The second screen

I studied hard to specialize in frontend, and then one day I wrote my first prompt to one of the early agents. A whole string of components materialized in front of me in a matter of seconds. That was the moment I sat there, eyes wide open, staring at the screen. “Oh.”

I remember how hard I had to elbow my way in to convince people I could do frontend. My career didn't start in frontend, but at the opposite end. Back then I didn't much like it, because the tasks were boring; frontend, on the other hand, I found fascinating. I had two monitors: while the prehistoric .NET project compiled on one, the other showed, full screen, Maximilian Schwarzmüller explaining React to me.

In consulting, if you're lucky enough to dig your heels in without getting fired, you can live a second life doing what you love most; otherwise you die, stuck plugging whatever hole needs plugging that week.
I dug my heels in.

Now that world, which I worked so hard to earn my place in, seems to have changed drastically.

These days, on the social feeds I read avidly to keep up, the conversation is no longer about the usual framework holy wars but about which provider to use: Anthropic, OpenAI, Cursor. That's what people talk about now.

At first I thought it was just the hype of the moment, a bubble bound to burst, but then I saw things really change.

Prolific, authoritative figures of the web, the people who taught me to read and write this craft, once deeply specialized in their own stack, suddenly became "AI Engineers" (to name a couple: Matt Pocock, Kent C. Dodds…).

"Frontend Masters", the well-known platform for frontend-focused courses, has now renamed itself "Masters.dev".

“What on earth is going on?”

Maybe Matt and Kent realized there's no point in learning to write good TypeScript and React anymore if there's a ready-made skill for every occasion; and maybe the "Masters.dev" marketing team realized that the commercial label of pure "frontend developer" is fading, and that panicking developers like me would rather take courses on how to use AI than on a new JS framework.

The multipolar world of frontend and backend, which in recent years had already started to blend with the arrival of fullstack frameworks and meta-frameworks, now seems destined to disappear.

After all, frontend hasn't been just HTML, CSS and JavaScript for a long time, not since those very frameworks arrived. So we adapted, and quietly accepted that those things are no longer all we do.

If being fullstack used to be a plus, at this precise moment in history it looks like a hard requirement.

And while technical social media talked about AI nonstop, on LinkedIn some random founder announced that, with four prompts to Claude, they had replaced 13 SaaS products.

It's only natural that statements like these sound threatening if this is your job: it looks as if society has figured out that software development is now within everyone's reach. But today plenty of people are celebrating with a demo in hand, believing they've built a new SaaS.

So it's no surprise what this climate does to the psyche of the developers reading these statements. And so developers pick sides: on one side there's FOMO, on the other, Luddism.

While the former check tech social media at regular intervals, hoping they haven't fallen behind on what to study, the latter, the Luddites, want nothing to do with it.

The arrival of AI isn't always met with enthusiasm. I know a circle of developers who resist this new way of working: "Why use these tools if they're going to get us fired?", they think.

Some developers do this job because, for them, coding is an exercise in style, beautiful and all-consuming. For some it's a real way to escape, like doing a crossword or a sudoku: a logic game that pits them against themselves. If you ask a developer who's hooked on coding to give up what they love doing more than anything else, they're unlikely to welcome the change.

I don't blame either side: there are days when I feel I'm on one side or the other, and days when I'd just like to vanish into some field to farm, live off the land, and not have to worry about every new model reminding me that we're living through a new industrial revolution, and that intelligence is no longer exclusively human.

So far, though, I've talked about developers in general, not frontend developers. Frontend developers who've made it out of the consulting grinder, who spent years bug-fixing hideous interfaces, now find themselves at a disadvantage, or so they see it. Frontend, when it isn't managed properly, has historically been marked by great architectural anarchy: design patterns, and even the most basic concepts of object-oriented or functional programming, take a back seat. So imagine spending ten years bug-fixing in consulting, on a nice contract for some fintech client: after those ten years, what exactly has this developer been trained in? React? jQuery? How useful will those specializations really be in today's projects, built entirely with AI?

But there's something missing both for those celebrating with a demo in hand and for those afraid of being replaced: awareness.

Writing software today is easy; understanding what you've written isn't.

The non-developer is holding code that works (because the only way they have of checking it is seeing that it does what it's supposed to: looks good, ship it!) and has no way of knowing whether, for example, it handles their users' data safely. They don't find out until it's too late and, when they do, they don't know where to start over.

That's the skill you can't generate with a prompt: not the ability to write code, but knowing what to ask for and recognizing when the answer is wrong.

This year the code I write by hand has shrunk drastically. When I started delegating bigger flows to the agent, my focus shifted: I no longer look at how it wrote a function or resolved a promise, but at what can have a real impact. More than once, code that "looked good enough to frame" was hiding a single, microscopic mistake capable of causing gigantic memory leaks in production.

It happened to me recently: the agent had built a plugin following my instructions, and there was a Map outside the plugin's function, bound to grow forever. I spotted it right away because it looked out of place; looking at it more closely in review, I found the bug.

How would a non-technical person have found it?

Along the way I had the misfortune of working on a few failed projects (in consulting, of course) inside dysfunctional teams: I went through hell. I'm sure a great many developers were trained exactly like that, floundering day after day, year after year, in tangled code, until they nearly lost their minds.

And that's why today, when I build new software with AI, I know what not to do, which boundaries not to cross and which strategies to adopt so I don't end up with nightmare code.

All of this is certainly new, and new things bring instability.

But instead of becoming prehistoric like that .NET project, I'll do one thing: sit down, get comfortable and switch on the second screen. Who knows, maybe Maximilian will turn my career around this time too.