KM4BI

Google Has More Data Than Almost Anyone. So Why Is It Bidding $10 Million On Old Emails?

Spirit Airlines’ work archive drew a $10 million bid, but emails and chat logs may teach agents the performance of work rather than its real value.

Published

Google won a $10 million bid for Spirit Airlines’ digital work archive. The package reportedly includes roughly 100 million emails, 500 million Teams records, code, and business documents. Buyers believe these traces could help train AI agents to perform knowledge work.

Work traces are not the work itself

A thirty-message thread may document an invoice resolution without showing that the valuable act was one person spotting a misapplied credit. A closed ticket may look like success even if a duplicate payment is found six weeks later. Corporate archives mix useful decisions, necessary coordination, bureaucracy, and mistakes.

That distinction matters for training. If an agent learns from every message without understanding outcomes, it may imitate the performance of work — updates, follow-ups, and meetings — instead of producing business value.

A market shaped by distressed companies

Spirit Airlines is bankrupt. More broadly, companies that sell their archives as a last resort may be the same organizations that failed to turn those records into durable value. Available training data could therefore overrepresent troubled companies while being used to define competent work.

Mercor, which reportedly bid $7.5 million against Google, is also investing in training environments where experts define assignments and success checks. That reinforces the central point: raw records are not enough. Someone still has to decide what a correct outcome means.

Workers need a role in the process

Spirit’s flight attendant union raised concerns about confidential employee information and sensitive data copied across systems. Workers should be able to identify records that must be excluded, understand the intended use, and have their expertise recognized when archives are converted into training material.

For businesses, the stronger path is not necessarily selling archives or teaching agents to mimic workplace conversations. The clearest gains come from repeatable tasks, structured data, reliable reference documents, and evaluations designed with human experts. Agents can process information and produce measurable outputs while people retain responsibility for context, judgment, and alignment.

Source