AI Won’t Walk Into the Old Building. The Business Will Walk Out.
Istanbul’s Grandbazaar since 1455, a 1954 mall in Detroit and one test for founders and investors: what AI agents take first, what stays scarce, and who keeps the value.
Hamal…
Most of you probably don’t know this word. Those of you who do, in English, are probably thinking of a star: the brightest one in the constellation Aries, about 66 light-years away and easy to find without a telescope on a clear night.
The hamal I am talking about doesn’t shine, yet I am going to use him as a metaphor for how we as founders can think about the agentic transformation of value creation, and how we as early-stage investors can ask the questions that guide our decisions as we make decade-long bets on what the future of work will look like.
The hamals I mean are these:
In Istanbul’s Grand Bazaar, the word means a man. (I am sure women could do the job. I have just never seen a woman hamal, period.)
Here is what today’s hamals look like, in a photo essay by the Associated Press.
A hamal is a porter, and in the old bazaar, Kapalıçarşı, he is still how many things get delivered. He wears a semer, a saddle of wood and leather strapped to his back, and he carries loads that belong on a truck: three trunks roped together, a wooden chest, a bale of foam mattresses so large that all you can see underneath is a pair of sandals. He walks bent forward, both hands on the rope across his chest.
Look at the old photo on the left at the top. Ten or so wooden crates are lashed to the porter’s back, stacked well above his head, and stenciled on every one: REFINED OIL. CAMEL BRAND. STANDARD OIL CO. OF NEW YORK, U.S.A. He is looking straight at the camera and almost smiling. Whatever ships and railways carried that oil most of the way from New York to Istanbul, the last part of its journey was a man’s back.
More than a century later, the hamal is still there. Istanbul has had trucks for most of a century and malls for decades, and none of it reaches the inside of the Grand Bazaar. When the Associated Press followed Istanbul’s porters in October 2025, it reported loads of 200 to 300 kilograms carried up to the fifth floor of an old merchant building.
Recently, a fellow early-stage investor sent me a framework for agentic business models and asked what I thought. Working through it made me think about this question:
As AI agents take over the work, what stays scarce, and who keeps the value?
When I was looking for an example, I found myself thinking of the hamals, which surprised me a little. The longer I looked at the hamal, the more he seemed to answer the question. Like a past messenger for a future puzzle.
It took almost a century for engines to take the premium out of muscle. Electricity carried that power to every corner of a building, wherever the work was.
Now, AI and compute are doing the same to the productive brain: following the procedure, processing, drafting, reconciling. Anything that can wear a collar, blue or white, will get automated.
But automation does not arrive everywhere at once. It goes first where the building was designed for it.
It can wait a very long time outside a building that was not.
Five and a half centuries under one roof
The Grand Bazaar has done business on the same ground since Mehmed II ordered its first hall built in the winter of 1455, two years after he took the city. People often call it one of the world’s first shopping malls. From the air it looks like one enormous building: a grid of terracotta roofs over sixty-one covered streets. Inside, the vaults are painted blue and gold, brass lanterns hang over cones of spices, and on an ordinary day somewhere between 250,000 and 400,000 people walk through. More than 4,000 shops. Eighteen gates. You will get lost. Everyone does.
Being first has its rewards: the building itself is now a large part of what the bazaar sells. But being first also means you were built before anyone knew what the future would need. The streets were laid out for people on foot. The hans, the merchant buildings around the bazaar, went up without elevators. Nobody in 1455 was planning for delivery trucks. So the bazaar kept the only delivery system that fits a fifteenth-century street: the hamal.
The mall that put the trucks underground
Five centuries later, a Vienna-born architect named Victor Gruen showed what building for the future looks like. His Northland Center opened outside Detroit in March 1954 as the largest shopping center in the world. Shoppers saw the stores. Under their feet ran several miles of service tunnels with 484 rooms, used mainly for truck deliveries and storage (and, at one point, a nuclear bomb shelter). Most malls built since have their own version: loading docks, service corridors, freight elevators. Goods move on one level and people on another. Nobody carries a refrigerator on his back through the food court.
The newer malls do not need hamals.
The builders of newer malls were not smarter. They could just see what the future would bring: cars, trucks and deliveries at scale. You cannot add Gruen’s tunnels to the Grand Bazaar now. You cannot widen a street that has been a street since the fifteenth century, or cut a freight elevator through an old han, without breaking the thing that makes it worth visiting. So the hamal stays.
Yes, their work is shrinking, and not (at least not yet) because humanoid robots are walking around the bazaar carrying the boxes. In a recent photo essay, the Associated Press found that work has begun to dry up as the large wholesale merchants moved out of Eminönü, the old trading district around the bazaar.
I think this is how it usually goes, and an important lesson for us:
Automation rarely walks into the old building. The business walks out of it.
Why now: the collar sits on the task
A collar (blue or white) is the sign of a procedure, and procedures are exactly what agents take over.
What they cannot take over yet comes in two kinds.
The first I call Gray Matter, for lack of a better term: deciding what is worth doing, handling the case with no precedent, owning the result.
The second I call is Standing: whose name is on the result when it goes wrong.
The collar sits on tasks, not people. Most jobs hold some of each, and the early numbers already cut across job titles.
Air Street Capital’s State of AI Report 2026, citing The Economist, shows employment between May 2023 and 2025 down 18 percent for data-entry keyers and 9 percent for bookkeeping clerks, while paralegals grew 11 percent. I would not have predicted the paralegals.
My guess is that their jobs hold more Gray Matter than the title suggests, or that law firms have not rebuilt their workflows yet. Or law firms figured out that they don’t have to hire new attorneys if paralegals can do more of the work with AI. I don’t know which. Maybe all.
The order is also reversed this time. Engines came for the hands first. AI is coming for the desk first, because screen work needs no robot.
The same report says physical AI is approaching “its GPT-2 moment,” and in the same pages describes a robot finishing a five-minute, ten-stage assembly in 2 of 10 trials. So the first hamals to go will not be the ones in the bazaar. They will be the ones at desks.
Two reasons a person is still there
I think most companies alive today are closer to the Grand Bazaar than to Northland.
They were built before agents, around people: the approvals, the handoffs, the systems that don’t talk to each other, the spreadsheet that travels between three departments by email. Inside every one of them are people doing a hamal’s job without the saddle. They rekey an invoice, copy a number from one system into another, chase a signature through a chain of inboxes. Those jobs exist because the building is hard to change. The work itself does not need a person.
The new companies are being built like Northland. An AI-native company designs its tunnels first: clean data, systems that talk to each other, workflows an agent can run end to end.
In an earlier article, The Streets Are Lit, I argued that the real gains come when intelligence can “sit at any point in a workflow, and the workflow is rebuilt around it.” That rebuilding is the tunnel. A company built this way never hires the hamals in the first place.
Get Serhat Pala’s stories in your inbox
Join Medium for free to get updates from this writer.
But the Grand Bazaar did not die when the malls came. What it lost, I think, was the everyday shopping, the socks and saucepans that are easier somewhere with a parking garage. You will not find an Apple Store or a Forever 21 inside. What it kept is what a mall cannot copy: the carpet seller who can tell you where a rug was woven by turning it over and reading the knots (Gray Matter), and the jeweler whose family name has been over the same shop for generations (Standing).
So people still do the work inside the Grand Bazaar for two very different reasons.
The carpet seller (and alike) is there because his judgment is scarce.
The hamal is there because the street is narrow.
The trouble is that, from the inside, the two reasons feel exactly the same. Both men show up every morning. Both get paid. Both can say, truthfully, that the bazaar could not run today without them. But one is protected by something he has, and the other by something the building lacks. And the building is not his.
There is a simple way to tell them apart. I call it the rebuild test:
If we started this from scratch today, with everything we now know, would this work still need a person?
Rebuild the bazaar tomorrow and you would still need the carpet seller. You would not need the hamal. Yes means merchant work. No means hamal work, however safe it feels today.
The test works at three sizes: a person, a company, a portfolio. Let’s run all three.
The test at the size of a person
Let’s start with me. It is only fair to run the test on my own week first, because it would be easy to apply it only to other people’s jobs. Full disclosure: I did not enjoy this.
What survives is narrower, and I would like to believe it is merchant work. I am not sure. The question nobody has asked is only scarce until a system has read enough founder conversations to ask it first.
Now yours. Write down the five things you spent the most hours on last week and run each one through the test. And because nobody is fully objective about their own job, take your answers to someone you trust and let them score you a second time. They are usually harsher. Listen anyway.
If you are early in your career, run it hardest. The same State of AI report I mentioned above finds that job starts appear to have slowed for 22- to 25-year-olds entering AI-exposed jobs. The cause is uncertain, but the entry level is where most of the hamal work in any office lives. Which of your tasks would survive a rebuild, and who around you does them well?
The test at the size of a company
At the size of a company, the test has a second half: who owns the tunnel? When work moves from the street into the tunnel, it does not disappear. It moves to whoever owns the tunnel. Every workflow rebuilt for agents, and every edge case a system learns from, moves a little more work underground. I call that the loop, and what matters is who keeps what it learns. Gray Matter does not vanish in that process. It concentrates. Decisions that used to be spread across a hundred clerks get made once, by the people who design the system, and then the system makes them again ten thousand times. Gruen decided once where the tunnels would go. Every delivery since has followed that decision.
In Surfing the AI Wave I wrote, “As AI makes one layer cheap, you climb to the next.” I still believe it. The bazaar makes me add one thing:
Whoever owns the tunnels keeps the value.
So why don’t incumbents just rebuild? The incentives answer it. The building is part of the brand. The people who would approve a rebuild run the old building, and many of them rose by running it well. And the cost of the hamals is small and visible, while the cost of a rebuild is large and uncertain. So the incumbent keeps its porters, sensibly, year after year, until the business walks out.
That leaves founders two good doors.
Build the mall next door: a company designed for agents from day one. Or buy the bazaar and rebuild it from the inside, which I wrote about in We Sold Hours and SaaS 2.0. Either way, the rule is the same.
Don’t build a business whose value depends on someone else’s street staying narrow.
While looking for examples, I came up with two businesses that show how differently this can play out. The first is an AI bookkeeping firm built by buying small local practices and automating them: it buys an old building and rebuilds it as a mall. The reconciling and the month-end chase for receipts are exactly what agents do well. (Those are the bookkeeping clerks, down 9 percent.) What stays is narrow: the hard cases, the worried owner in March, and the signature on the books.
The second is a home cooking service. An agent plans the family’s week, orders through Instacart, a robot cooks, and the family rates the meal. A kitchen is the narrowest street there is: every one is different, nothing is standardized, and the robots are not ready. So it is expensive and slow today, and I am honestly unsure about the timing. But each rated meal moves a little of the work into the tunnel, and whoever runs the purchasing ends up owning something much bigger than dinner: the household’s weekly demand.
The bookkeeping tunnel is easy to dig, and plenty of capable teams will dig the same one, so its scarcity has to come from owning the client and the signature (a.k.a. Standing). The kitchen tunnel is hard to dig. That is why it is expensive today, and why, if someone does dig it, few will follow quickly.
Scarcity lasts about as long as the tunnel is hard to dig.
Notice the bottom row. Better porters and hired diggers both work on someone else’s building, and both hand most of what they learn to someone else. Only the top row keeps it.
The test at the size of a portfolio
For those of us who write the investment checks, the test turns around. We are not asking whether a company still needs its people. We are asking whether its customers would still need the company:
If this company’s customers were rebuilt from scratch tomorrow, would they still buy from it?
That question sorts the AI pitches I see faster than any other. Many of them sell the bazaar better porters: tools that automate hamal work inside old companies. They are easy to sell, because the pain is visible and every bazaar has a budget, and some will be good businesses for years. But a customer rebuilt from scratch does not buy a better porter. It digs a tunnel.
A better porter’s market is the life expectancy of the old building.
Revenue will not warn you. A company selling better porters and a company that owns its tunnel can show the same growth curve for a while. The difference shows up later, when the business walks.
So the useful question is what protects each company, and for how long.
Sort your own portfolio, or your pipeline, with it. I suspect most of us would find more time than protection. The protection I think we underrate most is Standing. It is the least glamorous line in a pitch: someone has to sign, certify, insure or guarantee the result, and the customer wants to know who. A company that gets paid for its name on the result has something a model cannot copy.
When one company deserves a full diligence pass, I use five questions instead of one. Here are the two businesses again, placed on all five.
For founders, the six protections read from the other side. If I were building now, I would want to own at least one of three things: the customer, the learning, or the signature. The customer, so you are not living on someone else’s building. The learning, so every task makes the next one cheaper; if human touches per hundred completed tasks stay flat month over month, you are running a better porter. The signature, so that when the result matters, it is your name the customer calls. Own none of the three and you are selling better porters.
Where the truck can’t go
I don’t know how many hamals still work in and around the Grand Bazaar. I do know what kept them there for so long, and it was never their strength. It was the building.
If the bazaar could be rebuilt tomorrow with Gruen’s tunnels underneath, the hamal would be just as strong as he was the day before. He would simply no longer be needed.
I don’t want to make that job picturesque. The work is hard, poorly protected, pays very little, and already shrinking, and the people doing it will bear most of the cost when it ends. The porter in the Standard Oil photo is almost smiling. I have no idea what he was paid. But today’s porters most likely don’t smile to the camera. None I saw in that Associated Press photo essay do.
So the question I want to bring into the next pitch meeting is not how much of the work the agents do. It is where the truck can’t go, and why. Is it because the work needs a person, or because nobody has rebuilt the building yet? If it is the first, the company has something to defend. If it is the second, it has time, and I don’t know how much.
Where can’t the truck go in your company, and why?
