
Everyone's picking agents. Nobody's managing sessions.
The strategic question isn't which agent to buy. It's whether your company treats the session as an owned, first-class object—or throws it away.
Every company I talk to is having the same conversation right now. Which agents do we adopt? Which models? Do we standardize on one harness or let people choose? These are real questions. They’re also the wrong ones to obsess over, because they’re about the tool — and the tool isn’t where your company’s value lives.
The value lives in the session.
A session is the actual back-and-forth where work gets done: the context someone loaded in, the way they framed the problem, the corrections they made, the judgment they applied when the agent went sideways, the thing that finally worked. That’s not exhaust. That’s the most concentrated record of how your company actually operates that has ever existed. And almost every company treats it as disposable — a private chat window that closes and is gone.
I’ve seen this movie before. A couple years ago, “observability” wasn’t a category. There were logs, there were metrics, and there was this new idea that if you gave a request a unit of identity — a trace — and followed it across every system it touched, you could suddenly understand things that were previously invisible. The trace wasn’t obvious. Someone had to decide it was the primitive worth building around. Once they did, an entire discipline formed on top of it.
Agent work is at exactly that moment, and the session is the trace. It’s the unit of identity for a piece of human-agent work. Right now nobody treats it as first-class — the current tools store sessions and let you scroll them, the way you might tail a log file. But a log you can only read is a log you can’t learn from, collaborate on, or build on.
There’s a principle I keep coming back to from people thinking hard about how AI-native companies should operate: own your context. Your explicit knowledge, but also the tacit, institutional knowledge that used to live only in people’s heads. For the first time, that knowledge is being written down — it’s being typed into agent sessions every day. The catch is that most of the tools capturing it want to hoard it to build a moat. If your sessions live somewhere you can’t fully access, learn from, or reuse, you don’t own your context. You’re renting it back from a vendor.
So the strategic question isn’t which agent to buy. It’s whether you’re treating the session as a real, owned, first-class object — or throwing it away a thousand times a day.
We’re building Habitat because we think the session is the primitive the AI-native org gets built around. Everything good — collaboration, learning, oversight, reuse — becomes possible the moment you stop treating it as disposable.