Google's $10 million winning bid for Spirit Airlines' deidentified enterprise data, a proposed sale awaiting court approval, and calls at World Summit AI in Amsterdam to redesign organizations point toward agents working inside how companies coordinate. Agents can remove structural friction such as handoffs and re-explained context; political friction needs changed decision rights and incentives. Oversight must fit limited attention, and AI challengers are not neutral. The practical start is one cross-department outcome with a named accountable owner.
In August, Google was named the winning bidder, at $10 million, in Spirit Airlines' bankruptcy auction for its deidentified enterprise data. On October 8, members of Congress wrote to Google's Sundar Pichai and Spirit's Dave Davis about the proposed sale for AI training, described as roughly 100 million emails and 500 million Microsoft Teams messages, and asked for enforceable worker protections before any transfer. The sale still needs court approval.
Why pay for an airline's internal messages? They hold something the public web doesn't: a record of how an organization works.
What an airline's inbox knows
An operational archive shows what happens between departments. A delayed aircraft becomes a thread between maintenance and crew scheduling. A storm closes a hub, someone bends a rule, and someone else writes down who approved the exception.
The letter speaks of AI training, not enterprise agents, and I don't know Google's plans. Still, I read the bid as a clear signal of where we are moving: from assistants that help individuals toward agents that take part in how organizations coordinate.
What I heard in Amsterdam
I attended World Summit AI in Amsterdam last week and heard at least five talks making the same case, a second clear signal. My notes cover four of them.
The Beyond Automation keynote argued that faster individuals don't add up to enterprise return, because AI budgets go to one function at a time while work crosses all of them. Ask whether a workflow deserves to exist, my notes say, and you end up at the org chart, built around scarce people when it should be built around outcomes.
On The Intelligent Enterprise panel, David Martineau of Sia argued, as I understood him, that with similar models everywhere, the difference comes from reinventing processes built only for people. Giovanni D'Aniello, AXA's Group CIO, described core insurance systems as digitized paper processes. My notes have AXA running agents through a hub with orchestrators, guardrails and verification points, starting with a subset of claims.
The keynote The last mile is not artificial went furthest, placing the bottleneck in organizations whose structures, incentives and attention haven't kept pace with the models.
Two kinds of friction
One slide split the coordination tax in two. Structural friction is handoffs, status meetings, context re-explained and versions that drift apart. Political friction is silos, opaque data, control of the narrative and the people who benefit from the tax.
Agents suit the structural part. One can carry a permissioned record of a delayed flight across maintenance, crew scheduling and customer service and ask for what is missing, so nobody re-explains the situation at each handoff. It cannot settle who may approve an exception, or dissolve a silo someone is rewarded for keeping. That takes new decision rights, and incentives that reward resolving work rather than passing it on. Leaders must still tell necessary control, such as safety or legal approvals, from defended territory.
In *Interstellar* (2014), the robots TARS and CASE are not equipment stowed in a cargo bay. They are crew. They help fly the ships, run the numbers and take the dangerous jobs, and their settings are out in the open: TARS reports his honesty at 90 percent, and Cooper turns his humor setting down.
That is closer to what I think organizations need from agents than another chatbot. Each robot has a real place in the mission, everyone knows what it is responsible for, and a person can see and change the limits it works under. Nobody on that ship has to re-explain the mission at every handoff, and nobody pretends TARS is neutral just because it has no career to protect. The crew was designed with the robots in it. Most org charts still have no box for that.
A human role that fits human attention
The same keynote warned that people left in charge can lose critical thinking and the craft they stop practicing, and drown in agents asking for direction. "Human in the loop" means little if one person approves a stream of requests all day. The redesign must say which routine actions agents own, which decisions need a person, and how an exception arrives with the evidence to judge it.
The talk closed with AI monitoring AI, humans judging critical AI decisions, and AI challenging human ego and politics. I like that mutual scrutiny, but an AI with no career to protect is still not neutral: tested models often tell users what they want to hear, and any system reflects whoever deploys it. Its challenges must stay open to dispute, with a named person accountable.
People who fear for their jobs also stop sharing context and flagging an agent's mistakes, so trust has to be designed in from the start.
Redraw one part of the chart
Pick one outcome that crosses departments and redesign the organization around it. Take one category of insurance claims. An agent keeps the case context and coordinates checks across claims and finance, while people keep consequential and disputed judgment. One named person owns the resolved claim, and incentives reward closing it rather than passing it along. Start with a subset and measure finished claims, rework and the owner's review load.
Google bid for a record of how one airline coordinates its work. Every company already holds that record about itself. I think the leaders who pull ahead will use theirs to redraw the org chart for people and agents together, starting with one outcome.
