I’m not a web developer. I’m a composer and sound designer — and yes, I can dig into code when I have to. In my Unity projects I can find my way around in C#, and I know how to track down bugs in unfamiliar code (not always, but often enough). But that’s different from understanding the inner workings of a WordPress theme. I don’t know PHP. The structures, conventions, the typical pitfalls of that world — that’s not my territory.
That has changed.
The Problem
My website mathis-nitschke.com has been running the Daze theme by NordWood Themes since 2018 — a visually strong masonry layout that I chose deliberately and wanted to keep. The theme stopped receiving updates in 2019. At first that wasn’t a problem. Then the Instagram integration broke when Instagram overhauled its API in 2020. Then came PHP warnings. Then it got slow.
The usual advice: switch themes. But switching would have meant not just finding a new design, but also migrating content, redoing layouts, adjusting image sizes — a disproportionate amount of work for what was fundamentally a technical problem.
The Attempt
I updated the theme myself using Claude Code (Opus 4.6) — an AI agent by Anthropic that works directly in the file system of a local copy of a website, reading, analyzing, and writing code. I described the problem to it, not the solution.
The process was dialogic. Claude Code started by examining the theme: analyzing the file structure, identifying deprecated functions, tracing the broken Instagram integration, finding the causes of slow load times. Then it proposed a plan. I read it, asked questions, made decisions.
What interested me was less the technical dimension than the nature of the collaboration. I didn’t need to know what extract() does in PHP or why it’s considered unsafe — but I could follow the reasoning and decide whether the intervention made sense. I wasn’t the programmer, but I wasn’t just a client either. Somewhere in between: someone who understands their own project and now had the means to develop it further on a technical level.
What Changed
The result is Daze MN, a fork of the original theme, available on GitHub. The key changes:
PHP compatibility. Nine widget files and several theme functions were using extract(), a function now considered unsafe. All calls were replaced with explicit variable assignments, and deprecated database queries were updated. The theme now requires at least WordPress 6.0.
Instagram Graph API. The old integration relied on an unofficial access point that Instagram shut down in 2020. The new connection uses the official Graph API with automatic token renewal every 50 days. Reels are filtered out — only images appear in the carousel.
Performance. jQuery UI Draggable and the Facebook SDK were being loaded on every page without being used. Gallery scripts are now limited to single post pages. Google Fonts load with display: swap.
600 lines of demo code removed. The original theme contained extensive special cases for demo categories that don’t exist on a real website but were being evaluated on every page load.
The theme looks identical — just more robust, faster, and ready for the future. (If you think it’s still slow, well, it’s WordPress…)
What It Means
I published the fork on GitHub because I suspect other Daze users are in a similar situation: the theme is beautiful, but no longer maintained. The Instagram widgets are broken. PHP warnings are piling up. Switching feels disproportionate.
But there’s something more fundamental going on. In my work as a composer I’ve written that what interests me about technology is not what it replaces, but what it enables. The same applies here. Claude Code didn’t program for me — it enabled me to do something that was previously out of reach. The decisions were mine. The tool was new.
For artists who work with technology, this is a meaningful shift: not because AI can do everything, but because it lowers the threshold at which you stop having to ask experts and start intervening yourself. Updating the theme of my own blog sounds trivial. But it’s an intervention I simply could not have made a year ago.
