The Vibe Coding Revolution Doesn’t Ask for Permission

AI is a tsunami. You don’t negotiate with a tsunami.

For the past few months, I’ve been rediscovering my craft: programming.
Or rather, I should say: software creation.

Because today, if you’re using the right tools and methods, building software isn’t really about writing code anymore. It’s about designing—pure and simple.

You think, you formalize, you guide the (Deus-EX?) Machina that turns your ideas into working code. The execution speed is phenomenal—inhuman.

Sure, sometimes it hallucinates, makes mistakes, or heads in a suboptimal or unsustainable direction. But that’s exactly the point: the ex-programmer turned designer is there to tame the beast, to steer the stallion straight.

I know it, I see it, I hear it: there’s pushback in the dev community. A wave of resistance to the excitement. Some call it “fancy autocomplete” or even a “gimmick.”
Big mistake. It’s a huge misread of what’s happening. The whole profession is shifting.

AI won’t replace programmers—at least not in the coming years (decades? who knows). But programmers who embrace AI? They will replace the others. In fact, they’re already doing it, as I type these words.

All of this is emerging now under the name Vibe Coding.

Here are some notes I’ve jotted down along the way. I noticed they sparked a lot of interest when I shared them, so I’m collecting them here. Also, it’s the perfect excuse to launch this blog—freshly built… by Cursor (well, with Cursor…)

Vibe Coding: A Shift in the Developer's Role

Intro to Vibe Coding, by Garry Tan, CEO of YCombinator.
Crystal clear and perfectly explained. Start with this short YouTube clip—it nails the concept and the origin of the term.

Want to dig deeper? I highly recommend this podcast episode, also from YC:

The conversation is sharp and insightful. One caveat: they don’t really distinguish between pure vibe coding—i.e., not knowing what the machine is doing—and truly AI-augmented development. That difference matters. A lot. Even if the machine can generate working code on its own, it tends to drift (more on that later), and understanding the code you're shipping is non-negotiable. It’s the only way to build something sustainable.

The developer writes. The Vibe Coder reviews.

Scaling Up: Productivity Explodes

First major point: a complete shift in scale.
A single dev, with the right AI tools, can work 10, 20, even 100 times faster than before. I’m not being dramatic—I’ve lived it.

You can generate in minutes what used to take days. This isn’t “a little faster.” It’s a different game entirely. It changes not just how we work, but how we think about work.

Less Ego, More Agility

Direct consequence: less ego, more agility.

Before, we got attached to the code we wrote. It took time and effort to build something functional. With Vibe Coding, you can spin up entire features in a few minutes (with some tweaking). Deleting and restarting? Easy. Painless.

Pivoting becomes second nature. Honestly, it’s fun. AI frees us from emotional baggage—and unleashes real creativity.

From Software Engineer to Software Designer

The core of the job is shifting. The fundamental skillset of a developer is leveling up.
We become architects, product owners, Software Designers.

The power isn’t in typing lines of code. It’s in having a holistic vision, defining the right constraints, and guiding the AI toward the desired outcome.
It’s a deep change in posture.

The Developer's Role Is Still Critical

Make no mistake: this isn’t the end of programming. That’s not what I’m saying. AI can prototype, explore, and test. But building stable, long-lasting platforms? Not yet. Maybe never.

As prompts and iterations stack up, projects can drift. You end up with fragile architectures, hard-to-maintain messes.

Our job evolves—we become part Team Lead, part Software Architect. The team? The AI.

We define structure, enforce patterns, push for clean code separation and maintainable architecture.
In short, we bring the vision, the abstraction, the system-level thinking.

What About Juniors and Beginners?

Hot topic, but vital: what do we do with junior devs?

AI can definitely save them time. But it doesn’t teach them much.
Deep understanding of software engineering—patterns, abstractions, code boundaries—is still key to catching when the AI is off-track.

Without those foundations, juniors won’t see the traps.
And if they start coding with AI from day one, how do they actually grow?
Open question. A big one for schools and training programs. Personally, I think banning AI use during dev internships makes sense. Maybe even for all devs with less than three years’ experience.

Team Dynamics: Everything Changes

From 0 to 1: Bootstrapping

Before, you needed 5 to 10 people to get a project off the ground.
Now, two experienced devs with solid AI tools can build the whole thing. Especially when using serverless platforms like the excellent Railway.

From 1 to N: Scaling Up

At that point, you do need a bigger team—architecture, ops, monitoring, QA… but not at yesterday’s scale.
This is a different ballgame now.

It’s a Cultural Shift, First and Foremost

The biggest bottleneck I see isn’t technical. It’s human. Cultural.
Resistance is strong—even (especially?) among the most talented.

The idea that an LLM might do better, faster (radically faster!)—that stings. It hurts our developer egos. This was a high-skill job. It still is. But now? It’s highly automatable.
The backlash is emotional. Understandable. But pointless.

People downplay it. Ignore it. Push back.

But the tsunami is here. It’s rising. It’s fast.
It doesn’t ask for permission.

Tools & Methods to Get Started with Vibe Coding

Let me end this post with some hands-on pointers to try it for yourself. A quick rundown of what I’ve tested and validated:

• Basic manual mode: ChatGPT. Prompt, generate, copy-paste, test. Great for quick ideas and prototypes.

• Custom GPT: Train a model on your project’s source files and guide it with a system prompt. More context-aware. A step up. Requires the paid ChatGPT plan (~$20/month).

• VSCode + Copilot: This is real AI-augmented development. The editor becomes smart, prompts are contextualized. You accept suggestions, collaborate with the model. Game-changer. Starts at $10/month, free 2-week trial.
  Note: You can connect Copilot to any LLM—handy for testing setups.

• Cursor: Next level. The editor reasons, self-prompts, understands the whole project. It navigates your codebase to grasp context and changes. Almost never misses. Super impressive. Starts at $20/month.