The coworker, not the chatbot
A chatbot waits for your next message. A coworker remembers the last one, knows your tools, and gets the thing done while you're somewhere else. That gap is the whole product.
Ask most people what an AI assistant is for and you'll hear some version of the same thing: you type a question, it types an answer. Useful, sometimes impressive, but it stops the moment you stop typing. It has no memory of yesterday, no access to the tools where your work actually lives, and no way to do anything while you're in a meeting. It's a very good intern who forgets your name every morning.
A coworker is different, and the difference isn't intelligence. It's continuity.
Three things a coworker has that a chatbot doesn't
Memory that survives the conversation. A coworker remembers the decision you made last week, the way you like your updates written, the client who always needs a second nudge. Not because you re-explained it — because it was there.
Hands on your tools. The work isn't in the chat window. It's in your inbox, your CRM, your repo, your ad accounts. A coworker that can read and act on those is doing the job; one that can only describe what you should do is writing you homework.
The ability to work while you're gone. The most valuable thing a colleague does is the thing you never see: the follow-up sent, the report pulled, the draft waiting when you get back. Presence isn't a chat that's always open. It's work that happens whether or not you're watching.
Why this is hard to fake
You can bolt memory onto a chatbot and call it a coworker. People have. It doesn't hold, because the three things above aren't features you add — they're a posture you start from. A system built to answer treats your tools as something to talk about. A system built to act treats them as where it lives.
That changes everything downstream: how it's allowed to make mistakes, how it reports back, how much you trust it with the next task. A chatbot is judged on the quality of its sentences. A coworker is judged on whether the thing got done.
The test isn't "was the answer good." It's "did I have to do it myself afterward."
What we're building toward
Sauna is the attempt to build the second thing on purpose — an AI coworker that remembers, has hands on the tools you already use, and carries work forward when you've moved on. Not a smarter chatbot. A different shape entirely.
We'll use this blog to show the work: what we ship, what we learn, where it breaks, and the growth experiments behind it — the honest version, with the numbers that actually ran. If that's the kind of thing you want to read, you're in the right place.