AI and developers: who is still building, if AI writes the code?

If AI writes the code, who is still building? The same people — just no longer at the same point in the chain. Generative models write functional code, fast, and increasingly well; they don't decide on an architecture, don't check their own output, and answer for nothing in front of a client or a judge. The developer's job isn't disappearing because of it. Its value is migrating toward design, oversight, and accountability, and that shift changes what a Swiss SME must demand from a supplier or a new hire in 2026.

Published January 2026, reworked May 2026 — this version is a full rewrite.

What AI already does in software development, beyond debate

Stack Overflow's annual survey shows it: in 2025, a significant share of professional developers had already built generative AI into their daily practice^[1]. Its spread is no longer up for discussion. Three uses have moved from gadget to production tool.

Assisted writing comes first. An assistant proposes complete functions from an intent expressed in plain language, recognises the conventions of the project at hand, and absorbs repetitive code — standard structures, format conversions, boilerplate screens — that used to eat up a substantial share of the working day. Automated code review follows: a first machine pass catches likely bugs, security flaws, and convention drift, while human review focuses on business logic. Test generation completes the picture. The least-loved task in the job now produces a usable baseline in a few minutes — to be reviewed critically, but far better than the patchy coverage that time pressure produced everywhere.

I checked this on my own keyboard. FiscalDoc, the local application that files my tax documents and replaced a SaaS subscription costing around CHF 1,400 a year, wasn't written line by line: I built it in dialogue with a coding assistant, over three evenings. The redesign of mcva.ch, the site you're reading this page on, followed the same path.

If the machine writes the code, why pay a developer?

Because writing was never the hard part. In 2026, generative AI doesn't eliminate the developer's job: it shifts its value from writing code toward design, oversight, and accountability.

Design, first. Breaking down a fuzzy problem into workable sub-problems, choosing an architecture that will survive future changes, identifying critical dependencies: models don't absorb these decisions. They execute well what is well described to them — but someone still has to be able to describe it well.

Oversight, above all. A developer who accepts what their assistant proposes without review builds invisible technical debt. It doesn't show at delivery. It comes due months later, when code that no one truly understood needs to evolve. Knowing how to evaluate generated code, spot a subtle hallucination, reject a plausible anti-pattern: that has become a skill in its own right, and it's what separates developers who master their tools from those who simply prompt.

Accountability, finally. The obligations of the Federal Act on Data Protection^[2], the requirements of regulated sectors, the legal consequences of technical choices bind the company and identifiable individuals. A model carries none of that.

Should you still hire juniors?

Yes — but the trade-off has changed. A junior equipped with an assistant now produces a volume of code that used to require several years of experience. The gap between junior and senior is narrowing on pure coding. It isn't narrowing on the rest: judging the relevance of generated code, holding the overall vision of a system, anticipating edge cases that models don't foresee by design.

Fifteen years leading digital projects for major international brands taught me a simple rule: the cost of a project is decided less at the moment the code gets written than at the moment someone decides what needs to be written. A team without a senior accumulates debt it discovers at the worst possible moment. A team without a junior pays an experienced profile for tasks the tools handle very well. The balance is built project by project, and it deserves to be set before signing, not after. To place this trade-off within a broader approach, I laid out the process in What should a Swiss SME do about AI in 2026?.

What this shift changes for your software projects

The shift in the profession has a direct economic consequence: custom-built software is becoming a serious option again. When writing code stops being the bottleneck, a tool tailored to your processes costs a fraction of what it used to, and the generic SaaS subscription stops being the default choice. I laid out this comparison in SaaS versus custom-built in Switzerland, and it's the logic behind the solutions I build for SMEs.

Beware the reverse shortcut, though. A tool generated quickly and poorly supervised costs more than a subscription: it costs the trust of the team using it, then the price of a rebuild. The right question fits in a few words — who supervises, and under what rules?

Three things to watch before signing a project in 2026

Confidentiality tops the list. Some assistants send the analysed code — and sometimes the data it contains — to external servers under foreign jurisdiction. For any project touching personal data or trade secrets, the FADP requires addressing this at the scoping stage, not discovering it midway through.

Intellectual property comes next. Swiss law hasn't settled the status of code produced through dialogue with a model. Documenting tool usage, keeping structuring prompts, maintaining substantial human review: this is a prudent safeguard, and it's one of the points I lock down in every custom project before a single line is written.

Ongoing training, finally. A developer still at 2024's level is already behind on 2026 practices. This investment belongs with a technical team's normal operating costs, on the same footing as licences — I've devoted a full piece to it: Training your team on AI.

Key takeaways

— AI writes code; it doesn't design the architecture, doesn't check its own output, and carries no legal responsibility. — A developer's value is shifting toward design, oversight of generated code, and compliance: demand exactly those three. — Supervised custom development is becoming economically accessible to Swiss SMEs again; generic SaaS stops being the default choice.

FAQ

Will AI replace developers? No. It absorbs tasks — writing repetitive code, a first review pass, test generation — but it doesn't decide on an architecture, doesn't check its own output, and carries no legal responsibility. The job is recomposing around design and oversight, and those are the skills gaining value.

Is an AI-equipped junior enough for an SME project? For a simple, well-scoped internal tool, often yes. As soon as the project touches sensitive data, multiple integrations, or a question of long-term viability, oversight from an experienced profile remains essential: it's what prevents the invisible technical debt of unreviewed generated code.

Does AI-generated code raise a legal issue in Switzerland? The law isn't settled on the intellectual property status of generated code, and the FADP applies as soon as personal data passes through an external tool. The prudent approach is to choose tools whose data flows you control, document their use, and maintain substantial human review.

Is AI-assisted custom development still expensive? Markedly less than before: what used to be measured in months of agency work is now often measured in days, for an equivalent scope. My FiscalDoc application — built in three evenings, zero francs in operating cost — gives a sense of the order of magnitude on a small scale. The real cost shifts toward scoping and oversight.

And for your own software projects? The AI Usage Diagnostic: sixty minutes to map your actual workflows, identify what deserves custom software, what stays in SaaS, and what needs no AI at all. Book a diagnostic

Sources

[1] Stack Overflow, Developer Survey 2025. survey.stackoverflow.co/2025/ [↩]

[2] Federal Act on Data Protection (FADP), revision of 25 September 2020, in force since 1 September 2023. www.fedlex.admin.ch/eli/cc/2022/491/fr [↩]

Jérôme Deshaie is CEO and founder of MCVA Consulting SA, an augmented agency based in Valais. Fifteen years serving major international brands, now working directly with Swiss SMEs. Background.