Shutting-down startups are selling their Slack archives, emails, and Jira tickets to AI labs. Payouts range from $10,000 to $100,000 per company. The CEO brokering the deals puts it bluntly: "There's a feeling of a gold rush."

Why? Because the public data of the internet ran out. And what new AI systems need isn't clean text — it's the ten-o'clock-at-night conversation where someone says "this doesn't work" and someone else replies "I know, but the client needs it by Thursday."

This is an act of cartography. And cartography has a long history.

From Ptolemy to Google Street View, the question has always been the same: who decides what deserves to be recorded. Hawthorne showed that observing changes what's observed. Mintzberg needed weeks to study five executives. Bell tried to record everything about his own life and gave up.

And then, without anyone planning it, organizations began recording everything about themselves.

Hayek wrote in 1945 that useful knowledge never exists in concentrated form — only as dispersed bits possessed by separate individuals. Slack messages are exactly that. And now someone has figured out how to concentrate them.

Every message we write is, potentially, a training data point. The question is not whether this changes something. It's what exactly it changes.