I've built a few small things with AI.
One of them made money. Most didn't.
The lesson wasn't “build more.”
It was that building is getting easier, while figuring out what people actually want is still hard.
So this account is where I'll think out loud about AI products, useful workflows, growth, and the problems people keep running into.
I'll share things I try, things that surprise me, what works, what doesn't, and whatever I end up building next.
I don't know what the next product will be yet.
That's part of why I'm here.
I want to get better at noticing real problems before rushing to build solutions for them.
I recently came across a pretty interesting project called Ctx.
It tackles a problem that has always been annoying in software development:
Git is great at telling you what changed, who changed it, and when.
But why a piece of code was changed that way usually lives in commit messages, PRs, docs, comments, or simply in the head of the person who wrote it.
Over time, that context is easy to lose.
AI coding gives us a different way to deal with this.
Tools like Claude Code and Codex leave behind full sessions as they work. Those sessions record what task they were given, what they looked at, what they tried, and what they eventually changed.
Ctx indexes those sessions.
So the next time you come across a piece of code changed by an AI agent, you don't have to guess why it was written that way. You can just pull up the original session.
This is kind of brazen.
Unit 42 recently disclosed a ransomware attack.
The attacker used AI agents across multiple stages, from reconnaissance and credential hunting to lateral movement and persistence.
The report estimates that work that might normally take human operators around two weeks was completed in under 10 hours.
And even more brazen: after the attack, an 80-page security audit was left behind.