Google's reported $10M bid for Spirit Airlines' inboxes prices the data AI agents actually lack

The reported facts are few but pointed: Google has offered $10 million for Spirit Airlines' emails and work records, with the stated purpose of developing AI agent capabilities. Details on scope, anonymization, employee consent or whether the offer was accepted have not been reported, and we are not going to fill those gaps with speculation.

Start with the number. Ten million dollars is a rounding error for a company whose AI capital spending runs into the tens of billions a year. That asymmetry is the story. It suggests the buyer is not paying for volume — it is paying for a *kind* of data that money cannot easily manufacture.

Frontier models are no longer starved of prose. They are starved of process. An agent that books, rebooks, escalates and refunds needs to see how an exception was actually handled: the approval chain, the half-explained workaround, the thread where someone overruled the policy and why. Corporate email is the archaeology of operational decision-making, and almost none of it is on the public web. That is why an airline's inbox is worth more per byte to an agent lab than another crawl of the open internet.

The uncomfortable part is who is at the table. Employees wrote those messages. Customers, suppliers and partners appear inside them without ever having been asked. We are not alleging that anything improper has been proposed — we simply note that in a transaction like this, the parties negotiating the price and the people whose words constitute the asset are not the same group. Workplace communications have never had a clear market value before; that value is being discovered now, deal by deal, with no settled rules on consent, deletion or downstream use.

Our read: this is price discovery for operational data, and it is arriving before the governance does. Expect the rules to be written where they usually get written lately — by state attorneys general, privacy regulators and judges applying existing law to a situation nobody drafted for — rather than by any new AI statute. Short term, the friction is real: employees will reasonably ask what happens to a decade of their messages, and companies will discover their archives are both an asset and a liability.

Long term, we think this direction is right even if the first deals look clumsy. Agents that understand how work genuinely flows — not how a manual says it should — are what turn AI from a chat toy into something that absorbs the administrative drudgery we have already flagged as the most exposed layer of white-collar work. That absorption is how the abundance case gets paid for. But it only stays legitimate if the people inside the training data are treated as parties to the deal rather than as inventory. Right now, on the evidence available, they are not.