Somewhere between the cold brew and the 2am debugging sessions, a useful company was born.

Agency, a startup that grew out of San Francisco's increasingly crowded AI hackathon circuit, is tackling a problem that anyone who's played with AI agents has quietly worried about: you tell the bot to do something, it goes off and does stuff, and you have basically no idea what happened in between.

That's not a small problem. As AI agents get handed more responsibility — booking things, writing things, executing things on your behalf — the black box gets a lot more consequential. Agency's pitch is simple: transparency. See what your AI agents are actually doing, step by step, in real time.

It's the kind of idea that makes you wonder why it wasn't already table stakes. Then again, most of the AI industry has been moving at "ship it and figure out accountability later" speed, so here we are.

From a liberty-minded perspective, this is exactly the kind of tooling that should exist. Giving people actual visibility into automated systems acting on their behalf isn't a nice-to-have — it's a prerequisite for informed consent. If your AI agent is firing off emails, spending money, or making decisions in your name, you deserve a receipt.

The fact that this came out of the hackathon scene rather than a boardroom is also worth noting. San Francisco's tech culture gets a lot of grief — some of it earned — but the grassroots-to-startup pipeline still works when the problem is real and the builders are hungry.

Agency is early. But the instinct is right: in a world filling up fast with autonomous software acting on our behalf, tools that put humans back in the loop aren't just good product design. They're good sense.