Honestly, this post is something of a rant about spending today updating Antsand. I admit I was a bit too ambitious with the scope, and it took longer than planned.

Because Antsand is a quite sophisticated federated deployment engine, the AI I was using (Claude Opus 4.5) constantly broke existing styles every time it added new features or templates. It made me question my whole approach: Do I need better specs? Clearer prompts? Is my design too complex, or was I asking too much at once?

Eventually, after significant "human in the loop" testing and personally diving in to correct bugs, it got sorted. The key takeaway is the importance of feeding the AI detailed, incremental steps to prevent regressions. While it still offers a massive speed boost compared to days or weeks of manual coding, human intervention is still necessary, and it certainly helps that I know my codebase cold.

By the end of the day, I was just happy to be back to my original checkpoint where everything is normal again, including this site.