AI froze, and I coded: why developer experience matters

Whether you like it or not, AI is here. And you’ve probably noticed: it seems to pop up in every app and subscription these days.

And in the coding space, there are tools to help developers write code. And the scarier one, even getting non-coders to write (and ship) code (but that’s another story).

Within the dev community, there are two clear camps (and no doubt lots of little parties in between too). There are those who rely on AI to streamline their work, and then there are those who love to write code themselves. And of course, those in between. That’s me: I love to code, but also see the value in AI and how it can help me day-to-day.

At my core, I’m a developer. I do love writing code. Planning. Researching. Learning. Teaching. It’s all so awesome.

But also, with AI extending beyond just a tool that we can use but a service to integrate our own client projects with, I feel it’s important to not be left behind too, so having an understanding of how AI works, both as a tool to help my, but also a tool to integrate with, is so valuable.

Recently I have been using different AI products: from some open source models to Codex to Antigravity to Claude Code. And they’ve been helpful at times… and, well, also absolute failures too.


The quality of the code that gets produced seems to ride some really wild waves… one day it can efficient, smart, intelligent, and incredibly useful, but then other days it can produce real slop.

And then there are days like last Thursday.

When I’m using AI to help with some code, I always write a plan first. Sometimes I get AI to help with writing the plan, other times I write it myself, but ultimately it’s a text-based md file. I’ve done it for weeks. It’s been great.

But Thursday was not great.

Claude Code would read my plan, and then think. 20 minutes later, Claude was still thinking.

And then boom, my 5 hour window is completely used up, and Claude Code has output zero lines of code.

Right, I thought, 4 hours and 40 minutes until I can try again.

And as a side note, I tried again 5 hours later, and boom, 100% used within minutes, and zero lines of code. That was my whole workday’s allocation of time just gone with nothing to show for it.

But here’s the catch: when AI decides to think and output nothing, and is clearly having a moment, what do you do?

As a developer who loves to write code, it’s obvious: you write code.

But what if you’re someone who relies on AI to do your job. Not just as an assistant to work with, but truly depends on it. What do you do with your day?

I had a few features I wanted to work on, and wanted AI to help with some of that tedious scaffolding work, but I pivoted my plans for the day and worked on something else. First for 4 hours and 40 minutes, and then for the rest of the day.

The important thing though: I got work done.

Sure, it wasn’t what I had planned for the day, but work was done.

With decades of web development experience behind me, I was able to go back to what I am trained for. What I have educated myself for. What I have done for years and years. And that is write code.

I do enjoy working with AI alongside me - it can be incredibly useful - but also Thursday was a friendly reminder that when AI doesn’t want to play ball, remember that you’re a developer.


So go and develop.