"Your competitors just bought the same AI as you. Now what?"
Last week a post did the rounds on LinkedIn claiming the moat in AI isn't any single tool — it's stacking a handful of them together into a system that runs itself. A hundred-odd replies later, something unusual had happened: everyone agreed with the premise. Tools aren't the moat. The entire fight was over what is.
Three camps formed. One said the moat is architecture — tools get swapped out, but a well-designed operating model survives the swapping. Another said it's engineering discipline — one commenter pointed to a real security hole in AI-generated apps that exposed the data of more than 170 businesses, and asked the question demos never answer: who's awake at 2am when the automation chain snaps? The third camp said the tech was never the bottleneck at all. AI made building cheap. It didn't make customers cheap, and it didn't make a business ready to absorb what the tools produce.
Here's what struck me reading it. Every person in that thread is a builder, arguing about builders' moats. And every moat they proposed can be bought. Architecture is hireable — any competent firm will design you a durable one, and the same firm will design one for your competitor next quarter. Engineering discipline is hireable too; that's exactly what good contractors sell. Even distribution can be rented for a while, if you're prepared to pay for it. The whole debate was people with the same skills arguing over which of their skills matters most — and missing that a skill anyone can hire is not a moat for anyone.
Now bring it back to a business like yours — the family operation running showroom, factory, field and install under one roof. You've been watching this AI wave with a fair question: if all this capability is suddenly cheap, and the roll-up down the road can buy the same tools we can, where does that leave us?
It leaves you holding the one input that entire thread forgot to mention.
Every system those builders were arguing about is an empty machine until something goes into it. The stack is a commodity — they're selling it to everyone, that's the business model. The architecture is a service. The discipline is a day rate. What's none of those things is the twenty-odd years of judgment that prices your jobs, reads your builders, knows which supplier's lead time to believe and which customer is worth sharpening the pencil for. That's the input. Nobody in that comment thread can sell it, hire it, or generate it — because it doesn't exist anywhere except inside your operation.
The filter, if you want one: if a competitor can buy it, it's not your moat. Run that over everything an AI vendor shows you this year.
The tool — buyable by definition; it's for sale, that's why they're in the room. The build — buyable; someone built it for you, someone can build it for them. The judgment and the data that go in — that's the first thing on the list no competitor can purchase at any price. Which means the strategic question was never "which AI tool should we buy." It's "what do we know that nobody can subscribe to — and have we captured it anywhere?"
One thing that thread did get right, and it's worth taking seriously. The engineering camp's warning is real. AI-built systems break — quietly, in ways the demo never shows — and somebody has to be responsible when they do. For an operation like yours, that's not a reason to sit this out. It's the reason the owner's sign-off stays in the loop. A system built from your judgment, with you approving what it puts out until it's earned your trust, fails safe. A cheap automation nobody in the building understands fails at 2am.
The builders will keep arguing about the machine. They're good at machines — it's their trade. But the machine was never the scarce part. What's scarce is what you'd feed it. And right now, in most family operations, that's written down nowhere, held by one or two people, and walking out the door the day they do.
I've built a short assessment for multi-channel family operations working out where AI actually helps — and which part of their edge is worth capturing first. Ten questions. The readout comes from me, not a machine.
