a developer,” but now we’re realizing: The scarcest resource isn’t coders—it’s thinkers. AI has democratized execution, but decision-making demands even more expertise.

So don’t let tools mislead you
Too many teams are obsessed with “piling up features using AI,” but users don’t care. At the end of the day, tools are just amplifiers—garbage requirements powered by AI will only give you garbage products faster.

(Original article here: https://solutioning.rocket.new/vibe-coding-problem — worth a read.)

One last brutal truth:
Many AI startups today are the biggest victims of Vibe Coding—they’re frantically building with GPT but can’t articulate whose problem they’re solving. Now that’s dark humor.