The most over-built object in AI right now is the horizontal copilot — the assistant that promises to do everything for everyone and, as a result, does nothing in particular for anyone's actual business. When every team can assemble the same primitives on the same models, the horizontal layer commoditises, and the interesting question becomes: if not there, where does the durable value go? The market has answered, and the answer is unfashionable. It goes into boring verticals.
Horizontal is a crowded, shallow place
Follow the money and this stops being an opinion. In one 2026 tally of the agentic-AI market, vertical AI agents accounted for roughly 48% of deals and 55% of capital — the main investable surface, not a niche.1 The reason investors give is the one I keep coming back to: a vertical agent is not competing for a slice of the IT budget, it is competing for the labour line of the customer's P&L.1 Andreessen Horowitz's "AI inside" framing models up to a tenfold expansion in revenue per customer for vertical software that ships a credible agent, precisely because it moves from charging for a tool to delivering the work.2
A generic assistant saves an HVAC company a few hours of writing. A vertical agent that answers every inbound call, books the job at three in the morning, and chases the unconverted estimate generates real revenue the owner can see on a bank statement.3 Same underlying models. Completely different business.
The moat is the boring part
Here is why the horizontal players cannot simply reach down and take these markets. The value of a vertical agent lives in three layers the model alone does not have — the industry's data, the deep integrations into its systems, and the codified standard operating procedures of how the work is actually done — and those layers take tens of thousands of hours to build.3 A foundation model has no permissioned access to a specific clinic's schedule, a specific carrier's claims history, or a specific firm's playbook. The vertical company does, because it sits inside the workflow.3 The moat is unglamorous by construction: it is made of the exact tedious, domain-specific plumbing that a horizontal roadmap will never prioritise.
A dental front desk is a better business than a chatbot
Let me make this concrete with the least glamorous example I can find, which is also one we build into: the dental front desk. It sounds like nothing. It is a leaking bucket of money.
Industry estimates converge on an uncomfortable picture: the average practice misses somewhere between a quarter and a third of its incoming calls, and roughly one in three new-patient calls never reaches a person.4 When a new patient hits voicemail, only about one in seven leaves a message — the rest simply call the next practice on the list.4 Each of those missed new-patient calls is worth on the order of $850 in first-year revenue and several thousand in lifetime value, and across a year the leak can run to six figures for a single practice.4
No horizontal chatbot solves this, because the problem is not conversation — it is capturing a call, checking real availability, booking into the actual schedule, and following up, all wired into the specific systems a dental office runs. That is an owned outcome, not a chat window. And the market is wide open: roughly a third of dentists now use AI, but overwhelmingly for radiograph reading, while the American Dental Association's own account has front-office and administrative adoption still uneven, especially among the small and mid-sized practices that make up most of the field.5 The glamorous AI went to the X-ray. The money is sitting at the front desk. That gap is the whole thesis of Top Dentistry — fill every chair — and of running one focused company per vertical rather than a copilot for all of them.
Why the giants and the horizontals can't have it
A single dental practice is not a customer a billion-dollar forward-deployed org will ever embed with — it is too small, too fragmented, and it has no platform team to hand a deployment off to. That is the same tail I argued the hyperscalers structurally cannot serve in the land grab. The domain depth that makes a boring vertical unattractive to a horizontal roadmap is exactly what makes it defensible for a focused operator: the expensive machinery — the ingestion, the integrations, the telemetry, the senior team — transfers across verticals, while the domain knowledge that has to be earned does not.
Glamour and defensibility are inversely correlated here. The more boring the vertical sounds at a dinner party, the fewer well-funded teams are willing to do the unglamorous plumbing — and the plumbing is the moat.
Concretely, what owning this vertical looks like end to end: capture every call — including the one-in-three that go unanswered today and the after-hours ones that currently die in a voicemail box nobody checks — then check real availability, book straight into the schedule the practice actually runs on, follow up on the calls that did not close, and keep the reviews and the search presence healthy so the next patient finds the practice at all. Not a chatbot bolted onto a website; the whole loop from missed call to filled chair, wired into the specific systems a dental office already uses. That is the difference between a horizontal assistant that can talk about scheduling and a vertical one that is measured on whether the chair is full.
What this means if you are building one
Three sentences for anyone choosing where to point an agent. Pick the vertical that makes people's eyes glaze over, because glamour attracts competition and boredom repels it. Sell the owned outcome — the filled chair, the booked job, the cleared filing — not a chat window bolted onto the same problem. And treat the domain plumbing as the asset, not the overhead, because the codified SOPs and the wired-in integrations are the part a horizontal player will never bother to build and can never shortcut.
The whole services-as-software shift rewards depth over breadth, and depth has an address: one industry, one workflow, one outcome, done properly. The unglamorous wedge is not a consolation prize for teams that cannot build a horizontal platform. In this market it is the better business.
The honest caveat: "vertical" is not a magic word. A thin wrapper around a general model with an industry logo on it has no more moat than the horizontal copilot it is imitating — arguably less, because it also has a smaller market. The defensibility comes entirely from the depth: the permissioned data, the real integrations, the codified way the work is done, and the willingness to be measured on the outcome. Pick a boring vertical and do the shallow version, and you have simply found a smaller way to lose.