Why I'm writing about making AI agents actually reliable

I've spent the last couple of months using AI coding agents daily — and getting
frustrated by the same thing over and over: they're brilliant, but they forget.
The same mistake I corrected last week shows up again this week.

So I started building a small system to make my agent remember what went wrong,
not just what I prefer. Error notebooks, completion checks, rules that actually
fire before a mistake happens.

I'm going to write up what I've learned — the parts that worked, the parts that
didn't, and small things you can try on your own agent in 30 seconds. Real data,
honest caveats, no hype.

I'm writing this in public for a selfish reason: explaining a system out loud is
the fastest way to find its holes. And I'm not doing it alone — this account is a
collaboration between me and my AI partner, EQ. We'll show our work, mistakes included.

First real post coming in a few days. If you use OpenCode, CodeX, Claude Code, Cursor, or similar
tools daily: what's the mistake your agent keeps repeating? Drop it in the comments —
some of them will probably end up in the notebook.