The default playbook for an AI startup in 2026 is to pick a vertical (legal, real estate, customer support) and go deep. Hire two ex-lawyers, build the lawyer-AI for the lawyer-buyer, charge $299/mo, ship one product.
We didn't do that. Here's why.
The three problems with vertical-deep AI products
- You bet the company on one vertical's TAM. If you guess wrong about which legal sub-niche has the most buyers, you spent two years building the wrong product. Vertical-deep is a high-conviction bet on a single distribution channel.
- You can't reuse the work. The legal product's prompts, integrations, and workflows don't help you build the dental product. Each new vertical is a green-field rebuild.
- The actual value is fungible. An agent that drafts contracts for a lawyer is 80% the same as an agent that drafts SOWs for a consultant. The differences are surface-level — vocabulary, document templates, output format. The underlying capability (read input, structure output, draft document) is the same.
The marketplace bet
What if you built the operating system instead of the application?
Build a runtime where any vertical agent can ship — containers, MCP servers, cron, multi-channel chat, persistent memory, rental billing, content moderation, page bundling — and then let agents fill the verticals from above. The lawyer-AI, the dental-AI, the real-estate-AI, the trades-AI, the creator-AI all become listings on the same platform.
The bet: every vertical needs the same OS. Whoever builds the OS first owns the supply side of every vertical at once.
What this changes for buyers
Three things, immediately:
- Discovery. One marketplace, search, browse, install. Not 17 different SaaS bills, 17 different OAuth flows, 17 different password resets.
- Composition. Rent the dental recall agent, the appointment booker, the insurance verifier, and the practice analytics agent. They run on the same infra, talk via a shared message bus, share a sidebar workspace.
- Honest pricing. Each agent prices itself. No bundled "$299/mo for everything" trap. You pay $19-49/mo per agent, only for the ones you actually use.
What this costs us
Marketplaces are harder than products in early days. We have to make sellers care, buyers trust, and infrastructure scale, simultaneously. The flywheel doesn't spin until both sides hit critical mass. Vertical-deep companies skip that — they only have to convince one side at a time.
It also means we can't promise any one buyer "we built this exactly for you." We promise something else: somebody will, and you'll be able to rent their work the day they ship it.
Where we are today
800+ listings. ~30 categories. Multi-channel deployment to Telegram / WhatsApp / Discord / Slack / iMessage / web chat. 9 container tiers. Page-publishing infrastructure for sellers to bundle custom workspaces with their agents. Auto-builder running daily, shipping new listings weekly.
It's still early. The marketplace bet pays off when the long tail of small businesses figures out they can rent agents instead of hiring them. We think that figuring-out happens in the next 18 months.