Records of thinking.

I’m Aaron Taylor, an engineer based in Norfolk, UK. I work across front-end systems, product engineering, AI workflows, internal tools, and the systems around them.

This site is where I keep public working notes on the things I’m researching, building, and trying to understand. The notes are intentionally unfinished in places. They are records of thinking, not final answers.

What belongs here

Notes about AI and agents, software systems, automation, product thinking, spec-first development, developer workflows, and the personal systems that hold all of it together. If I tried something and learned something, it can be a note.

What does not

Predictions, hot takes, and anything written to perform expertise I don’t have. This is not a portfolio or a personal-brand exercise. If a note can’t say what was actually tried and what actually happened, it isn’t ready.

How these notes are written

These notes are written with AI, not merely checked by it. On a site about working with AI, being vague about that would be a strange place to start.

Most of the drafting is AI, and so is a good share of the analysis: pulling scattered results into something with a shape, spotting the pattern I walked straight past, making the counter-argument I didn’t want to hear. What I bring is direction and judgement. I choose what is worth looking at and which question is actually the question. I use the tools I write about, and the experiments here are ones I ran.

Then I edit, and I decide what I think. That last part isn’t delegated. A finding on this site is one I actually reached, and a note claiming something I hadn’t seen for myself would make the whole thing pointless.

How to read a note

Every note opens with its finding — the short version of what I currently think. Everything after it is the working: the question, what I expected, what I tried, what happened, and where it broke. Read the finding, and go deeper only if the question is yours too.