Introduction
Late 2022, I was trying to prompt AI better so it would code the way I do. Late 2025, I spend my time building the systems that shape its work. This journal traces the path between the two.
What I take away from it fits into one idea: useful AI isn’t about the prompt, it’s about the environment (context, rules, feedback). When the environment is clear, it moves forward; when it’s fuzzy, it spins its wheels. I call it AIX, AI eXperience, by analogy with Developer eXperience: the same tooling logic, but designed so AI can spot its own mistakes, correct them, and keep going. The concept is developed in the Guardrails chapter.
Failures take up as much room here as the rest: almost every chapter contains something I ended up throwing out.
If you want to talk about this, ask a question, or work with me: augustin.bengolea@gmail.com.
How to read this journal
Fastest path: let an AI read it for you. The document contains a prompt that tells it how to extract the essentials.
If you’d rather read it yourself, two paths:
- In order, the 2022 → 2025 thread: chapters build on each other, the turns are visible, the reasoning holds. This is the clearest read.
- Cherry-picking, each chapter stands alone: go straight to whatever interests you. The technical topics each have their own chapter: context architecture, agent orchestration, guardrails.
This is not a polished manifesto: it’s field work, not a showcase.