It takes time to learn how to work with AI

I had drinks last week with a former colleague — a developer I've known for years. He's been in the industry a while, writes good code, and is genuinely skilled at what he does. He's also skeptical of AI.

I spent most of the evening listening (and taking notes). And what I heard sounded really familiar — because it sounded exactly like me a year ago.

The denial sounds the same every time

What I said a year ago is exactly what I heard last week:

It makes too many mistakes.

And:

It can't do what I can do.

And:

I write better code than AI."

And the thing is, they're not entirely wrong. AI does make mistakes. It does require a lot of hand-holding. But the framing — that it sucks and isn't worth the effort — is where they lose me. Because I said the same things. And I was wrong.

What's happening here isn't stupidity or stubbornness. (Though I often feel both stupid and stubborn.) It's a natural human reaction to something that feels threatening. When you've spent years honing a craft — learning to write clean code, debugging complex systems, building things by hand — and someone tells you a machine can do it better and faster than you, the instinct is to push back. To protect what you've built. To insist that your expertise still matters.

And your expertise absolutely still matters. Just not in the way it used to.

I went through this a year ago

I work for a company is building AI tooling for developers building websites. My CEO was pushing AI adoption harder, faster, and earlier than most. That meant I was forced into the struggle before a lot of other developers were.

And I struggled. AI tools were rough in early 2025. Getting an agent to do what I wanted felt like pulling teeth. I spent more time correcting mistakes than I saved by using the tool. And the output was mediocre at best. It was genuinely frustrating, and I understood the temptation to just go back to writing code the old way.

But I stuck with it. Partly because I had to, and partly because I started to see where it was heading. Over months, something clicked. I learned how to frame problems clearly, how to provide the right context, how to work with the tool instead of fighting it. And now? I'm running multiple projects in parallel, hitting a flow state I haven't felt in a decade.

When I look at developers just starting their AI journey today, I see myself twelve months ago. And I know what's coming for them.

Better tools don't shrink the learning curve

Here's the thing I hadn't considered until that conversation last week: the learning curve is real. Yes, the tools are objectively, dramatically better than they were a year ago. Models are smarter. Context windows are bigger. The ecosystem is more mature. And it still takes time to learn how to work with coding agents.

Anyone starting out is still going to fumble. They're still going to fight with the agent. They're still going to feel like it's slower than just writing the code themselves. Because the struggle isn't really about the tool — it's about learning a fundamentally different way of working.

You have to learn to form your thoughts before you prompt. AI can't read your mind — your thoughts are erratic, half-formed, full of assumptions you haven't stated. Working with an agent means learning to externalize your thinking in a way you've never had to before.

You have to learn when to push back and when to let the agent run. You have to build a sense for what it's good at and where it falls apart. You have to develop new instincts — and that takes time, no matter how good the tool is.

It's like learning a new language. The textbook can be better, the teacher can be better, but you still have to put in the hours. There's no shortcut through the awkward phase. The learning curve may be smaller than it was a year ago, but it's still there.

Vulnerability is the price of admission

What I think makes this particularly hard for experienced developers is that it requires vulnerability. You have to be willing to say: Yeah, you know what? I don't know it all. I am a very seasoned veteran of writing code manually. And now I have to learn to do my job in a different way.

That's a big ask. There's real ego wrapped up in the craft of coding. Years of debugging, years of learning frameworks, years of building mental models for how systems work — and now you're being told that the value has shifted. That the thing you spent a decade getting good at isn't the thing that matters most anymore.

I get why people resist that. It hurts to admit it. It hurt me. But hanging on to the old way is going to hurt more in the long run.

The time to start is now

I don't think anyone can define exactly what "indispensable" looks like in this new world. I certainly can't. But I do think it has a lot to do with being ahead — learning to direct AI effectively, evolving how you work, and making yourself someone who can use the tool better than your peers.

The developers who start struggling now are the ones who'll come out ahead. Not because they're smarter, but because they put in the time while others were still in denial.

Here's the blunt version: if you are not willing to figure out how to solve front-end web development problems with AI, you're not going to last in the industry. You're going to have to find another job. And that's a scary thing to say, but I think it's true.

The game has changed. Humans aren't replaceable — but the role that humans play is evolving. We need people, but we probably need people doing different things than they're doing today. The question is whether you evolve your job now, on your terms, or wait until someone evolves it for you.

I'd rather struggle now and come out the other side than pretend the wave isn't coming. And if you're reading this and you're still in that denial phase — I get it. I was there. But the struggle is worth it, and it's time to start.

So go build something. It'll be frustrating, but you'll be happy you made the leap.

And don't forget to have a little fun along the way.