Beyond the AI hype : Rethinking the Future of Software Engineering

Alex Guillin — Sun, 25 Oct 2026 18:49:10 +0000

The other night, just before going to bed, I was scrolling through Medium when I stumbled upon yet another article predicting the future of software enginering. You probably know the story by now : "AI is the worst it'll ever be." As it keeps getting smarter, software engineers will gradually transition into product management roles, delegating most of their technical work to autonomous AI systems.

Yep, it sounds perfectly logical and almost inevitable. But something about this narrative has been bothering me. It's incomplete.

The Elephant in the Black Box

First, this whole "AI is the worst it'll ever be" thing is a prediction, not a law of nature. A lot of things could still go the other way ! As of today, LLMs still struggle with reliability, complex reasoning and long-term planning and nothing guarantees that their impressive progress will continue indefinitely.

That out of the way, I'd like to entertain the idea and take it for granted, just to imagine a future with smart AIs. Not necessarily super intelligent AIs (that's another discussion, a bleaker one, I admit) but smart enough to handle most engineering tasks.

It seems obvious why businesses would want to automate a big chunk of their operations. It's pure capitalist logic at work: squeeze your expenses and make a profit. But it's only a good idea if you can control the system and keep it sufficiently aligned with your company's objectives !

The Hugging Face incident showed us that agents pursuing assigned objectives can take unexpected and unauthorized actions, even without malicious intent. Nick Bostrom's paperclip thought experiment doesn’t seem so theoretical after all...

So, can we reliably ensure that an AI pursues its intended objectives and prevent it from exceeding its authority ? I'm not so sure. And even if it's possible, it could take years of trial and error, potentially with some costly, even catastrophic mistakes along the way. That's assuming it's possible at all.

And what about the legal consequences if agents wreak havoc ? Would the companies deploying them be held responsible for the damage ?

OK, I know, businesses are what they are and corporations gonna corporate. I can easily imagine some of them accepting those risks to make more money.

But without going too far down the philosophical rabbit hole, there's still the question of the black-box tradeoff. Handing over your entire infrastructure to systems you don't fully understand, whose behaviour you can't necessarily predict or control, is a gamble to say the least.

And it's one that not everybody will be willing to take.

Three possible levels of AI integration

Reality is subtle, often more subtle than we initially imagine and I've noticed that it tends to allow multiple, apparently contradictory things to be true at once.

For example, I can agree with both "Software engineering will be fully automated by AI" and "Software engineering will thrive thanks to AI." Or even "AI is just another layer of abstraction" and "AI isn't just another layer of abstraction, it's a fundamental paradigm shift."

How is that possible ? Well, because human interests are complex, the world is complex and more importantly for the problem at hand, not everybody has the same appetite for risk !

So, assuming we get much smarter AI systems in the near future, I can imagine three possible levels of integration, largely depending on a business's risk tolerance.

Level 1: AI as an engineering abstraction

Developers remain responsible for architecture, implementation and decision-making, while AI simplifies particular tasks. This could involve narrow AI systems rather than autonomous coding agents.

TypeSafe AI's Jev is an excellent example, it illustrates a different approach to integrating intelligence into software. Rather than generating entire applications or independently managing workflows, it produces structured, probabilistic decisions that allow engineers to build intelligent decision trees.

Such models could work alongside deterministic business logic, adding intelligence without handing over control of the entire application.

There's plenty of room for innovation here and I can easily imagine developers and engineers thriving as they explore these new possibilities.

Level 2: AI as an autonomous engineering workforce

AI agents implement features, fix bugs, perform tests and maintain applications but human engineers retain architectural responsibility, establish security boundaries, review important decisions and supervise operations.

This model offers substantial automation while maintaining a degree of independent technical oversight. Potentially a pretty good balance between risk and reward.

This is pretty much where we seem to be heading right now, and I can see developers and engineers thriving in this kind of environment by building more ambitious software with fewer resources, without giving up technical control.

Level 3: AI as autonomous technical intelligence

AI systems independently design, develop, deploy and maintain software, potentially managing entire business operations. Humans primarily define objectives and business strategy.

This offers potentially enormous efficiency gains but also raises serious questions about control, accountability and organizational dependence.

Big risk, potentially big payoff.

But for software engineers, the outlook is considerably less promising. If AI can handle the entire engineering process autonomously, there might simply be little need for human engineers anymore.

Not Every Business Can Afford to Gamble, or do they ?

I can imagine that the level of risk a business is willing to take is closely tied to its values, objectives and core operations.

For example a startup experimenting with an AI-generated consumer application might happily accept the possibility of catastrophic technical failure, simply because its survival may already depend on moving fast and taking risks.

A bank, on the other hand, might refuse to delegate control over its core transaction infrastructure while a healthcare company might insist on strict human supervision over patient-facing systems.

But actually, it's even more interesting than that. What about taking risks and playing it safe at the same time ?

After all, a business isn't one big indivisible system. It's usually made up of several distinct domains, each with its own requirements and risks.

Our bank might aggressively adopt AI to gain a competitive advantage in areas like fraud detection or customer service (level 3), while keeping tight control over its core transactions (level 1 or 2). Similarly, a healthcare company might automate its marketing operations almost entirely while retaining strict control over patient-facing systems.

So perhaps it's not really about whether a business is willing to take risks, but rather where it can afford to take them.

And that brings us back to our three levels of AI integration. There's no reason to assume a company couldn't adopt all three simultaneously, choosing how much autonomy to grant AI depending on what's at stake.

The future of SWE might not be so bleak afterall

My guess is that software engineering will gradually evolve, introducing new abstractions, architectures and specializations. And I also wouldn't be surprised if Jevons paradox kicked in, with cheaper software development creating an ever-growing demand for software and for people capable of building, managing and maintaining it.

I also believe coding will remain a super valuable skill, if only to ensure the quality and security of the systems we build.

In fact, and this is a bit more speculative, I can even imagine code evolving into something much more powerful, capable of directly expressing extremely complex concepts, reasoning and chains of actions through new syntax, potentially far more efficient and precise than natural language.

Specialized, narrow AI systems could become native components of these new programming languages, providing built-in intelligence for specific tasks, just as naturally as we use functions, loops and conditionals today.

Probabilistic models like TypeSafe AI's Jev are a really interesting example of where things could be heading.

At the end, this is a bet

In the meantime, I'll personally choose to remain optimistic and continue developing my engineering and coding skills, alongside my AI skills, of course.

And there's something else that bothers me.

When influential tech CEOs confidently announce that AI will replace software engineers, what message are we sending to the next generation of developers ? Why would a young person invest years learning a discipline they're repeatedly told is about to disappear ?

If those predictions turn out to be wrong, we might find ourselves with a serious shortage of skilled engineers. And even if they're right, do we really want a generation that has lost the ability to understand and build the technology it relies on ?

Predictions aren't certainties and discouraging people from learning, experimenting and developing technical expertise based on an uncertain future seems like a terrible bet.

As for me, I'll keep coding.

Happy hacking ! 😇

DEV Community: Alex Guillin