Google paid $10 million for a dead airline's traces

Google just agreed to pay $10 million for a dead airline's data.

Spirit Airlines shut down in May. Last week Google won the bankruptcy auction for its internal records: 100 million emails, 500 million Teams chats, 30 million lines of custom software, 7.5 billions of passenger transactions and much more.

I keep thinking about what's in that package. Not facts. Traces.

How problems surfaced. Who got looped in. How a decision moved from a chat thread to a code change. How work got done.

You can't scrape that from the internet at any price. And if you're training AI agents to do enterprise work, it's exactly what you need.

So deploying agents into your workflows is only half the play. The other half is logging everything: what the agent decided, where a human stepped in, what got changed, whether the fix held.

That record is your tomorrow's proprietary training data.

Analysts estimate 50 to 70% of the humanoids produced in China this year will spend their lives collecting training data, not doing paid work. The robot is just a pipeline. The real product comes later.

When you feed your own loop the same way, every iteration starts from what actually worked last time in your environment.

Companies that keep that loop running are building the one dataset nobody can buy at auction: how their organization actually works, and how to make it work better.

That's the moat.