A low-IQ take on AI replacing software developers

I like AI as much as the next guy, but this hyperbole around AI is driving me nuts.

In case you've missed it, various thought leaders have been warning that AI development is on such an upswing that within the next 6–18 months, most regular office jobs will be automated by AI.

Call me slow, but I can't envision how this would be done in practice.

I'm not talking about some hypothetical scenario in the distant future. The thought leaders have stated very clearly that all of this will happen during 2026–27. To me, that means it has to happen using the tools and technologies (LLMs) we already know.

A few thoughts from a software developer's perspective:

Backlog zero

Never in my 15+ years in the industry have I seen software developers this busy and excited about software development. You'd think outsourcing your thinking and implementation to agents would shorten your workday. In practice, many software developers actually work more than before. There are people who are legitimately anxious if they don't have a model running 24/7 in the background planning, iterating, or implementing something.

This aligns well with the notion that once the unit price of something drops, the demand for it increases. We now have the opportunity to experiment and build features that were previously out of reach due to the time and resources required.

If you have ever worked in IT, you know that every piece of software with real users has a backlog almost as heavy as the node_modules directory. Companies literally employ first- and second-line support, project managers, and product owners, and have developed all kinds of processes to manage this constant stream of bug reports and feature requests. The idea of backlog zero has been a complete utopia, not to mention everything that never even made it onto the backlog because software development is so slow and expensive.

Even if productivity skyrockets, why would you settle for the quality and feature set of the old world? Do you really think people are going to stop once their current backlog is empty?

"I'm not a doctor, but I'll take a look"

Often you hear predictions that in the future (#soon) everyone will be coding because it's finally within reach. I hate to be the realist, but I find the thought ridiculous. The average person who was incapable of scheduling a VCR recording in the '90s, burning a DVD in the early 2000s, or changing the Wi-Fi password in the 2010s is going to suddenly start prompt-engineering personalized software that will replace the services they currently use?

If it's not the average person who will create their own tailored software, then who is it? Are small-business owners going to vibe-code their booking system and membership management? Is the middle manager, who currently spends 30+ hours a week in meetings, going to suddenly clear their calendar and start cranking out business-critical systems? Is the CEO's 14-year-old tech-savvy nephew going to be trusted with building the new bespoke ERP for the company? Or is it a small group of AI-native software developers who are going to become domain experts in all industries and disrupt the whole market? And what stops the people who got laid off from vibing and setting up shop themselves?

Or maybe it's the AI itself. Maybe the LLMs will simply gain consciousness overnight and start calling all the shots. But why would they be interested in working for you and toward your goals?

Human-readable code

All current AI coding solutions generate human-readable code. If machines are truly taking over (remember, any time now), why are we still optimizing for developer ergonomics? Why generate a terabyte of React and ship a virtual DOM to the browser when the same result could be achieved without a separate library? Why produce JavaScript at all when you could output WebAssembly?

If we were truly headed toward a world where machines write all the software, you'd expect the focus to shift from developer experience toward raw performance optimization, or for code to disappear entirely into low-code/no-code platforms. We're nowhere near the paradigm shift people are predicting.

Faster horses

Another thing I find puzzling: what third-party AI-powered applications have genuinely eliminated tasks? Most AI tools are additive and enable you to create more, faster. Sure, you can create meeting minutes automatically or let the AI assistant play calendar Tetris on your behalf, but you still have to sit in those meetings. And file that TPS report.

The end of the code artisan

Before someone accuses me of being an AI denialist, let me say that one group of developers should be worried: the code artisans. You probably know the type - the person who throws a tantrum if they have to touch someone else's code, gets anxious at the sight of a for-loop, constantly invents more elaborate ways to build HTML forms for the company's internal ERP, and switches jobs every year to work with "modern technologies." The future looks rough for those whose job satisfaction comes from the code itself rather than the business value it produces.

Final thoughts

Let's say the utopia becomes real. One of the AI companies introduces their new Besserwisser 13.37 model. Paired with the Kludge Kode agent, it takes the world by storm, replacing regular office jobs left and right. If you are the company building this model and agent, where would you set the price? At $20, $200 or even $2000 / mo (+tax), this would be the deal of the century. No investor would settle for that.

Like I said, maybe I just lack the imagination. Maybe it is the next model that will finally show how it's done!