A generic AI agent is the wrong shape for real estate
Every conversational platform now ships AI agent templates. They are almost always the same three: a receptionist, a sales agent, a support agent.
They have to be. A horizontal platform sells to a dental clinic, a driving school and a furniture shop, and the only persona that fits all three is one that fits none of them properly.
We only sell to one industry, so our templates can arrive already knowing it.
The difference is a choosing problem, not a quality problem
Put the two choices side by side.
A brokerage owner opening a generic platform picks between "Sales agent" and "Support agent", and then spends an afternoon writing a persona prompt, guessing at escalation keywords, and discovering three weeks later that nobody told the AI not to negotiate on price.
The same owner opening ours picks between "Listing enquiry concierge" and "Off-plan sales concierge — knows payment plans, handover dates, and books a call with your sales team".
That is not a feature comparison. It is the pitch, and it is structurally hard for a horizontal competitor to copy without building thirty vertical template packs.
What a template has to carry
A template that is only a persona paragraph is a snippet, not a template. Ours carry four things, and the last two are where the value is.
A persona, written for the domain rather than translated into it.
A mode. Off-plan runs in assist mode with booking off, because that conversation should end in a call with a human, not a calendar slot. The after-hours receptionist runs fully autonomous, deliberately — it exists to answer at 11pm, and a draft nobody approves until morning is the same as no answer.
Escalation keywords, in both languages. The shared safety floor is roughly lawyer, legal, complaint, refund, RERA, Dubai Land Department, and the Arabic equivalents. The per-template additions are where the judgement lives: an off-plan desk needs "cancel my booking" and "escrow"; a lettings desk needs "eviction" and "deposit dispute".
Suggested knowledge stubs — titles only, with placeholder bodies the tenant must fill. Commission structure, payment plans, viewing policy, areas covered. This is the one that gets argued about, and the argument matters: a stub the tenant edits beats a plausible fact the AI might quote. We would rather the concierge say "let me check with the team" than invent a service charge.
Applying a template does not switch anything on
The last design decision is the one we would defend hardest. Choosing a template prefills the settings form. It does not save, and it does not go live.
The owner reads the persona, reads the escalation keywords, edits whatever they want, and saves. Then they can run it in a sandbox that reads live data and writes nothing.
A template that quietly changed how your AI talks to your clients would be a faster onboarding and a much worse product.