A Danish tech executive’s declaration that he will “probably never hire a software engineer again” has ignited a fierce debate about the future of programming jobs, the reliability of AI-generated code, and whether the entire craft of software engineering is on the verge of being hollowed out by tools that let anyone — literally anyone — build applications by describing what they want in plain English.
The executive is Morten Bødskov, CEO of a Danish startup called Workbase. In a post on X that quickly went viral, Bødskov laid out his reasoning with blunt confidence: the company’s AI-generated codebase is now roughly 95% written by artificial intelligence, and the remaining human engineers on staff are being transitioned into product-focused roles rather than traditional coding positions. “I will probably never hire a software engineer again,” he wrote, adding that what he now looks for in candidates is product sense, not the ability to write Python or JavaScript. The post, first reported by Futurism, drew thousands of reactions — some admiring, many skeptical, and a good number outright hostile.
Bødskov’s comments land at a moment when the software industry is grappling with the practical implications of so-called “vibe coding,” a term coined earlier this year by AI researcher Andrej Karpathy. The concept is simple and, to veteran engineers, either thrilling or terrifying: instead of writing code line by line, a user describes what they want an application to do, and an AI model — typically something like OpenAI’s GPT-4, Anthropic’s Claude, or one of several specialized coding assistants — generates the code automatically. The user doesn’t need to understand what the code does. They just need to know what they want.
That’s the pitch, anyway.
The reality, according to many working engineers, is considerably messier. And the gap between the pitch and the reality is where the real story lives.
Start with what Bødskov actually said. His argument isn’t that AI can replace all software engineers everywhere. It’s narrower than that: for his particular company, at its particular stage, the combination of AI coding tools and product-minded generalists is sufficient. Workbase, which builds AI-powered sales tools, is a small startup — not a company running millions of lines of legacy code or managing infrastructure at scale. The distinction matters. A 20-person startup deciding it doesn’t need to hire traditional engineers is a fundamentally different proposition than, say, Google or JPMorgan Chase reaching the same conclusion.
But the reaction to his post suggests it struck a nerve far beyond the startup world. On X, software engineers pushed back hard. Some pointed out that AI-generated code is notoriously difficult to debug, that it frequently introduces subtle security vulnerabilities, and that a codebase built almost entirely by AI tools amounts to technical debt on an industrial scale — debt that will eventually come due, probably at the worst possible moment. Others noted that Bødskov himself doesn’t appear to have a deep engineering background, raising questions about whether he fully understands the risks of the approach he’s championing.
“This is going to end in tears,” one engineer wrote in a widely shared reply. “Not because AI can’t write code. It can. But because nobody at that company will understand the code when something breaks.”
The criticism isn’t purely theoretical. Recent months have seen a growing body of evidence that AI-generated code, while impressive in demos and prototypes, introduces real problems in production environments. A report covered by Futurism noted that studies have found AI-generated code to contain more bugs and security flaws than human-written code, particularly in complex systems where context and nuance matter. The code works — until it doesn’t. And when it doesn’t, diagnosing the failure requires exactly the kind of deep engineering expertise that companies like Workbase are choosing not to hire.
This tension sits at the heart of the vibe coding phenomenon. The tools are genuinely powerful. GitHub Copilot, Cursor, Replit’s AI features, and a growing roster of competitors can produce functional code with startling speed. For prototyping, for building internal tools, for spinning up a minimum viable product to show investors — these tools are extraordinary. Nobody serious disputes that.
The dispute is about what happens next.
Software, in the real world, isn’t just about getting something to work once. It’s about maintaining it over years. It’s about scaling it to handle millions of users. It’s about ensuring it doesn’t leak customer data, crash under load, or silently corrupt a database. These are engineering problems, not product problems, and they require people who understand systems at a level that current AI tools simply don’t.
Or do they? The counterargument, advanced by Bødskov and others in the pro-vibe-coding camp, is that AI tools are improving so rapidly that today’s limitations are tomorrow’s footnotes. If Claude or GPT-4 can write 95% of a codebase today, the argument goes, next year’s models will handle 99%. The remaining 1% can be managed by a small number of highly skilled engineers — or perhaps by the AI itself, with humans serving as reviewers rather than writers.
There’s some evidence for this view. OpenAI’s latest models show marked improvement in code generation accuracy compared to versions released just six months ago. Anthropic’s Claude has become particularly adept at handling longer, more complex coding tasks. And new tools like Devin, billed as an “AI software engineer,” are attempting to handle not just code generation but debugging, testing, and deployment — the full lifecycle of software development.
Still, the gap between “impressive demo” and “production-ready system” remains vast. Devin’s own launch was marred by reports that its capabilities were significantly overstated in promotional materials. And even the most optimistic AI researchers tend to acknowledge, in private if not always in public, that fully autonomous software engineering is years away at minimum — if it’s achievable at all.
The labor market implications are already materializing, though, regardless of where the technology ultimately lands. Hiring data from multiple sources shows that entry-level software engineering positions have become dramatically harder to land over the past 18 months. Companies that once hired dozens of junior developers are now hiring a handful and equipping them with AI tools, expecting each person to produce what previously required a small team. The result is a market that’s increasingly bifurcated: senior engineers with deep expertise remain in high demand, while junior engineers face a job market that’s contracting around them.
This is the part of the story that doesn’t get enough attention. When a CEO like Bødskov says he’ll never hire another software engineer, the implication isn’t just about his company. It’s about the pipeline. If startups stop hiring junior engineers, where do senior engineers come from in five or ten years? The craft of software engineering has always been learned primarily on the job — through code reviews, debugging sessions, production incidents, and the slow accumulation of hard-won knowledge about how systems actually behave. Remove the entry point, and you don’t just eliminate junior roles. You eventually eliminate the senior roles too, because nobody is learning the skills needed to fill them.
Some in the industry have begun calling this the “hollowing out” problem. It’s not unique to software — medicine, law, and finance have all faced versions of it as automation has crept into entry-level tasks. But software’s version is particularly acute because the technology doing the displacing is advancing at a pace that makes planning difficult. A medical school can reasonably predict what skills a doctor will need in ten years. A computer science department trying to predict what skills a software engineer will need in 2030 is essentially guessing.
Bødskov, for his part, seems unbothered by these concerns. His post on X was unapologetic, even provocative, in tone. He framed the shift as an inevitability that smart companies should embrace now rather than resist. And he’s not alone. A growing cohort of startup founders — particularly those building AI-native companies — have adopted similar stances, arguing that the traditional model of large engineering teams writing code from scratch is already obsolete.
The question is whether they’re right, or whether they’re confusing the early, forgiving stages of a startup’s life with the brutal realities of scaling a product that real customers depend on.
History offers some cautionary parallels. The no-code movement of the late 2010s promised something similar: anyone could build software without writing a line of code. Tools like Bubble, Webflow, and Airtable gained massive followings. And they delivered real value — for certain use cases. But no-code tools never displaced traditional software engineering. They carved out a niche. The hard problems — performance, security, reliability, integration with complex systems — still required engineers who understood what was happening under the hood.
Vibe coding may follow the same trajectory. Or it may not. The AI models underlying today’s coding tools are categorically more capable than the no-code platforms of five years ago. They can handle ambiguity, generate novel solutions, and adapt to feedback in ways that rule-based systems never could. The ceiling is higher. How much higher is the open question.
What’s clear is that the conversation has shifted. A year ago, the debate was about whether AI would affect software engineering at all. That debate is over. It already has. The new debate is about degree: how much of the work can AI handle, how quickly it will improve, and what role human engineers will play in the systems of the future. Bødskov’s viral post didn’t settle that debate. But it crystallized the fault lines in a way that’s hard to ignore.
For now, the safest bet is probably somewhere between the extremes. AI coding tools are real, they’re powerful, and they’re going to reshape how software gets built. But the idea that a company can operate indefinitely with zero engineering expertise — that product sense alone is sufficient, that understanding what you want is the same as understanding what you’ve built — is a bet that a lot of experienced technologists would decline to make.
Then again, experienced technologists have been wrong before. And Bødskov, whatever his critics say, has the advantage of putting his thesis to the test in the market. If Workbase succeeds — if it scales, if its AI-written codebase holds up, if its customers are well-served — then the argument changes. Not because one startup proves a universal rule, but because it proves a possibility. And in technology, proving a possibility is how everything starts.
The engineers watching from the sidelines aren’t wrong to be concerned. But they’d be wrong to assume this is just hype. Something is shifting. The only question is how far it goes.
WebProNews is an iEntry Publication