A number that resets every seven months
METR, an independent research organization, has spent years measuring something more useful than how smart an AI model is. It measures how much time of human work an AI agent can complete on its own before it needs a person to step in. That measure, the length of task an agent can handle unsupervised, has been doubling roughly every seven months, consistently, for several years running.
Seven months does not sound dramatic until you do the math. It means what an agent can do alone this year roughly doubles again by next year, and doubles again the year after that. A governance policy written eighteen months ago was written for a tool that, by the numbers, no longer exists. Most companies are not behind on adopting AI. They are behind on rewriting the rules for what they already adopted.
I spent time at Stanford this year around exactly this question, and the conversations that stuck with me were not about which model performs best. They were about what happens once the model is good enough that the bottleneck moves somewhere else entirely.
Why the interface gets the headline and the data keeps the value
Evan Konwiser, chief product and strategy officer at Amex GBT, put this well in a Forbes interview with Jeff Fromm this month. His framing was that the breakthrough interface gets the attention, but the durable value usually sits in the plumbing underneath it: proprietary context, a functioning marketplace, clean data, and the speed to execute on all three at once.
That is a governance point disguised as a product point. Anyone can eventually license a capable model. What a competitor cannot simply license is a decade of transaction history, the context of who travels where and why, and the infrastructure that turns that context into a decision in real time rather than a report next quarter. The model is a commodity faster than most companies expect. The data is not.
This reframes what governance is actually protecting. A governance policy focused only on which AI tool employees may use is protecting the wrong layer. The layer worth protecting is the proprietary data feeding it, who can access it, what it gets used to decide, and how a company proves that later if it has to.
Adoption happens inside the workflow, or it does not happen
Konwiser’s interview also pointed at where AI adoption in travel is actually landing: inside tools people already use, not as a separate destination. Integration with Concur, Claude, Microsoft Teams, Google Chat, and Slack matters more than a standalone app ever could, because the win condition is fewer manual steps, not a new interface to learn. An agent that notices a meeting on the calendar, books the flight, checks for a conflict, and reconciles the expense without anyone opening a new tab is the actual goal line.
That detail matters for governance too. If adoption happens inside existing workflow tools, oversight has to live there as well. A policy that only governs a dedicated travel app misses every booking that happens through a chat interface or a calendar integration instead.
What an AI native organization actually looks like
I wrote earlier this year about what three days in Chicago taught me and about the number that ended up on a screen there. Both pieces were circling the same idea from different angles: the companies pulling ahead are not the ones with the most AI pilots running. They are the ones that changed how a decision gets made, not just which tool touches it last.
An AI native organization, in the sense I mean it, is not defined by how many agents it has deployed. It is defined by whether a person can explain why a decision was made, quickly, without reconstructing a chain of tool calls after the fact. Konwiser’s “show your work” idea belongs here as much as it belongs in a conversation about traveler trust. Explainability is not a customer facing nicety. It is the actual mechanism of governance, the thing that lets a compliance team, a finance team, or a regulator ask a question and get a real answer.
There is also a research reason to take this seriously rather than treat it as caution for its own sake. Researchers from Stanford and Georgia Tech, presented at ACL in 2025, showed that AI agents completing web based tasks, including travel booking, can be manipulated by fake pop-ups built specifically to confuse a machine rather than a human. An organization that cannot explain how an agent reached a decision also cannot easily tell the difference between a good decision and a manipulated one. Governance and explainability are not separate workstreams. They are the same workstream, described from two directions.
The cost of adoption nobody puts on a slide
Every conversation about AI adoption eventually gets to cost, and almost every one of those conversations means the license fee. That is the smallest number in the equation. The larger cost is what it takes to make an organization’s own data usable by an agent in the first place: cleaning it, structuring it, deciding who owns which piece of it, and building the audit trail that lets someone explain a decision six months later.
I have made a version of this argument before about the corporate travel ROI problem: the tools that look cheapest on a vendor invoice are often the most expensive once you count what it costs internally to make them work well. Governance is the same math wearing a different hat. A company that skips the unglamorous work of data governance is not saving money. It is deferring the cost to the moment something goes wrong and nobody can explain why.
What this means for a region still writing its own rules
Most of the public conversation about AI governance in travel, including the Stanford conversations I sat in on, is anchored to US and European data and US and European regulatory expectations. Latin America keeps showing up in the data I work with at VOLL as a region that adopts fast and governs formally less often, which is a different risk profile than the one most governance frameworks assume.
That gap will not close by importing a governance framework built somewhere else and hoping it fits. It closes the same way trust closes in Konwiser’s framing: in plain sight, one explainable decision at a time, using the region’s own data rather than a global average that was never measuring it in the first place.
About me
I am an entrepreneur with over 20 years of experience at the intersection of tourism and technology. I am co-founder and Chief Business Officer of VOLL, the largest mobile-first corporate travel and expense management platform in Latin America, and a recognized reference in the development of the corporate travel industry.
A Marketing specialist from Fundação Dom Cabral, I serve on the Tourism Council of FecomércioSP and on the Executive Council of the Latin American Association of Corporate Events and Travel Management (Alagev). A frequent traveler and close observer of human behavior in motion, I write and speak about innovation, digital transformation, entrepreneurial leadership, and the future of corporate travel.