Skip to content

Vibe coding was only the beginning

Chapter 00 · Craft · Published · 6 min read
Written by
Alexandre Cormeraie

Author

Cormeraie.

Alexandre Cormeraie
If you're discussing vibe coding with your product owners, you're not structured — and you're at risk. Vibe coding isn't a technology: it's a way of working, and it's not the right one. Here is what I think the next ecosystem looks like: two profiles, one loop, three agents.

Vibe coding was never made for the enterprise

First misunderstanding to clear up: vibe coding is not a technology. It's a way of working — and it's not the right one. The models running underneath are powerful, and they improve day by day, week by week, release after release. That's not what I'm disputing.

Vibe coding was the first step: the one that proved to everyone that this works. It did its job. Now comes the structured version, with the right people at the controls.

What comes next, I call vibe creation. Non-technical people, with no knowledge of code whatsoever, will create software directly — without ever seeing a line of code. They'll care about the design. Does it work well? Is it fast enough? But code, they will never see. That's why keeping the word "code" in the term is a misuse: there is no code left in what these people do.

And that is exactly where the risk sits today: we put people who are not vibe coders on vibe coders' tools. Those tools are made for people who know code and move faster because of them. They are tools for people who know how to use them — who know how to prompt, who know how to do all of that. People without that knowledge will take far more time with them, not less.

The word “code” is already one word too many.

And it's not only a matter of lost time. There is not enough tracking, not enough structure, not enough follow-up. The apps pile up, they burn budget through tokens, and nobody holds the thread. That's what being at risk means.

Which means, in passing, that the CFO will have to change the budgets. Traditionally there's the licence budget and the development budget, and now there's the hosting budget. When you consume AWS or Azure, those costs land on run-side IT budget — and that's where AI spend will land too. So you need to find, among your projects, the ones that let you start moving the lines: shrink the dev budget to grow the IT budget, and make room for AI.

This deserves its own chapter. I'll come back to it.

Vibe coding was only the beginning. You may well choose vibe creation.

Tomorrow's ecosystem: a tech lead and a PO

The right people at the controls are the vibe creators. Because all of this is creation: we're going to create, create, create great things. And to be a creator, you don't need to be a designer — nor to be only a designer, nor only technical. You need to be multi-subject. That's what I call being a digital native.

So tomorrow's ecosystem rests on two profiles. A tech lead and a PO.

The PO is that digital-native profile. They can do design, understand the technical foundation, and evolve their product against the UX being asked for, while respecting the brand. Not a developer in denial: someone whose job is the product, and who now has the means to build it themselves.

The tech lead isn't a developer either. It's a profile who knows the foundation, knows the right things to do, and is there to secure the whole and set the framing. They run several projects at once — that's the whole point: their mission repeats from one project to the next.

Design and development can be handed to AI. That's the important point, and it's what makes the other two possible. Provided there is a technical foundation, with the right tools, the right organisation and the right rules.

The digital-native method

The method fits in two loops.

The first is the initial loop: the tech lead structures the project and creates the agents. It's the only moment anyone works on the base itself.

Then the second one spins. The PO lands on a platform. They won't even see a line of code being written — they arrive on code that already exists; you never start from zero. They update, change their design, modify whatever they like, within the rules the tech lead has put in place. The agent knows those rules for them. When they say "good for me", it lands with the tech lead, who runs validations, security checks, and ships to production.

And that loop spins, and spins, and spins, as much as you want.

initial loop
tech lead structure + agents
the loop that spins
the PO updates
“good for me”
tech lead validates

prod

big changes: start back here
[Artifact 03.01: The two loops]

If there are big, drastic changes, you go back to the initial loop: the tech lead may need to do heavy updates to their agents and to the project's structure to absorb them. But as long as you stay inside the frame, nobody needs to go backwards.

Only two profiles can do this.

That designer-developer barely exists

And to meet these needs, you have to bring a solution. For the tech lead, you need the best generative-coding tools — the ones that let them structure and run their mission across many projects at once.

For the product owner, you need far less technical tools. What are a PO's best reflexes? Leaving comments on Figma, and making small tweaks. So let them comment directly on a live front end, make design changes, and have it just go — exactly as if they were on Figma giving feedback to their designer-developer.

Except that designer-developer barely exists. Behind it, there's an AI, structured by the tech lead.

the PO's annotation
designer agent — the brand
tech agent — the features
fix agent — the standards

framed by the tech lead

[Artifact 03.02: What sits behind the annotation]
A designer agent, structured on the brand. A tech agent, structured on building new features while respecting the stack. A fix agent — design, bugs — making sure the code holds to the company's standards. Three agents, which you can subdivide for specific details, but that is the stack that works.

In the end, it's fairly simple. We were expecting something bigger.

No tool answers this

The hardest part is that today, no tool answers any of this. There is no platform where a tech lead frames things once and a PO then works without ever crossing a line of code. So that tool has to exist.

That's what I'm trying to build with agentation.fr. It's a one-guy company, but that's the vision behind it.

Not convinced?

Doubt what I say? Challenge me.

Pick an AI: it receives the essay and puts it to the test, argument against argument.

a good sparring partner beats applause