About
fathomforge is one person.
I'm Carlos M. I'm a security engineer, and fathomforge is the name I publish tools and writing under.
My interest is narrow and specific: systems that take actions on their own, and what it costs when they take the wrong ones. Not the science-fiction version of that problem — the mundane version, where a retry loop runs for two days because nothing in the stack was designed to notice.
I run five autonomous agents in production, and most of what I write here started as an incident in my own logs. One agent received a bare "Hello", invented a task involving two people who don't exist, went looking for somewhere to post it, and made 148 events off that single message before I restarted it. Another hallucinated a URL, got a 404, and retried it a few hundred times per image — quietly, for days.
The tools came afterwards, because the alternative on offer was a kill switch, and a kill switch you're afraid to arm is just decoration.
How I work
I learn in public and I show the arithmetic. If something here is wrong, I would genuinely like to know — the writing gets corrected rather than quietly deleted.
Everything I build is open source and free. I don't run analytics on this site, there are no cookies, and no third party is contacted when you load a page. That's not a marketing position; it's the same standard I hold the tools to, and it would be strange to apply it to a plugin and not to my own website.
How these posts are written
The incidents, numbers and arguments here are mine. They come out of my own logs, my own debugging, and my own wrong turns. The prose is drafted with an AI assistant working from my notes, and I read, edit and verify every claim before anything publishes — if a number appears in a post, I've checked it against the source.
I'd rather tell you that than have you work it out later. The part that carries any weight is the reasoning and the evidence, and both of those are mine.