Why I only write about what I've shipped

A few years ago I read a widely-shared write-up on system design where the advice was clearly frozen in time — architecture patterns from a job the author had left a decade earlier. Nothing wrong, exactly. Just untouched by a production incident since 2016.

That's the thing I keep coming back to when I write here: the gap between knowing something and having recently been wrong about it in front of real users. Production has a way of correcting you that a blog post never will.

So the rule I hold myself to is simple — if I haven't shipped it, I don't write about it as settled. Half of what I believed about RAG pipelines six months ago doesn't hold anymore, because the patterns that worked in early 2025 mostly don't survive contact with real traffic.

This is also why I'd rather post fewer, more specific things than a steady stream of generic advice. If something I wrote a year ago looks different from how I'd write it today, that's not a redesign — that's me admitting the old version was incomplete.