Writing · May 2026 · 3 min read

The Rented Part

Why “which model do you use” is the wrong question.

Nobody asks which spreadsheet a rival runs. Excel never gave anyone an edge. The edge was the model your team built inside it: twenty years of your deals, your debt structures, your investors’ questions, settled into rows and formulas nobody else has.

The spreadsheet was the rented part. The model was yours.

AI splits the same way, and the split is easy to miss because the intelligence is so new and so loud. The intelligence is rented. It arrives by subscription, it improves without your permission, and every firm in your market can rent the same brains from the same three labs for roughly the same money.

Whatever advantage AI brings your firm, it cannot live there.

No model stays best

Whichever model is best today will not be best for long. The title has changed hands every year since these tools arrived, and nothing suggests it will stop.

The useful news is that you do not need the winner. Today’s models are already good enough for this work, provided the work is set up so they cannot be wrong quietly: the same steps every month, a check at every move, your own model doing all the math.

So there is nothing to wait for. What a firm needs is not a smarter intelligence. It is the system that holds the intelligence to your standard.

The owned part

That system is built, not rented. In an installed workflow, the AI does two jobs, reading documents and drafting words. Everything around those two jobs is fixed: which documents, in which order, checked against what, graded to whose standard.

Those pieces are built in your shop, against your own past packages. The checks that catch a misread. The maps that tie every label to a cell in your model. The calibration that makes the output come out your way, not the average way.

Swap the intelligence and all of it holds. That is what owning the process means.

Starting over

Without a system, every new model means starting over. A new tool, new habits, a shop re-learning its own month, and the learning thrown away when the next thing arrives. That is what the last few years of AI have felt like inside most firms: motion without accumulation.

An installed workflow accumulates. The steps, the checks, and the calibration do not care what the labs released this quarter. If a better model is ever worth switching to, the switch is an afternoon of testing, not a rebuild. Most of the time it will not be worth it, because the work was never waiting on a smarter model.

Which model do you use? One that is good enough, held to your standard. The model was never the point.

We build the owned part for real estate operators: the steps, the checks, and the calibration that make any model’s work come out as yours.