Your first AI agent lives in a folder
A step-by-step guide, with no coding, from the ChatGPT chat to your own agent: a folder with instructions (AGENTS.md), memory and skills. For Claude, ChatGPT, Antigravity and Goose.
Experiments, niche solutions and findings. Every post is a plain page: no script is needed to read it, and each one looks the way its story asks.
A step-by-step guide, with no coding, from the ChatGPT chat to your own agent: a folder with instructions (AGENTS.md), memory and skills. For Claude, ChatGPT, Antigravity and Goose.
Una guía paso a paso, sin programar, para pasar del chat de ChatGPT a un agente propio: una carpeta en tu ordenador con instrucciones (AGENTS.md), memoria y habilidades (skills). Sirve para Claude, ChatGPT, Antigravity y Goose.
A map generator that asks a decision model one question per tile, about 3,000 times per test run for 25 cents, and the evaluation loop in Galtea that took it from a green blur to a village, one measured change at a time.
For years, writing for the web meant writing for a ranking. Now language models read everything that is public, so a plain page about a small problem can become the answer someone gets. That is why this blog exists, and why its posts are plain HTML.