"We are fucked" and some other half-assed thoughts on AI and coding

"We are fucked."

Those were the first three words of a blog post I started a few weeks ago. I never finished or published it.

I wrote those three words after a friend of mine, who has 0 coding experience, was able to put together a complete application to manage the schedule of photographers at an event.

He had no clue about how the code behind his app worked, what it did, or why it did what it did. He didn't need to. And the scariest part? The application worked, looked beautiful, and was better than anything I could have produced in double the time.

It left me thinking: This is it. The robots have arrived to take my job.
All those years, learning layer after layer of good architecture, all those programming languages and best practices ... poof ... obsolete.

It made me feel down. Even though I kept cheering for my friend whenever he showed off a new feature of his app, deep down, I felt the end of an era creeping up. It was like a magician standing on stage, explaining to the audience exactly how the tricks worked. My tricks.

And it hit me: soon, everyone would be able to do the tricks that once made me special.

But instead of drowning in sorrow, I decided to open up. I started talking to ex-colleagues, asking them how they feel about all this.

And I started coding, like I had something to prove. Assisted by robots. Side by side. Not rivals, but partners.

That unpublished post was written in fear.
This one is not.

Here's what I've learned since:

Becoming the conductor

I still love programming. I think in code.
But I've come to accept that it's no longer, and will continue to be less, the core of what I do.

By using powerful tools like Claude Code (an AI coding assistant from Anthropic), I've become more of a conductor. I tell the AI what to build based on my experience with tooling, architecture, and libraries. And that’s made me far more valuable to my clients than if I were typing every single character by hand.

By planning together with the AI, reviewing its changes, and quickly iterating on ideas, I've been able to create features that were estimated to be weeks of work in days or less.

Large language models are incredibly good at the boring parts.
Creating something like a context in React, for instance (don’t worry if you don’t know what a React is), is a repetitive task I’ve done more times than I can count. It’s not hard. It’s just tedious.

It might take me 30 minutes to do manually. With an AI coding agent? Probably in under a minute.

That’s 29 minutes I can now spend on things that actually matter—like product thinking, polish, or moving faster toward the next idea.

A faster brush

I love creating. Whether it's coding, photography, or 3D printing a dickosaurus - I love adding things to the world that weren't there before.

But when it comes to creating, there’s something that really matters, and that’s often overlooked: speed. Not because we're in a rush, but because momentum is creative fuel. Ideas are fragile, and the longer they sit unfinished, the more they wither.

By using AI to code (or even vibe-code), I can sketch ideas much more quickly. Not perfect, far from it, but quick. I can describe an idea, like "make a game of Mario where you play as the Goombas", and within minutes, I’ve got something running. And in the case of Mario and the Goombas… well, let’s just say there’s a reason Nintendo didn’t make it themselves.

But the point is: I didn't waste two weeks building it. I spent ten minutes, had a good laugh, tried to pivot it into a 3D game (which was even worse!), and moved on. The ability to quickly validate ideas and discard them is what makes the next good idea possible.

I used to get stuck trying to refine an idea to the point of death. Now I just make three versions and throw two away.

Working with an AI coding agent feels like having an endless sketchpad.
Most of the sketches won’t make it to the fridge—but without them, the masterpiece would never happen.

Slop as superfuel

There has been a lot of writing on AI slop, the sludge of low-quality output of generative AI, flooding the internet. And yeah, I agree: that’s a problem. To be fair, the internet was already in a bad enough state even before AI slop got involved.

But I'd like to make a case for slop too.

Because in the hands of a creator, slop isn't pollution, it's raw material.
It's noise you can sift through, remix, reject, or refine. It lowers the cost and barrier of experimentation to almost zero. You can try out five different directions, see what sticks, and discard the rest without hurting any feelings.

It's creative superfuel.

Slop only becomes a problem when you confuse it with the final product.
Treat it like a sketch, and it becomes a tool, not a threat.

The unresolved

Of course, I don't want to paint an overly rosy picture.

We're still at the very beginning of a revolution.

It's been less than 3 years since ChatGPT was unveiled to the world, and we're now seeing new model releases weekly.

We do not yet know how this shift will affect the world around us. The impact on designers, writers, coders, actors, engineers or anyone whose creativity can now be cloned by a machine is still unfolding. These models consume enormous amounts of energy and water, and are all owned by big tech, requiring infrastructure the size of Manhattan.

There's also a growing divide between those who have access to these tools and know how to use them, and those who don't, even though they might be just as creative and driven.

A co-worker recently raised an interesting question:

We can now all produce more, and save time, but do we get that time as free time in return?

That one stuck with me.

We’re building faster than ever, but are we rushing past the craft? The joy?
Are we buying time we never cash in?

And if this pace continues, will we eventually run out of work? How will this impact us as a society?

These are significant and unresolved issues. We can’t ignore them. And we shouldn’t. However, they should also not paralyze us.

I'm not obsolete, I'm just getting started

I don't have all the answers. But I know this:

AI coding tools haven't made me obsolete. They've made me faster. More daring. More curious. Less afraid to fail.

They gave me back the part of programming I fell in love with: imagining something... and watching it come to life.

I still hold the baton. I still choose the direction. But now, with AI as my co-creator and sketchpad, I can explore faster and build more than I ever could alone.

That unpublished post was written in fear. This one is the start of something.

And to revisit the magic trick analogy: the trick isn’t gone. It’s just no longer a solo act. And honestly, I’m excited, because lowering the barrier of entry to creating digital experiences and delight might just change the world for the better.

As an epilogue, my current AI coding stack (July 2025):

I have a Claude MAX subscription, which I use in conjunction with Claude Code.

Previously, I used Cursor (and Windsurf), but now I exclusively use stock VS Code. I find stock VS Code to be much more modern and faster than the ancient version that Cursor once forked.

I have Claude Code running in a side panel inside VS Code. Additionally, I usually have one or more extra Claude Code instances running in a terminal. Because I hate waiting around while one Claude instance is busy thinking 😀 I use some MCP servers, but nothing too exciting (context7, Figma, Sequential Thinking, GitHub, and Playwright).

I also have GitHub Copilot Pro, which I receive as a gift every month due to the open-source contributions I made at my previous company. I use Copilot primarily for tab completions inside VS Code.

I tried many other tools, such as CodeRabbit, to increase code quality, but I've found them to be of little to no additional value compared to asking the same question to a normal coding agent. "Please review my staged changes" is something every agent can handle quite well. You achieve better code quality improvements by using a strict linter config (ultracite) and good typings, in my opinion.