Existential Joyride #

The surge of AI (LLMs) in software engineering - especially web development - has felt like both an existential crisis and an absolute rush. Depending on the day, AI is either replacing all of us, enhancing all of us, or it is a complete gimmick. I imagine how jarring that feels depends a lot on where you are in your career. I feel lucky to be far enough in to have some context, while still being new enough to feel open to change. But I keep coming back to the idea that LLMs are "democratizing" code or the industry as a whole.

I have always felt like software engineering has a relatively low barrier to entry. My own path was unconventional, yet it still led me to great opportunities in tech. I've enjoyed working with peers from a variety of backgrounds. So now I find myself wondering, for such an open career with no show-stopping certifications or similar restrictions, does software engineering really need to be more accessible? Is this an optimization or is it marketing verbiage by vendors selling tokens?

Challenges Ahead #

Whether or not the barrier to entry exists, has existed, or should exist, I have two main concerns about the framing of the rapid LLM-ification of software engineering, especially for those seeking to get started in this industry.

Education #

I don't mean in the formal sense like college education, but rather in the what-programming-teaches-you sense - especially when it comes to critical thinking, organization, problem solving, and so on.

We still interview on data structures and algorithms not because these are unsolved problems, but because they display critical thinking skills and the ability to break a large problem into small, logical steps. It's like practicing scales in order to learn music. What does AI-first development and outsourced reasoning as a whole replace that mental training and exercise with?

Responsibility #

It's one thing to vibe code your pet project for friends and family or build an internal tool at work. It's something else entirely to ship such a project out to the general public in exchange for their hard earned money and/or sensitive data. If something goes wrong, is there enough human context to even solve the problem, or is that just another prompt for the LLM to solve? I wonder if we'll see developers pointing to the LLM agent as being responsible. As far as I know, there's no legal answer here yet and I imagine developers and AI vendors feel pretty differently about it.

And yet... #

I do genuinely enjoy shipping products, features, and tools that are helpful for people. That has always been the sort of overarching goal with what I do and I find that LLMs massively improve my abilities. I think that is how a lot of people feel.

The tools are changing, and maybe that's just technology, but something about this feels different.