Distribution is the moat AI can't replicate
Dheer Gupta argues that the models got cheap, but getting noticed and getting trusted did not. I think he's right. And for a small service business in a small town, that changes what it means to build a moat at all.
The essay, and why it stuck with me
Dheer Gupta's piece is called "The Only Moat AI Can't Kill." The argument is tight. Every technical advantage a company used to have is getting flattened. Anyone can spin up a model. Anyone can clone a product. Anyone can generate content at scale. What does not get flattened is distribution. Getting noticed by the right person. Being trusted enough that they open the email, show up to the meeting, and sign the contract.
His argument maps cleanly onto what I already see happening in the community I work in. Most of the new work I take on doesn't show up through an ad or a funnel. It shows up through somebody who knows me thinking of me at the right moment.
What distribution looks like in a small town
I run my business from St. Thomas, Ontario. It is not San Francisco. There is no venture ecosystem. The chamber dinner is the quarterly event that matters. Half of the introductions I've made in the last year happened because somebody saw me at an event, remembered a conversation from three months earlier, and walked a stranger over to meet me.
That entire motion runs on something AI cannot do for you. Somebody in your community has to think of you when a need comes up. They will not think of you because of a search-engine ranking. They will think of you because you showed up to the thing. Because you followed through when nobody was watching. Because you were useful once, for free, and they remembered.
In a small town that stuff doesn't compound in months. It compounds in years. It is also very hard to fake, because the community knows who actually shows up and who just posts about showing up.
Why this isn't a small-town quirk
You might read the above and think it only applies to places like St. Thomas. I don't think that's true.
Every industry has a version of the small town. Real estate brokers have their regional board. Contractors have their suppliers and their site-visit chain. Accountants have the two or three lawyers and bankers who refer them. Even in big cities, the real distribution for a service business is a network of maybe fifty people who trust you enough to hand you their clients. That network is your moat. AI does not get you in. It might help you serve the network better once you are in.
The technology flattens the product. The relationship is what survives.
AI scales the work, not the relationship
Here's where I think Dheer's essay lands particularly hard for people like me.
My AI infrastructure lets me do more for the people I already know. It drafts faster. It researches faster. It lets me respond at three in the morning when I happen to be up, which the person on the other end notices. The thing it does not do is make strangers trust me. That still takes time, reliability, and showing up.
I've watched operators try to skip that step. They point an AI at a list of strangers and fire. They scale the bad version of outreach. The response rate does not go up, because the bottleneck was never "send more." It was always "be worth receiving."
If you are going to scale anything with AI, scale the thing your existing network already wants more of. Scale the followups you usually forget. Scale the research they asked you to do. Scale the second draft of the proposal so you have time to have coffee with them before you send it. That is AI serving distribution. The other way around does not work.
An agent whose only job is to give
A couple of nights ago I built a small agent I named Gabriel. His job is to find one non-profit or community organization in the region each day, research one concrete thing they need, and quietly help. No pitch. No follow-up. No tracking. If the help gets used, good. If it doesn't, also good.
I did not build Gabriel as a distribution strategy. I built him because it felt right. But the more I sit with it, the more I notice that Gabriel is the only piece of software I've ever built that maps cleanly onto the moat Dheer is talking about. He is not scaling outreach. He is scaling showing up. Those are not the same thing, and the small-town version of me knows exactly which one actually moves the needle.
What I take from all of this
If your work depends on a stream of new humans trusting you, AI is not your moat. Your moat is the week you spent helping someone for free, the year you spent showing up to events, and the time you picked up the phone when you didn't feel like it.
AI can make every one of those things less expensive in time and energy. That is real. Just don't confuse the force multiplier with the thing it is multiplying. Take the cheaper time AI gives you and put it into the relationship layer. That is the one part nobody else can replicate.
← Back to all posts