Every time I tried to learn how AI agents actually work, I'd get two
paragraphs in and hit a wall. POMDP. Policy. Stochastic. Dropped into the
text like I was supposed to already know them. I'd close the tab and tell
myself I'd come back later. I rarely did.
So a few weeks ago I stopped collecting tabs and started writing the
explanation I wished existed. A small wiki, in plain language, where each
topic starts easy and the heavy stuff (the math, the edge cases, the
notation) sits one click away. Nothing thrown at you cold.
I used a Claude Code workflow to research and draft the modules, then went
through everything by hand to cut jargon, fix what didn't hold up, and keep
the terms consistent. That review step turned out to be where most of the
actual learning happened.
Two things caught me off guard.
The first: how little the vocabulary is standardized. I assumed words like
"agent", "tool use", and "memory" had settled definitions. They don't.
Different sources use the same word to mean noticeably different things, and
a lot of the confusion beginners feel isn't them being slow. It's the field
not agreeing with itself yet.
The second is older than AI, but it landed hard here. Explaining something
simply is brutal at exposing what you don't really understand. More than once
I started a section I thought I knew, got to the part where I had to say it
plainly, and couldn't. Those were the sections I learned the most from.
The wiki is a static site (Astro + GitHub Pages), bilingual EN/ES, free, no
signup. Right now it covers the model underneath agents, the loop, memory,
retrieval, tools/MCP, multi-agent setups, evaluation, and security. It grows
as I learn, so some parts are solid and some are still thin.
If you read any of it, tell me what's unclear or what I got wrong. The wrong
parts are the ones I most need pointed out.