As Software Developers it feels like our days are numbered.
But despite all this I really don’t think they are. In fact, I believe Software developers will be flourishing in 5 years and here’s why.
The Underlying Models Are Not Improving
The innovations happening around tools in the software AI space this year have been immense. Cursor has evolved dramatically with new features shipping regularly — the most recent being a “Dreamweaver like UI Visual editor” (I’m not going to get into that now — but seriously wtf?!).
Press enter or click to view image in full size
Claude code continues to evolve too, every tool we use now has some MCP available (which hardly anyone uses) and other UIs we need to interact with all seem to have some kind of AI integration nowadays — Jira, Github, Gitlab etc.
Like I said, it’s been quite a year.
But… these are all innovations in tooling. The underlying AI models haven’t really got any better.
There was the completely underwhelming ChatGPT 5 release. More recently Anthropic have released the Opus 4.5 which hasn’t brought any significant improvements. They’re still really useful in the right context but they aren’t making any meaningful difference from what we had before.
We’re at the point now where we’re tweaking and tuning for optimisation rather than bringing any kind of major innovations to the underlying models.
Up until now companies like OpenAI and Anthropic have been relying on throwing massive amounts of data at the models and seeing improvements happen that way. And it worked. But it’s not working anymore regardless of how much data they have to train on.
The models just aren’t going to get much better.
This then begs the question — Are the current set of tools and AI models good enough to replace the majority of Software engineers? Not a chance!
Software engineers make and MAINTAIN software for all sorts of things. Aeroplanes, power stations, military weapons, financial markets, banks, utility services, the internet — the list goes on and on. Pretty much whatever you see in your home will have interacted with some software during it’s lifecycle.
I mean as I write this it just sounds like lunacy. Absolute madness that some people think this is even a possibility. The amount of complexity and responsibility in these systems is mind blowing.
What About Innovation?
I don’t find this spoken about too much in the AI world. Sure generative AI can make you more productive, especially if you know what you are doing in the first place. It’s a force multiplier.
But what about innovation? I’m not seeing any.
Why not? Because generative AI fundamentally can’t do innovation. It repeats patterns it’s seen before and it remarkable at it but it can’t truly innovate.
Innovation is critical for software evolution. Look at where we started with computers back in the days helping Alan Turing crack the enigma machine compared to where we are now. That happened thanks to many small and large steps of innovation. All this innovation emanating from human minds.
Before you start screaming though that “AI can innovate!!” let me address that. AI can feign innovation but I would argue it’s not “true innovation”. What do I mean by this?
AI could get lucky with innovation. It could stumble across novel ways to combine ideas that could lead to some kind of innovation. It could be used to spot patterns that humans have missed in the past. But there would be limits to how far it could take this.
AI can’t pursue a goal over a long period. It can’t spend years on a problem “chipping” away at it, mulling it over in it’s mind, looking at it from 100 different angles until it suddenly sees that leap that no one has spotted before. This is where true innovation comes from.
Who’s going to create the next Linux? Where will the next great leap forward happen in gaming like it did when John Carmack and John Romero created Doom? I’m pretty sure it won’t come from AI.
Software Is More Than Just Writing Code
This year, with all the AI present in our workspace, I have realised one thing over and over again. Writing the code is just one small part of what makes up the skillset of a Software developer. It’s an important and crucial part but still only part.
To build great software you need a whole bunch of other skills.
- Soft skills — You need to be able to talk to people, ask the stakeholders the right questions, know when and how to say “No”, have a high level of “Technical communication” to explain your code at the appropriate level depending on with whom you are communicating with. Human interaction is a critical part of making software.
- Plan and Prioritise — When working on any meaningful piece of software there are usually quite a few people involved. Gathering enough information from all these people about the current situation of the software, what’s been asked to do, what needs to be done and then making sure whoever will code the thing is crystal clear on what they need to do is a lot and it isn’t easy. LLMs can’t do this currently. They can aid in many of these steps but coordinating and piecing it altogether is a long way off.
- Sort the signal from the noise — This happened to me recently with a situation we had in a prod app. It had become really slow suddenly. My boss used AI to diagnose the issue and it found an issue. The problem was it wasn’t “THE” issue that was causing the slowness. I knew this because I had spent the last month working on the app and I could see from my experience that this issue wasn’t “THE” issue. My context window was much larger than the LLMs. Humans eventually fount the root cause and it was resolved relatively quickly.
- Domain knowledge — This is something that humans are really good at. Building up knowledge over long periods of time that can then be used to make informed and well judged decisions. When you work on something for 5 years you build up a whole bunch of useful knowledge all without trying to hard. This could be a frontend developer who builds up an intuition about good UX/UI or a games dev who has a real sense for what makes good gameplay.
All these skills are not easily replaced. I would argue in fact that current versions of AI are not designed to replace these skills. The interesting thing is we’re in a place right now where it’s becoming more and more evident what the “special” skills are that humans bring to their jobs.
Conclusion
Even with everything going on today with AI I still think Software developers will have big parts to play in the future of software. What I have listed above are major hurdles to the plans of the big players in the AI space.
There are of course some caveats to humans sticking around in software but it’s no different to any major technological change that’s happened before. That is the skill to adapt.
Humans, like they have done for all their history, will have to evolve. As Charles Darwin pointed out “It is not the strongest of the species that survives, nor the most intelligent; it is the one most adaptable to change” and indeed that looks like the case again.
You will need to learn to work with generative AI, in the right hands it is a true force multiplier. But you need to learn its limits and where and when to use it.
I would still argue humans need to learn their craft. More than ever for the ones that know their craft deeply will be more and more valuable as the future unfolds.
But let’s put the hype about AI aside and appreciate the human mind and everything it is capable of because it is a far more extraordinary thing than ChatGPT or Claude code will ever be.
