# Vibe coding is over. The hard part was never the demo.

Google now teaches vibe coding to a million-plus people in a free five-day course. When the thing you were proud of becomes a weekend class, that skill just stopped being your edge. But here's the part the headlines miss: vibe coding was always good at the easy 80% — the demo — and useless at the 20% that decides whether software survives. The skill that's actually scarce now isn't generating code. It's the judgment to know whether the code you got is any good.

- **Author:** Fedor Chishchin
- **Published:** 2026-06-06
- **Tags:** careers, methodology, ai-native
- **Source:** https://fedorthinks.com/blog/the-demo-was-never-the-hard-part

---

Here's the clearest sign that an era just closed: Google and Kaggle are running a
free five-day course that
[teaches vibe coding to over a million learners](https://blog.google/innovation-and-ai/technology/developers-tools/kaggle-genai-intensive-course-vibe-coding-june-2026/),
explicitly aimed at people "without formal programming experience." Building an
app by typing what you want in plain English is no longer a secret trick. It's a
weekend class with a capstone project about running a virtual farm.

When something becomes a free course for a million people, it stops being an
advantage. That's not an insult to vibe coding — it's just what commoditization
looks like. And it forces the honest question for anyone whose value was "I can
get the AI to build the thing": if everyone can now do that, what exactly are
you for?

## Vibe coding was always great at the easy part

Strip the romance away and vibe coding does one thing extremely well: it gets you
to a working demo, fast. Describe an app, watch it appear. That genuinely used to
take real skill and now takes a sentence. Wonderful.

But a demo is the easy 80% of software. The hard, unglamorous 20% is everything
that comes _after_ the thing first runs — and that's exactly where vibe coding
falls down. The data is brutal. As AI writes more of our code,
[code churn is up 41% and duplication has quadrupled](https://medium.com/@ahmed.hafdi.contact/vibe-coding-is-over-what-comes-next-is-much-harder-9fe95b509850),
while the careful refactoring that keeps a codebase alive has collapsed. A
security firm built fifteen identical apps with the five most popular AI coding
tools and found
[sixty-nine vulnerabilities, six of them critical](https://medium.com/@ahmed.hafdi.contact/vibe-coding-is-over-what-comes-next-is-much-harder-9fe95b509850).
The summary that keeps showing up is the right one: vibe coding can _build_ a
product. It cannot _maintain_ one.

So the demo got free. The part that was always hard — keeping software correct,
secure, and changeable over months — got, if anything, harder, because now
there's far more sloppy code to keep alive.

## The scarce skill is judgment, not typing

This is the reframe that matters for your career. When generating code costs
nothing, the value moves to the one thing the generator can't do for you: knowing
whether what it produced is any good.

The most valuable engineers right now
[aren't the ones writing the most lines](https://medium.com/@ahmed.hafdi.contact/vibe-coding-is-over-what-comes-next-is-much-harder-9fe95b509850) —
they're the ones who can direct the AI and _evaluate what it gives back_.
Architectural judgment. Security intuition. The ability to take a vague problem
and break it into clean, reviewable pieces before any AI touches it. That's the
skill the free course doesn't hand you, because it can't be transferred in five
days — it's built from fundamentals. As one write-up put it bluntly: without
fundamentals,
[vibe coding becomes blind trust](https://medium.com/@ahmed.hafdi.contact/vibe-coding-is-over-what-comes-next-is-much-harder-9fe95b509850),
and that's the risky kind.

This is the same split I keep coming back to: the difference between
[knowing what and knowing how](/blog/knowing-what-knowing-how). The AI is a
spectacular "how" machine — it produces the code. You're paid for the "what" and
the "is this right" — what to build, whether the output is sound, where it'll
break under load or attack. That judgment is the part that didn't get
commoditized, and the flood of generated code makes it more valuable, not less.

## What this means if you're early in your career

There's a real worry under all this, and I won't wave it away. If the AI does the
beginner work — the simple functions, the boilerplate — how do you ever build the
judgment that only comes from doing that work yourself? This is the
[disappearing bottom rung](/blog/the-bottom-rung-is-gone) problem, and it's
genuine.

The answer isn't to refuse the tools; that ship sailed, and Google's teaching the
course. The answer is to use them while deliberately keeping the understanding the
tools would let you skip. Read the code the agent wrote. Ask why it chose that
approach. Break it on purpose. Learn the data structures, the security basics, the
way a database actually behaves — not to out-type the AI, which you can't, but to
out-_judge_ it, which you can. The goal isn't to be the person typing. It's to
become the person who can tell, in ten seconds, that what got generated is quietly
wrong.

## The takeaway

"Vibe coding is over" doesn't mean the tools go away — they just became
electricity, available to everyone, taught to a million people for free. What's
over is the brief window where _using_ them was the skill. The hard part was never
producing the demo. It was everything the demo doesn't show: keeping the thing
alive, safe, and honest after the magic wears off.

That part still belongs to the people who understand what they're looking at. The
free course makes a million more code generators. It doesn't make a single
engineer who can tell good code from a future breach — and that, now more than
ever, is the job.
