In the previous post I wrote about the two answers this industry has already tried, and why neither one is somewhere to stand.
I spent part of a trip to Cartagena talking with my sister-in-law, who's an architect, and my brother, who works in thermal and acoustic insulation. Windows and facades.
None of it was about software. They talked about how the buildings behave.
The same chain doesn't behave the same way in every city. A property in Cartagena sits in heat and salt and humidity. One in Bogotá doesn't. Which way a facade points changes what happens inside it. The envelope, meaning the windows and walls and everything separating inside from outside, decides how hard the building has to work to stay comfortable. None of that is guesswork. It's measurable, and these are people who measure it.
Then I brought maintenance up, and not by accident. I'd been carrying the question from the last two posts, and I wanted to know if there was something in all that knowledge that could be of interest for the hotel. When something needs servicing, when a seal is done, when a treatment on a facade needs redoing.
I don't know how those calls actually get made. I've never seen their systems and I didn't ask. So everything from here is me wondering, and I want to be upfront about that.
But here's what I found myself imagining.
If you're deciding when to service something, you're probably looking at how it has performed, and where the property is, and what season it is, and how full the hotel has been. And then somebody who has watched that particular building for years does something with all of it that isn't written down anywhere.
Maybe that's wrong. Maybe there's a system that does the whole thing. I'm planning to ask.
The shape is what matters, whether or not I have the hotel right. The systems hold facts. A person holds the reasoning that turns those facts into a decision. And the reasoning is the part nobody writes down.
The systems hold facts. A person holds the reasoning that turns those facts into a decision. And the reasoning is the part nobody writes down.
That's especially true across Latin America, where a lot of companies run several systems that were never designed to meet, bought at different times for different reasons, each holding one piece of the picture. The only thing holding all of it at once is somebody's memory.
There's a phrase going around that comes at this from the other side. The AI doesn't know your company. It's the shortest version I've seen of why so many of these projects stall. The models are extraordinary at general knowledge. They know nothing about why your team stopped doing something a particular way, or which of your buildings always surprises you.
The governance products are about whether an agent is allowed to act. Who it acts as, what it's permitted to touch, whether the action was logged, who can shut it off. That work is serious and I'm glad people are doing it properly.
There's a second category that does go after context, and it has a name now. Gartner said context engineering was in and prompt engineering was out about a year ago, so I'm not pretending nobody thought of this.
The part that worries me is a different one. Not where the knowledge sits, but who keeps it true, and whether the people whose knowledge it is can change it themselves.
Because you can have complete control over what an agent is permitted to do and still watch it make a confident, fully authorized, wrong decision, if the reasoning it needed was never available to it.
So the what-if I came back with is this. What if the thing we're missing isn't permission, it's context? Not more autonomy, not tighter control. Just the same understanding the person has, in a form an agent can use.
And kept true. What the operations team knows in August isn't what they knew in March, and a building that behaved one way for five years behaves differently after a renovation. Knowledge nobody tends stops being knowledge fairly quickly.
Which is also why I don't think this replaces anyone, and I'd want that clear. The whole thing depends on the person staying, still deciding, still correcting it when it drifts. It creates a permanent need for someone's judgment instead of removing one. If they leave, you don't inherit their reasoning. You inherit a copy of it that's already going stale.
If they leave, you don't inherit their reasoning. You inherit a copy of it that's already going stale.
So the two things meet. Everyone can build the workflow now. What feeds it is what I'm not sure about.
I don't know if that's right. It's a what-if built on one conversation about facades and a lot of guessing about the rest.
The obvious next question is what it would actually look like. Not as a product, as a shape. And the interesting part is that I'd already built a version of it for myself without realizing what it was.
That's what I want to write about next.