·3 min read
Vibe Coding for building features: what I see from the bleachers
Vibe Coding for building features: what I see from the bleachers.
Heads-up: I know I'm not a product manager. I don't spend my days prioritising a backlog or wrestling with a roadmap. But I wanted to write about how AI can be a game changer when it comes to generating new features, based on what I see in colleagues who are on the product side, and on understanding a bit how AI actually works under the hood. So take this for what it is: the opinion of someone watching from the technical side, with curiosity and quite a bit of humility.
And to avoid kicking off with my own theories, let me give the floor to someone who knows. Andrej Karpathy, co-founder of OpenAI, was the one who put a name to all of this on February 2, 2025. He described it as a kind of programming in which «you fully give in to the vibes, embrace exponentials, and forget that the code even exists» (Andrej Karpathy, X, Feb 2, 2025). Something like surrendering to intuition and forgetting that the code is even there. Sounds like a joke, but it points to something serious.
The same old problem: ideas stay on the whiteboard
What strikes me the most about my product colleagues is the sheer number of good ideas that never get tried. Not because they're bad, but because validating them needs a prototype, and a prototype needs time from someone who can build it. Spec, backlog, wait for a development slot... and by the time the slot lands, the idea has cooled off or the context has shifted.
That's where vibe coding feels like a real game changer to me: it brutally shortens the distance between "I just had an idea" and "I've got something to show".
What shifts: trying three ideas instead of one
And this has an effect that to me is the most interesting of all. When prototyping is cheap and fast, you stop betting everything on a single idea. You can put together three versions of a feature, show them, and let real feedback decide — instead of arguing it out in a meeting on the back of opinions.
What used to be "we pick one and pray" becomes "we try a few and see". Again, I'm speaking from the outside: but that sense that many more ideas now get touched is exactly what I notice in the people who are in product.
Generating ideas for new features
Where it helps (and where it doesn't, so much)
To be honest, AI doesn't help me equally with everything. Where I notice the big push is precisely where the bottleneck used to be: prototyping something you can touch and showing it to validate with real people. Exploring several ideas at once and iterating on feedback also benefit, although there's still a lot of human craft involved there.
Which phase does AI help me most with when building a feature?
Not equally in everything. Where I really notice the push is exactly where it used to hurt: going from an idea to something you can show.
Personal and subjective scoring, seen from the technical side rather than from product.
What you shouldn't forget (and here I'll be cautious)
I don't want to sell you smoke. A vibe-coded prototype is just that: a prototype. It's for learning fast and cheap, not for shipping to production as-is. Karpathy said it himself: he saw it as good for throwaway weekend projects, not for serious systems without review. Building the actual feature — with its security, its tests, its maintenance — is still the work of people who know what they're doing. And that, far from taking value away from AI, is what I find beautiful about it: AI removes the friction of the "let's see if this is any good" part, and frees up the human brain for what really matters.
What I'm taking with me
To wrap up, from my humble non-product-manager spot: what blows my mind about vibe coding isn't that it writes code for me — it's that it turns ideas into real conversations much sooner. And in product, from what I'm told, almost everything is played out right there: in how many ideas you manage to put in front of a user before you run out of time.
I don't decide anyone's roadmap. But if this saves some product colleague the frustration of watching good ideas die on a whiteboard, that feels like more than enough reason to write a post.
Thanks for reading this far. See you in the next one.