Why now
The supply layer is being built in 2026: ~63%3 of all dated walled-garden activity lands in 2026 H1, and the curve is still climbing. Yet the agency layer is open in every one of the 333 mapped markets except France — Orpi’s Kleio is the only network-scale exception. And access technology is commoditizing (Web Bot Auth is becoming free plumbing), so the durable moat is licensed, neutral supply.
³ Placy studies — walled-garden census and agent-network map: ~88 portals · ~330 networks · 33 countries, 2026.
If Placy becomes the neutral, licensed, agent-direct structured-supply layer for European real estate — listings sourced from agencies, published as one open contract, served through an MCP registry with verified identity and provenance — then it wins a defensible share of the agent-readable supply layer by 2027. Because supply is fragmented with no MLS, agencies are motivated to bypass portals, AI-search demand is already routing to whoever exposes structured listings, and access tech is commoditizing — so the durable moat is licensed neutral supply, not technology.
Isn’t this just Kleio?
Kleio proves the model works at network scale — and we say so above. But it is single-vendor, welded to Orpi, and France-only. Placy is neutral and cross-network, and the agency layer is open everywhere Kleio isn’t. Being neutral is the one thing it structurally can’t be.
Won’t the incumbents do it themselves?
They can’t be the neutral layer. Each portal is one silo, and neutrality is against their interest — Zillow’s own ChatGPT app is engineered so the data never trains the model and every action routes home. An AI platform wants one rail, not fifty widgets.
Is there real demand?
It is here and measurable: idealista launched its own app inside ChatGPT4, Hemnet shipped AI conversational search5, and Rightmove is re-platforming search on Google Gemini6. The question is no longer whether demand exists — it is who owns the supply layer that serves it.
How big is this really?
We won’t sell the $300B–$1.4T “AI in real estate” figure — that’s TAM theater, bundling smart buildings and back-office AI. The honest anchor is global PropTech at ~$40–47B7, and the economic model is CarEdge’s: licensed listings, revenue from the qualified lead — a referral model, not ad impressions.
⁴ idealista newsroom, Mar 2026. ⁵ Hemnet Group, Jun 2026. ⁶ Google Cloud blog, 2026. ⁷ Global PropTech market, 2025 — clustered across four independent research firms.
What we don’t know yet
Two gaps, stated plainly. There is no independent, real-estate-specific benchmark for AI-referral conversion — we treat cross-vertical evidence as the leading indicator, nothing more. And European consumer data is thin: the US is a 2–4 year leading indicator, never a proxy for Europe.
How we’d know we’re wrong
We publish our kill criteria. If any of these holds, the thesis fails:
- AI platforms standardize on bilateral portal deals, and agencies don’t defect at scale.
- A single vendor or an MLS-style governance play captures the agent-direct layer first.
- Agent-direct supply proves too thin for AI platforms to prefer it over portals.
We test each of these deliberately and early.