If you've spent any time on Substack or Medium over the last twelve months, you'd be forgiven for thinking that the entire software development profession is about eighteen months away from extinction. The prophecies of the AIpocalypse are coming thick and fast, each one more hyperbolic than the last. AI is coming for your job - your career is over - best retrain as a tree surgeon while you still can.
Most of it is alarmist nonsense, of course, peddled by individuals who are deeply invested in frontier labs and who stand to gain significantly from the narrative that artificial intelligence is about to consume the world as we know it.
Sam Altman, CEO of OpenAI, says there's a "high probability that AI will replace coding jobs in a gradual but accelerating manner". And he's not the only one saying it. Dario Amodei, CEO of Anthropic, told The Economist at Davos that AI models could do "most, maybe all" of what software engineers currently do within six to twelve months. Mark Zuckerberg, Marc Benioff and a surge of other big tech CEOs are all repeating the same prophecy.
But there's a pattern here. Every one of these predictions comes from somebody running or invested in an AI company. They have a direct financial interest in convincing you - and more importantly, potential investors - that AI is about to replace the most expensive line item on every tech company's balance sheet. I'm not saying they're full of shit, but I'd certainly take their timeline predictions with a fistful of salt. I don't believe AI will be the end of software engineering. Yes, I think it will change it forever - perhaps beyond recognition - but the discipline itself will still be here and the people who can do it well will be more valuable than ever.
Personally, I find it incredibly exciting. Cast your mind back to the arrival of frameworks like Rails and Node (yes, I'm that old). Before Rails, building a web application meant writing everything from scratch - database abstraction, routing, authentication, session management, the whole shebang. Rails took all of that grind and packaged it up so that developers could focus on building what actually mattered - the product. Node did something similar for real-time applications a few years later. Neither framework replaced developers. They simply freed them up to create more, faster, and with less of the grunt work.
AI - especially coding models and agents - is just the next generation of that same idea, only far more powerful. The difference is that it won't be constrained by a bunch of opinions baked into a specific framework. It won't enforce a particular architecture or file structure or routing paradigm. It'll be able to follow any system or pattern you like, which means you'll get the speed of a framework without the rigidity.
What this means in practice is that the donkey work - the undifferentiating, unsexy, flat-pack assembly of boilerplate code that currently eats up a disproportionate amount of engineering time - gets handled by AI. That frees up developers to do the bit that attracted them to the discipline in the first place - inventing things. Solving interesting and unusual problems. Designing systems that are genuinely unique, rather than assembling pre-fabricated components into the same boring shapes everyone else is using.
The same goes for design. If you're a product designer who's spent the last five years dragging and dropping Figma components from a design system library, AI might just be the thing that breaks you out of that hell. AI will finally let designers create genuinely distinctive interfaces with the same speed and consistency they've grown accustomed to with component libraries - only without the creative straight-jacket.
The common thread here is that AI takes away the repetitive slog and frees up humans to innovate, create and explore. And it's that creativity - the ingenuity, the empathy, the ability to understand why a problem matters and not just how to solve it - that AI (derivative as it is) will never be able to emulate. Large language models don't think; they mimic human thought processes through statistical osmosis. They're incredibly good at pattern-matching, but they're woefully bad at pattern-making.
The bit that worries me
It's not all sunshine and rainbows. There are two problems with the "AI will handle the boring stuff" narrative that I think are genuinely dangerous if we don't address them.
The first is the junior talent pipeline. Traditionally, it was junior developers who did all the donkey work. That wasn't punishment - it was apprenticeship. You learned your craft by doing the tedious, repetitive foundational stuff until you understood why it works the way it does, at which point you were ready to start making decisions about how to do it differently.
If AI automates all of that entry-level work, then what happens to the juniors? Where do they start? How do they learn? And if they never learn, where does the next generation of senior architects, system designers and technical leaders come from? You can't promote people into roles they've never been trained for, and you can't train people by removing the opportunities to learn.
The second problem is more subtle, but potentially more damaging. AI has learned to code from humans. Every pattern it follows, every best practice it applies, every architectural decision it makes - all of it was learned from human-generated data, much of it from platforms like Stack Overflow. And Stack Overflow appears to be dying.
Monthly question volumes have plunged from peaks of over 200,000 in 2014 to under 50,000 by late 2025, back to the levels the site saw at its launch in 2008. Traffic has dropped roughly 50% since ChatGPT hit the market in November 2022. Developers aren't posting questions any more because they're asking Claude or ChatGPT instead. Which creates a rather uncomfortable paradox.
AI learned to code from human developers sharing knowledge on platforms like Stack Overflow. Developers have now stopped sharing knowledge on Stack Overflow because they're using AI instead. So where does the next generation of AI training data come from? If the well of human-generated knowledge dries up and AI models are increasingly trained on other AI models' output, you end up with what researchers call model collapse - a degenerative feedback loop where each successive generation of model produces output that's slightly less accurate, less diverse and further removed from reality than the last.
A study published in Nature found that this effect is real and measurable: models trained recursively on AI-generated data suffer irreversible defects, losing the ability to represent the full range of human knowledge. It's the AI equivalent of photocopying a photocopy. Each generation gets that little bit blurrier.
The long game
As a product leader, it's tempting to look at AI coding agents and see nothing but cost savings. An assistant that doesn't sleep, doesn't go off sick, doesn't hand in its notice for a 20% pay rise at a competitor. But if we're not careful, we'll optimise ourselves into a corner. I've always viewed software engineering as a fundamentally creative pursuit. It involves taking complex problems and tight constraints and inventing solutions that work within them. That - to me - is no different from design. And you don't nurture creativity by removing every opportunity to practise it.
We need to think beyond the quarterly headcount savings and consider the longer-term health of our industry. That means investing in junior talent even when AI can do their current work faster. It means keeping humans in the loop not just because we should, but because the entire AI training pipeline depends on it. And it means resisting the lazy assumption that just because AI can write code, it can replace the people who understand what that code is for.
In 1960, Nobel laureate Herbert Simon predicted that computers would make the programming profession extinct within a decade. Sixty-six years later, there are more software engineers than at any other point in history. I suspect that in ten years from now, we might all look back at Dario Amodei's Davos prediction and chuckle at it in much the same way.
