Boris Cherny, the development manager of Claude Code at Anthropic, said on a podcast that he has not written a single line of code himself since November 2025. He has not become lazy. It simply is no longer necessary: he does everything through Claude Code. Someone who built his career on writing code, and who helped develop the tool, no longer writes code. I think that says a lot about the times we live in.
I am not a developer. I am a programme manager and I never wanted to be anything else. Even so, over the past months I have built two working web applications: finediningfinder.nl and the site you are reading now, miliarium.nl. I did it without a team of developers, without an external agency and without ready-made website builders, with AI and Claude Code in particular, in a way the tech world now calls ‘vibe coding’.
Below I describe how that was possible, and why I expect it to change the market for software development profoundly.
What is vibe coding, exactly?
The term ‘vibe coding’ was coined in early 2025 by AI researcher Andrej Karpathy. He used it to describe a way of developing software in which you give yourself over completely to the AI: you describe in plain language what you want, the AI generates the code and you steer based on what you see and feel, the vibe, without digging into the underlying code yourself.
That sounds light-hearted and perhaps even a bit irresponsible, but behind that casual description lies a serious shift in working method. Traditional software development requires an exact translation of a business need into technical specifications, then implementation in a programming language, and then testing, debugging and documenting. With vibe coding that chain is shortened considerably: the distance between idea and working product is measured in hours instead of weeks.
The AI works less like a search engine that gives you the right Stack Overflow answers and more like a conversation partner that understands what you mean, states its assumptions and suggests alternatives. The conversation is the development process.
Twenty years of learning what users really need
To understand why vibe coding works for me, I need to step back for a moment.
In more than twenty years as a programme and project manager in the financial sector, on assignments involving payment systems, core banking implementations and SEPA/SWIFT infrastructure, I have learned one thing that outweighs the rest: the gap between what a client says and what a user needs is almost always bigger than anyone thinks.
My job was to bridge that gap. I talked with directors about strategy, with process analysts about workflows and with end users about the daily frustrations that never make it into a specification. The result was user stories, structured descriptions of what a user wants to achieve, in which context and with what goal. Those stories went to software vendors such as Temenos or ACI Worldwide, who turned them into features.
Over the years I have drifted, without noticing, from pure planning and coordination towards something closer to product management. I represent the client towards the vendor, not only on planning and budget but on substance too. I check whether what is being built fits business reality. I am the one who says: “This is technically correct, but the user is going to get stuck here.”
Being able to say precisely what software should do, for whom and why, turned out to be the key that vibe coding unlocks.
From user stories to AI prompts: a smaller step than you think
When I first worked with Claude Code, something struck me. My prompts for the AI looked suspiciously like the user stories I write for a vendor.
A user story has a fixed structure: As a [type of user] I want [action or functionality] so that [business goal]. It is all about that so that, stating the goal behind the request. A developer without that context builds literally what is written. A developer with that context makes better choices.
It is the same with AI prompts. “Build a search function” produces something different from “Build a search function that lets a user filter by cuisine, city and price range, because the goal is that someone finds a relevant restaurant for a specific occasion within thirty seconds.” With the second prompt the AI thinks along on interface choices, error handling and user experience.
Thinking along like that, formulating functional requirements that reflect what the user intends, is not something you learn in a weekend. It comes from years of work in the space between business and technology. That is why I think vibe coding is so effective in the hands of someone with that background.
finediningfinder.nl: an idea that became reality
A concrete example. finediningfinder.nl is an application that helps people find fine dining restaurants in the Netherlands, filtered by location, cuisine, Michelin stars and other features. I had had the idea for a while: as a lover of good restaurants I saw a gap in the market myself, because existing platforms are either too generic or too commercially driven.
In the past this idea would have stalled at “interesting, but I have no developer for it.” Having an agency build an MVP easily costs tens of thousands of euros and takes months, with the risk that what is delivered does not match what you had in mind.
With Claude Code it went differently. I started by describing the core of the application: what must a user be able to do, what data do I need, how should the interface feel. I did that in business terms and not in technical ones. From there we built the application step by step, the AI and I. I checked every iteration against the original user need, and where it deviated I steered by describing what was missing and why, without writing code.
The result is a working application that looks professional and does what I want it to do. The lead time was a fraction of what a traditional project would have required.
miliarium.nl: the website as a lab
Miliarium.nl, this website, also came about through vibe coding. That was a deliberate choice. I did not only want a blog, I wanted to understand what it is like to build and run a complete web environment with AI assistance.
In doing so I noticed that vibe coding is not a one-off development moment but a continuous process. The website grows with my thinking. If I want to add a new section, I describe what I want. If I want to organise the navigation differently, I explain why, from the point of view of user experience. The AI adjusts, I review and I steer.
I increasingly realise that this is the iterative development process that has been best practice in the professional software world for decades: agile, scrum, continuous feedback and step-by-step improvement. The difference is that I now go through the cycle myself and do not depend on an external team with an agenda of its own.
The professional relevance: what this means for my work
For me vibe coding is not a hobby experiment, it has direct professional relevance.
As a programme manager I usually work in complex environments with multiple vendors, international stakeholders and tight regulatory frameworks. There it is valuable to be able to build prototypes quickly, proofs of concept that show how a solution could work. They bridge the communication gap between business and IT better than a PowerPoint presentation.
There is something else. Vibe coding has deepened my understanding of software development in a way that years of project management had not. I now have a better sense of what developers run into, which choices are made when an application is built and why certain user stories are harder to implement than they look. That makes me a better conversation partner for the vendors and development teams I work with.
The wider meaning: democratising software development
Now the big question: what does this mean for the market?
My position is that vibe coding profoundly democratises access to software development. That does not mean everyone can build complex enterprise systems tomorrow, and that nuance matters. It does mean that the ability to turn a good idea into a working digital product is no longer reserved for people with a technical education.
What you still need is the ability to formulate precisely what you want, for whom and why. That is domain knowledge and experience, and an understanding of user needs and business context: the skills that lead to good user stories in a traditional project environment.
The developer as a craftsman who writes code is, in this new landscape, partly replaced by the product thinker who understands what needs to be built. The AI takes care of the implementation. For people with technical knowledge that need not be threatening, because good engineers remain indispensable for complex, large-scale systems. But it does shift who is at the wheel when new products are developed.
The anecdote I began with shows that this is already becoming usual among tech professionals themselves: Boris Cherny, the man who co-built Claude Code, no longer writes code, because the AI does it better and faster. If that becomes the norm in Silicon Valley, I expect it is a matter of time before it becomes the norm in every sector.
An honest caveat
Vibe coding is not a miracle cure without catches. There are limits.
The quality of what you build stands or falls with the quality of your instructions. A vague idea produces vague software. Without a feel for user experience you build something that functions but does not work. And for large-scale, critical systems, such as the payment infrastructure I am professionally involved in, vibe coding is not yet a mature alternative to structured software engineering with all the quality assurance that comes with it.
For exploring ideas, building prototypes and making early-stage products I do think it is a big change. And it is developing fast.
Conclusion: this is only the beginning
I build software. Five years ago I would not have thought that possible. For me it is not a curiosity or a technical experiment either, but a serious activity that leads to products people use.
That has become possible through two things together: the AI capabilities of platforms such as Claude Code and the skills I have built up over twenty years of programme management. The first is changing fast and becoming more accessible. The second is not easy to copy.
Vibe coding makes building software more accessible, but the quality of what you build is still determined by how well you understand what your user needs. The technology changes, the core of good product thinking does not. That is why I expect this to be only the beginning.
Jan-Willem writes on miliarium.nl about technology, management and where the two meet. He is Senior Programme Manager and Managing Partner of Ductus B.V., a boutique IT consultancy specialising in payment systems and financial technology.