Your rivals can buy your AI. They can't buy what you point it at.
I scored 105 real estate workflows to find where AI gives a firm an edge its rivals can’t copy. Used the way most firms use it today, none of them do. Pointed at what a firm already owns, many more could.
This piece started with a tool and ended with a jolt. For the past few weeks I’ve been building a map of where commercial real estate firms might find durable advantage as AI spreads, scored against every workflow in my RIRA library. When the scores came back I was surprised, a bit shocked in fact, by how little of the gain looked likely to stay with the firms doing the work. Customers got most of it. Landowners, software vendors and contractors got some. The firms themselves got remarkably little. Then I looked again, and the shock turned out to be about how firms use AI. What AI can do for them is a better story.
Executive summary
Every firm can buy the same AI, so most of the gains end up with customers, as the savings from Buffett’s new looms did. Two questions decide what a firm keeps: can rivals get the same capability, and how fast does the gain get repriced? Slow repricing gives firms that lean in years of real money. A durable edge sits only where AI works on something the firm owns: rights, unpublished evidence, relationships turned into records, and its own methods.
WHAT I SCORED
The 105 workflows come from my RIRA Master Workflow Library, across institutional real estate, rental living, workplace and master development. For each one the question was the same. Once your competitors have the same AI, who keeps the gain?
I scored all 105 with Claude Opus 5.5 High. Twenty were then scored blind by three models (Claude, Gemini Argon and ChatGPT-6.1 Sol, High), none of them seeing the others’ answers. They agreed well on the big call: whether a workflow’s best realistic outcome is a commodity or a differentiator (a Fleiss kappa of 0.64, for those who like their statistics). They agreed poorly on finer calls about where things stand today. And three models are not three practitioners. Treat what follows as structured judgement, dated November 2026, rather than measurement.
As firms typically use AI today, with the same tools on the same data, none of the 105 workflows gives an advantage rivals can’t match once they buy those tools. Some of those gains still pay for years before they’re priced away. In my scoring, 49 of the 105 could become a durable edge with redesign, but only where the firm owns something its rivals can’t quickly buy. Reviewers judged some of those calls generous, so treat 49 as a ceiling.
Put like that, the first finding sounds close to obvious. If everyone buys the same tools and points them at the same data, nobody gets an edge.
It is also, as far as I can see, how most of the industry is deploying AI.
THE LOOM
Charlie Munger told the story in a talk at USC in 1994. The managers of Berkshire Hathaway’s textile mill came to Warren Buffett with a new loom that did twice the work of the old ones. Buffett’s reply: “Gee, I hope this doesn’t work because if it does, I’m going to close the mill.”
The looms worked. As Munger put it, “the savings didn’t go to you”. Every mill could buy the same machines, so the cost reduction went straight through to customers in lower prices. The man who bought the equipment got nothing for it except the right to stay in business.
Buffett explained the timing in his 1985 letter to shareholders, the year the mill closed. Competitors had made the same investments, “and, once enough companies did so, their reduced costs became the baseline for reduced prices industrywide.” How long that takes is what this piece is about.
Most of our 105 workflows are looms.
That is the normal fate of a general-purpose technology in a competitive industry, and I argued as much in July in ‘Better, Faster, Cheaper, Fewer‘: efficiency gains are the ones certain to be competed away. The scoring confirms it workflow by workflow. The loom is still compulsory. The mill that skips it closes anyway, just sooner.
TWO QUESTIONS, FOUR BOXES
JLL’s 2026 Global Real Estate Investor AI Adoption Survey gets halfway there. In acquisitions, it says, efficiency gains get competed away and the window for an edge is short, so invest in differentiation. In asset and property management, gains tend to stay with the company, so invest in effectiveness. What it misses is that the same activity can land in different places, depending on who is paid, and how.
So I ask two questions. Can the firms you compete against get the same capability? And how fast does the gain get repriced, passed on to clients and customers?
| Easy to replicate | Hard to replicate | |
|---|---|---|
| Repriced quickly | Parity. Necessary just to keep up. Buy cheaply and expect no excess return. AI-assisted market analysis for bids. | Edge. Build and protect. A rent-review adviser whose unpublished evidence wins instructions, on fees still set by tender. |
| Repriced slowly | Harvest. Move fast and bank the gain before it's passed on. A managing agent's efficiency savings on a fixed fee, until retender. | Compound. Invest patiently. This box is rare. Pricing and renewals learning from tens of thousands of rental homes. |
Each box calls for a different stance. JLL’s split gives you two of them. The repricing clock adds the third and fourth.
As firms deploy AI today, every one of the 105 workflows sits in the left-hand column. That is less grim than it sounds, because Harvest is real money, sometimes for years. But advantage over your rivals only exists on the right.
WHY THE CLOCK RUNS SLOWLY
How fast a gain gets repriced depends on who is paid, and how. Take one improvement: AI cuts the work of running a multi-let office building.
-
The managing agent on a fixed fee keeps the saving on its own labour until the contract is retendered. That’s Harvest.
-
The occupiers get any cut in recoverable costs through the service charge, more or less at once. That gain was never the firm’s to keep.
-
The owner keeps savings on costs it bears directly, such as empty space and non-recoverable spending. They flow into income and value, until buyers start to price them in.
One improvement, three beneficiaries, three clocks. Which is why JLL’s view that management gains stay with the company is only partly right in a UK multi-let building. Fee models set the pace elsewhere too: hourly billing passes savings on almost at once, while percentage-of-assets fees and fixed retainers hold them until the next fundraise, retender or review.
In October, in ‘Show Me the Incentive‘, I borrowed Munger’s other famous line, ‘show me the incentive and I’ll show you the outcome’, to ask who inside a firm gets the gain from AI. Turned on the firm itself, it explains the industry’s pace. Nothing forces a manager on a percentage fee to hurry, so slow adoption is rational.
Which is exactly why the firm that does hurry gets years of advantage over peers who are content to wait. And it’s why AI-native newcomers will do well. A firm with no legacy cost base can price analysis, production and coordination at the new cost from day one. What it can’t do is conjure the buildings, consents and capital that incumbents already hold.
THE RIGHT-HAND COLUMN
What makes something hard to replicate? The answer in the Durable Advantage Map is a ladder. An advantage lasts longer the more it rests on rights, place and institutions, and the less it rests on people, data or tools. Rivals buy the same tools within months, and data histories can be bought or go stale. Title, consent and committed capital can’t be created by AI at all.
In the blind review the two independent models read ‘today’ differently from me: as AI already pointed at assets a firm holds. Under that reading they called a handful of workflows differentiating now. They agreed with each other on only one, so I won’t pretend this is a count. But the three strongest cases share a pattern, and it’s instructive.
Acquisition due diligence. Under exclusivity the buyer sees evidence no other bidder sees, while it can still retrade or walk away. AI that reads that evidence faster and harder than a rival could is an edge, for that deal. It rests on a contractual right.
Rent review. A large landlord or adviser holds unpublished comparables, unredacted side letters and a record of past arbitration and expert determinations. AI pointed at that evidence argues a better case than AI pointed at the published market. It rests on records nobody else can buy, because they were never published.
Origination. The only workflow both independent models called differentiating today. Its edge is relationships with owners, knowledge of who has debt maturing and a history of past approaches. JLL found the same thing from the other side: among investors whose edge is relationships and proprietary deal access, only 22% fund AI for relationship management, against 43% funding AI only in direct deal sourcing. Most are pointing AI at the part of origination every rival can buy.
Origination shows how a firm climbs the ladder. Relationships held by people walk out of the door with them. Wire AI into the approach log, the debt-maturity tracking and the signals that prompt a sale, and that edge becomes an institutional routine the firm keeps when people leave. Log which signals led to a conversation, which conversations led to an instruction and which deals closed, and the firm can start to improve how it originates the next one, provided it does enough of them to learn from.
Method. There is a fourth kind of ownership, and it is the one most firms can start on now. A firm that builds its own methodology into Skills (how it underwrites, runs a rent review or tests a business plan) is already using AI differently from most of its peers. That method is an institutional routine, a rung below rights and unpublished records. Rivals can hire your people or infer the method from what you deliver, and each new model narrows the gap. Capturing what happens each time the method is used is how a firm improves it, and the edge lasts only as long as it improves faster than others can copy.
FOUR ROUTES
In each of these the edge comes from ownership, and AI is how a firm puts what it owns to work. That gives four routes out of the left-hand column, in rough order of how soon they pay.
-
Wire in. Point AI at what you already hold privately: records, evidence, relationships, rights, and the methods your best people use. This is cheapest and quickest because the asset already exists. If the JLL figures are any guide, most firms haven’t connected it yet.
-
Consolidate. Real estate services are fragmented. When AI cuts the cost per asset, per unit or per instruction, the firms that use the lower cost base to grow take share from those that don’t. Profit per unit falls, but more of it ends up in fewer firms.
-
Climb the ladder. Turn what sits with people into routines and records, and turn temporary leads into rights, positions and contracts.
-
Build what has newly become viable. This is the bigger pie I argued for in ‘From Boardroom Privilege To Baseline Utility‘: institutional-grade service at a price that opens markets ten times the size of the one served today. Its profits are invisible to any analysis of today’s workflows, mine included, because those workflows don’t exist yet.
WHERE THERE IS NOTHING TO WIN
Some domains have no durable upside in the scoring at all: fund finance and governance, property financial management, insurance, owner-side valuation, resident protection and workplace finance. Do the work well, automate it fast and buy the tools cheaply. Bank whatever the fee structure lets you keep, and look for your edge somewhere else.
THE MILL AND THE MILL-OWNER
For most of this industry, then, AI is a forcing function far more than a source of profit. Customers will collect most of the gain, as they did from Buffett’s looms, and the industry’s profit per unit of work will shrink.
That doesn’t mean every firm shrinks with it. The years a slow repricing clock gives you are the most valuable thing AI hands an incumbent, and most will spend them waiting.
Buy the looms. Bank the harvest. Then spend those years putting your AI to work on the things your rivals can’t buy.