The most valuable thing at your company disappears every time someone closes a chat

Somewhere in your company right now, one person has quietly figured it out.

They’ve worked out the exact way to get a great result from an agent for some task that matters — the context to load first, the way to frame the ask, the sequence of steps, the correction that keeps it from going off the rails, the moment to take over by hand. It took them a dozen frustrating sessions to get there. Now they’re fast and the output is excellent.

And when they close that tab, all of it is gone.

Not gone from them — they’ll remember. Gone from everyone else. The next person who needs to do that task starts from zero, re-deriving the same lesson through the same dozen frustrating sessions. Multiply that across every task and every person and you’re paying an enormous, invisible tax: your organization re-learns the same things over and over and never compounds any of it.

Why can’t you see how other people use AI agents?

We used to have a name for the opposite of this — institutional knowledge — and we had places it accumulated. Runbooks, wikis, the senior engineer you could roll your chair over to. It was imperfect but it existed. The agent era quietly broke it. The most valuable know-how in your company is no longer written in a doc or demonstrated at a whiteboard. It’s happening inside thousands of private sessions, and it evaporates the instant each one ends.

This is the part people underrate: the knowledge that matters most in an AI-native company is tacit. It’s not the answer the agent gave — it’s how a skilled person got the agent to give it. That’s exactly the kind of knowledge that never made it into documentation even in the old world, because it was hard to articulate. The difference now is that it’s actually being externalized — typed out, step by step, in the session — and then thrown away.

The best way to use an AI agent is to start from one that already worked

The move that fixes this is one I’d call encode your expertise: capture the patterns of what works, make them legible, and put them back in front of the next person right where they work. Not a wiki nobody updates. The good session itself, promoted into something reusable — a template, a shared tool, a starting point — so the person after you inherits your dozen sessions of hard-won learning instead of repeating them.

Every agent session at your company right now starts from zero. It doesn’t have to. The knowledge is already being written down. The only question is whether you keep it.