26 June 2026

Putting AI in its place

ClaudeOpenAIAI

I have a confession - I built a website 😊. Furthermore, I built it with my own code and got AI to review it, not the other way around. Here is something else thoroughly strange... I used CSS 😱. Not SCSS, PostCSS, CSS modules. Because it's an Astro site, I just included the CSS in page and component files.

It seems positively weird to be building a website right in the middle of the "chatbots will replace user interfaces" rhetoric. However, if we don't build things ourselves with the tools we have then we lose the "craft" of development. A stripped back project with minimal dependencies was the ideal opportunity to test my skills outside of the AI bubble we inhabit.

Why do I tell you all this? I say it, not to blow one's own trumpet (there's not much to trumpet anyway), but to let people know it's okay to write your own code. Given the current "vibe code everything" movement, writing code and being an actual programmer is a revolutionary act.

Revolutionary, but also 100% necessary. If you're outsourcing your brain and thinking to AI then you're diminishing the power of your brain. How exactly? By reducing your ability to think, reason and problem solve.

Consider for a moment the analogy of translati ng a text into say, Icelandic. You pass it through your favourite LLM and get a response. You yourself don't speak Icelandic, so how do you know the intended meaning is mapped over from English as accurately as possible? There are all kinds of nuances we individually bring to the English language, so the AI version might be strictly correct, but it won't be exactly what you were trying to communicate.

The same is true of development. You outsource everything to AI and get some code in response. Now, if you then go on to study it, learn what the code is trying to do and contrast it with accepted patterns, then bravo, you used AI to get better. If however you just accepted everything verbatim, then you run the double of risk of (a) not learning anything and (b) committing bad code. And believe me, LLMs can and often do spit out some bad code.

Forging a better relationship with AI

I do find it amusing to watch the AI hype cycle we've been through in such a short period of time. Firstly, AI was going to be so good that we'd all be losing our jobs. Then, we were all moving from VS Code to Cursor, Antigravity or Windsurf (I've circled back to VS Code as many have).

Now have very powerful agentic coding platforms drawing from incredibly advanced LLMs. However, as the era of unfettered LLM use is gasps its final breath, we're having to be very judicious with how we implement AI into our workflows. Tokens are not cheap anymore. AI companies grafted all that data from us when we were giddy writing everything with AI. Now they have that data the economic reality is it was largely a bubble all along.

Circling back to writing code by hand. I was struck by how freeing this was. Sure, it's only a one page website for my wife's craft business operation, but I had to go back and revisit how to use flexbox correctly, what clamp() is and why it's great for targeted typography sizing. I used color-mix function, which is something I had not used before. This led me down a rabbit hole of what the heck oklabs and oklch is and how that transforms the range of colours available.

After I mocked everything up, I did use Codex as a sounding board and sense checker. The refinement done at this stage, in collaboration with Codex, enabled me to learn even more and speed up the production time of the project.

Will I use this very hands on approach all the time? No, but at the end of the day we need devs who can dev without AI in order to make sure AI generated code is actually good and meets organisational standards.

Also, I feel if you want to be fulfilled as a developer, be able to get things done and stay mentally sharp, try not to just outsource everything to the machines. If you do this, then you make yourself vulnerable to job displacement. Always be learning!

Back to blog