Six months ago I started maintaining this site as a side hustle. Two to three hours a week, mostly nights and weekends, including all the hobbyist content around it. Not a lot of time. And yet somewhere in that process I started noticing something I didn't expect: I was actually learning software engineering.
Not in a "here are the fundamentals of computer science" kind of way. More like the way you learn things when you're on the job and something breaks and you have to figure out why. Except compressed. Weirdly, uncomfortably compressed into what should have been a pretty shallow experience of just poking at an AI until a website works.
That tension is worth pulling on.
The Sharp Edges Show Up Fast
When you're actually maintaining an app, even a simple one, you start finding bugs. Some are obvious. Some are hiding. And what you begin to realize is that every app has these sharp edges, places where things can go wrong, where attack surfaces open up, where assumptions you made at the start turn out to be wrong. As a working software engineer, you'd know this intuitively because you've spent years pattern-matching against exactly these situations.
As a vibe coder, you find out the same way. You just find out faster.
I've been talking to my agent in plain language: "Hey, can you check to make sure this is actually doing what I think it's doing?" And what happens is the agent goes and builds a test harness, runs a scan, looks for the thing I was vaguely worried about, the same discipline that turns into an actual post here most weeks. The discipline is baked in. I don't have to know the name of the methodology. I just have to have enough awareness to ask the question.
That's a real shift. The abstraction isn't just "natural language instead of code." It's natural language instead of years of accumulated disciplinary knowledge about how to check your work.
Watching the Agent Teaches You How to Think
Here's the part that surprised me most. Because I can see what the agent is doing, step by step, I'm actually learning. I'm learning how to decompose a problem into smaller chunks. I'm learning which kinds of tasks the agent handles well and which ones it fumbles. I'm developing intuitions about where things might go sideways before I ask it to look.
Those intuitions feel like software engineering instincts. Not fully formed ones. But genuine ones. The kind that would have taken me years to build through traditional on-the-job experience.
It's also teaching me to think across disciplines simultaneously. Security, testing, architecture, code quality: these used to be distinct specializations that people spent careers developing. Now I'm getting exposure to all of them at once because my agent is navigating all of them at once, and I'm watching it do it.
The AI Have-Nots Gap Is Really a Curiosity Gap
I've started thinking about who's getting left behind as AI accelerates, and I don't think the dividing line is what most people assume. It's not technical background. It's not access to tools. It's curiosity. It's the same reframing behind a finance intern spending her summer vibe coding automation instead of shadowing someone: the on-ramp was never the CS degree, it was the willingness to jump in.
If you're willing to jump in, to tinker, to accept that you're going to hit bugs and weird edge cases and moments where you genuinely don't know what just happened, then AI gives you this hyper-abbreviated version of on-the-job learning. You pick up skills fast. You develop judgment. You start building things that would have been out of reach.
If you're not curious, if you're waiting for AI to feel safe and obvious and simple before you engage with it, the technology is moving past you. Not because you're incapable. Because you're not in motion.
I think about this a lot when people ask me whether AI is going to displace jobs. My honest answer is: not the way most people fear. What I think we're actually entering is a period of massive expansion in how much software gets built, and who builds it, and what kinds of problems get solved. The unlock isn't replacing engineers. It's making it possible for someone like me, spending two hours a week on a side project, to develop real engineering intuitions through practice.
That's a different story than the one about displacement. It's a story about democratization. Skills that used to require a computer science degree or years of mentorship are becoming accessible through curiosity and a willingness to mess around and pay attention to what happens.
I don't think that means the craft of software engineering stops mattering. If anything, watching my agent work has made me more interested in the fundamentals, not less. But the on-ramp has changed dramatically. You just need to jump in.
Are you learning things from your AI tools that you didn't expect to learn?
Top comments (0)