On this page
I still remember the times in high school during the pre-ChatGPT era where you still spent the majority of your time manually writing code by hand or looking through Stack Overflow to see if somebody else came up with a solution for your problem, before trying to solve it on your own. It was tedious work and immensely slow, compared to today, but you could see step-by-step how your work slowly but surely evolved into something usable.
I love engineering and the process of building something on your own and that's why I started my software engineering apprenticeship back in August 2021. An apprenticeship felt more meaningful to me than going to university: a software company had already hired me, since I already knew a bit about coding, and I could ship my first changes to production fairly quickly.
About a year into the apprenticeship came the release of ChatGPT in November 2022. While it was mostly just useful for quickly looking up things, we all know what happened in the years that followed. LLMs became an integral part of software engineering and significantly changed the workflow of most software engineers nowadays.
Early on, mostly random Twitter users, but sometimes also better-known people, claimed that developers could be jobless within 6 months. Well we're a couple of years into these 6 months and developers still exist, but me being a junior at that time, it really increased my fear of having done that apprenticeship for nothing — maybe it won't affect me now, but in a couple of years. In the end, I just dismissed it, as I still liked my job and the signals weren't big enough for me to see that I couldn't find a job anymore soon. Instead, I utilized LLMs as a tool to increase my own capabilities in building stuff at my job and in my free time and it actually made engineering more fun, as I was able to build things much faster than before.
However, I feel recently, especially with the releases of Claude Fable 5 and GPT-6, that agents have become more capable when it comes to long-running tasks and the output quality has increased a lot. I still strongly think that human oversight is necessary to correctly steer the agents to cover all important aspects of software development like privacy, safety, accessibility and quality, but that required oversight shrinks year by year. The best way I can describe the current state is that projects which used to take 20–30 engineers can now feel doable for 1–2 experienced engineers with AI tools.
What does the software engineering future look like?
Well, I don't know. I've never felt as uncertain about anything as I do about AI. I think there's a good chance that LLMs can become so good in the future that they can run unsupervised. I'm not saying that they should, but at least that they could.
Following capitalism's usual push for efficiency, I think software engineers might continue to become more high level, ultimately becoming something like product engineers with a technical background. But in the case of agents running unsupervised, product managers might also become redundant, because an agent could also make these decisions and do the research.
However, I think an important part that many people miss in this discussion is that software is kind of a collection of opinions, decisions, past experience and actually, creativity. There are various different approaches to building an online shop or an ERP system. There's just no right or wrong for those. It's kind of like the various smartphones out there, some have a stronger focus on gaming, some on a nice user experience and some on privacy. SaaS products still have the advantage of history, so it's more likely that issues you might encounter in your own vibecoded solution are already fixed there, so you're safer there.
Is software engineering becoming more democratized?
Some people argue that AI will be a tool so that it gets easier to solve much bigger problems than before. I can partially agree with that, but it requires money. I feel like if you want to do serious engineering work in your free time, you at least need a $100 plan, which was actually not a requirement before (though you had to spend a lot of time learning and your output was slower). For businesses, that decision might be easy, but for individuals, especially in this economy, it's not always a no-brainer. And then there's also the rapidly increasing cost of hardware, which makes it more expensive for people to afford a device to leverage AI for software engineering. Also, in my experience, mostly the frontier AI labs offer decent models that are somewhat usable for non-technical people to vibecode with, which makes us dependent on them, as these models are closed source or require expensive hardware specs.
AI could give people more power and impact
On a good note, I think AI does help the open-source community become more serious competitors to closed-source projects and more useful. In some sense, that helps people become more independent from major tech firms too and get control over their data back. I think it nowadays shows more clearly whether a product has been purely vibe-slopped to just get money, or whether somebody really cares about the product (though still kinda vibecoded to be clear), though that might not be as clear in the future anymore. So the distinction then could move more to the company and how it operates on a moral basis. In the example of Adobe, people have been mad at Adobe for making it really difficult to get out of a subscription and that there's no option anymore to outright purchase a copy of their video editing software ("You'll own nothing and be happy"). The recent example of ArtCraft, a copy of the Adobe Suite that an experienced engineer reverse-engineered with AI, shows clearly that we now have the possibility to get back ownership. It's still very buggy right now, but I guess a couple of months of bug fixing could actually make it usable. And it's a nice example of what I said earlier: the AI did a huge part of the work, but it still took someone experienced to steer it.
AI could make everything much worse
There's no clear plan for what happens when a lot of people who do huge parts of their daily work digitally can be replaced by AI. We don't know if and when this will happen, but I strongly suggest that we should have a plan for that. If people get laid off and have no income anymore, then doing the stuff I mentioned in the previous section might become impossible, as people won't be able to afford the hardware or subscriptions, and we actually might become more dependent than ever on large companies. And with how fast robotics has been advancing recently, physical jobs might not be safe either. If robots actually take off, there could be very few of them left, or their salaries would drop, since many more people would apply for them. That worries me the most and I think we shouldn't rely on major companies doing "the right" thing. While sometimes it might have worked out, there's no guarantee and in times of such immense uncertainty, we should have a plan.
Aren't you building your own SaaS company??
Yeah, I am. And yes, I see the irony after the ArtCraft example. Some people might argue that SaaS doesn't really have a future (like my ex-boss), but I still believe it does. A vibecoded clone can copy features, but not the testing and care that went into them, and ArtCraft shows that pretty well right now. That's also why I'm considering going partially open source, or at least offering a fairly open API and app system, so customers can adapt my app to their needs instead of rebuilding it themselves. I think we're entering a new generation of SaaS that is much more customizable to the customer. And also because I still love engineering and would like to create a place that aligns with my personal values. I am anxious about whether this will work out or not, but at least I'm trying! And I think that's the most important thing: Just do what you wanna do, nobody knows what will happen in the future, so it doesn't help to hold yourself back just because you don't know whether it will still be valuable soon.
Wrapping things up
I think realistically, it's impossible to estimate the impact of AI on our personal future. I don't know what the right laws look like, and I don't think anyone fully does yet, but we need to start that conversation now and push lawmakers to act, including on global guidelines for the development and usage of AI, before it's too late. As for how we use AI: I strongly think we should treat it as a tool to primarily help us in doing the work, to focus on what we actually wanna do, while still valuing the human craft and quality behind that work, with a good moral compass of overall helping humans rather than simply automating everything away just because we can or to extract more money out of people. Far too many people are doing things they don't enjoy and I think we should change that.
AI should be a tool for us, as humanity, to reach for the stars and not make us irrelevant.
Written by me, proofread with the help of AI.