1/42
Looks like no tags are added yet.
Adil's goal for customer friction
Any sales question answered in five minutes now. By about 2028, buyers ask a voice assistant and buy through the conversation. His test: a buyer asks if a home has a bathtub or a shower, and if no one can answer in five minutes, that's a technology failure.
Zillow app in ChatGPT
Launched Oct 6, 2025 for U.S. users. New construction is only planned, and tours and agents are handled on Zillow.
Zillow connector for Claude
Announced Oct 19, 2026. Brings listings, home values and affordability tools into Claude, with next steps through Zillow. One secondary source says the tools are U.S.-only, and Zillow's page doesn't mention new construction.
Edgewise Eva
An AI sales agent for new-home leads that hands buyers to staff when ready. The Inman article about it is sponsored content written by Edgewise's CEO, with no performance numbers, so treat it as a vendor description.
SAMI AI
Calls and texts new inquiries within seconds, qualifies them on timeline, budget, location and intent, and routes them to sales staff. Found only through the vendor's own listing, with no results data.
Astro AI
A staff-facing assistant built on ChatGPT and New Home Star's training content, launched Feb 2, 2026. It answers sales professionals' questions, so label it a staff tool, not a buyer-facing one.
ResVR and Landmark Homes
ResVR is a Calgary company whose voice assistant (May 27, 2026) sits inside 3D show homes. Landmark Homes (Edmonton and Calgary) was its first full production deployment. We couldn't confirm Landmark uses the voice assistant itself.
llms.txt
A plain-text file that points AI agents to a site's key data. Morrison Homes (Calgary) publishes one pointing to its models and pricing page. It's the cheapest example of making listings AI-readable.
Data the buyer-facing tools depend on
Live inventory and price, plans and options, availability dates, buyer constraints, and contract terms. Adil's examples map onto it: live pricing, bathtub or shower, a $450,000 budget, 90 days left on a lease.
Zillow builder feed
Records for builder, subdivision, lot, plan and spec, scored on listing quality such as inventory, completeness and timeliness (spec v3.5, May 2022). It mirrors the "one ID per lot" foundation step. We found no statement that it powers the ChatGPT or Claude experiences.
Publishing listings in ChatGPT and Claude
Submit a connector. OpenAI opened submissions Dec 17, 2025, and Claude's directory is open to paid plans. The catch: being listed isn't the same as being surfaced, because users have to name or enable the app (our inference for lesser-known builders).
2025 builder mystery shop
None of 50 builders sent a personalized email within five minutes, and 22% never sent one. Also 20% replied within 30 minutes and 24% within four hours. Builders with online sales counselors did better. Consultancies ran it, and the coverage doesn't state the country.
Cotality 2026 trust survey
U.S. trust in AI to help find a home fell from 30% in 2025 to 16% in 2026. The same survey found 55% of homebuyers use generative AI monthly, 68% would verify AI output, and 44% would pay a human to verify it. The trust figure is U.S. only.
Moffatt v. Air Canada (2024)
A B.C. tribunal held the airline responsible for its chatbot's wrong answer ($650.88) as negligent misrepresentation. It's a small-claims decision, not binding precedent, and we found no Alberta case. Lesson: the assistant should quote only from live records.
E-signing a purchase agreement in Alberta
Generally yes by consent, but land documents that must be registered are excluded. RECA calls e-signed agreements valid, subject to new case law. Confirm specifics with counsel.
Selling a new home in Alberta
A new home can't be sold or offered without warranty coverage, and builders need a licence. New condos carry a 10-day cancellation right. We found no cooling-off right for other new homes, and the warranty act doesn't mandate deposit protection. Confirm with counsel.
Audit, Assist, Automate
A way to reduce human involvement step by step. Audit: AI flags, people decide. Assist: AI recommends, people approve. Automate: AI acts, people review exceptions. For Rohit, AI would draft a first reply and staff would approve it.
Local competitors
Landmark is ResVR's first production customer, Morrison publishes an llms.txt, and Trico and Hopewell use OpenHouse.ai. We didn't check Jayman, Cardel, Shane, Brookfield, Mattamy, Excel or Avi, so say "as far as we could find" about anyone else.
Rohit's gap and the closing line
Likely ahead on back-end data, behind on buyer-facing AI. Closing line: the assistant is the easy part; the data it reads from is the product. The back-end lead is our inference, so ask Adil to confirm it.
MCP (Model Context Protocol)
An open standard that lets an AI assistant plug into outside data and tools. Adil's version: a pipe between an AI agent and a database. His warning: if the data is messy, it's like giving a Chinese speaker an English book. Connectors for ChatGPT and Claude are built on it.
Connector
An app that links an AI assistant to a company's data, such as Zillow's listings. Claude calls them connectors, and ChatGPT calls them apps. This is the "connect" path on slide 5.
RAG (retrieval-augmented generation)
The AI first looks up relevant passages from your own documents or data, then answers from them, instead of guessing. Adil's fix for unreadable data. Morrison's llms.txt says its site is optimized for it.
Semantic model
Documented definitions and relationships, so a word like "sale" means one thing everywhere. An assistant can't answer reliably without one.
Three generations of AI
Gen 1: analytics on messy data. Gen 2: structured data plus a semantic model (Ask BuildBase). Gen 3: agents that act. The buying-by-conversation vision is Gen 3.
Agentic AI
AI that makes decisions and takes actions, such as sending a contract, instead of only answering. It's slide 5's 2028 goal, and nothing we found does this for purchases.
Human in the loop
A person reviews or approves what the AI does before it takes effect. Safe but slow. Adil says AI moves at the speed of the human in the loop. Our rule: a person approves every contract.
Prompt injection
A hidden instruction inside content the AI reads, which it may obey. Adil's example: a long email whose last line tells the AI to send a payment. It matters for any assistant that reads buyer emails or forms.
Hallucination
The AI states something false with confidence, such as a wrong price or feature. Under Moffatt, the company owns the wrong answer.
Interface-less
Adil's term for dropping the browse-and-filter website. The buyer talks to an assistant that picks up constraints (area, budget, lease end) from conversation. It's his 2028 vision.
Speed to lead
The time between a buyer's inquiry and the first meaningful reply. It's the basis of Adil's five-minute test, and the mystery shop measured it.
Online sales counselor (OSC)
A staff member who answers web inquiries before a buyer reaches the sales centre. 70% of surveyed builders had one in 2025, and those builders performed better.
Lead qualification
Finding out whether an inquiry is a real prospect, by asking about timeline, budget, location and intent. It's what SAMI automates.
Feed (direct feed)
A standard-format data stream that carries all of a builder's listing information. Update it once and it flows to every platform that reads it. Zillow's builder feed is an example.
Lot, plan, spec
Lot: the parcel. Plan: the home design. Spec: a home built ahead of a buyer, often move-in ready. These are the record types in Zillow's feed, and they link to the "one ID per lot" foundation step.
Negligent misrepresentation
Making a statement without taking reasonable care that it's accurate, where the listener relies on it and is harmed. It's the basis of the Moffatt v. Air Canada ruling.
Drip pricing
Advertising a lower price than the buyer ends up paying, by leaving out mandatory fixed fees. The Competition Act covers it. Our inference: relevant if an assistant quotes a base price.
Implied consent
Consent inferred from circumstances. Federal privacy law (PIPEDA) allows it, but Alberta's PIPA has no such concept and requires specific forms of consent. Our inference: collecting income or mortgage details through an assistant needs clear notice and consent.
CASL
Canada's anti-spam law. Commercial texts and emails need consent, sender identification and an unsubscribe option. Voice calls fall under separate CRTC rules. It applies to AI-initiated follow-up texts.
Electronic Transactions Act (Alberta)
Gives electronic records and signatures the same validity as paper, with exceptions such as land documents that must be registered. It decides whether a purchase agreement can be e-signed.
New Home Buyer Protection Act
Requires builders to hold a licence and every new home to have warranty coverage before it can be sold or offered. A check any buying-by-chat flow must respect.
RECA
The Real Estate Council of Alberta, which regulates real estate professionals. Its bulletin says e-signed purchase agreements are valid, subject to new case law. Selling your own property is exempt from the real estate licence requirement.
CREA DDF and Realtor.ca
DDF is CREA's permission-based system for distributing members' MLS listings. Realtor.ca is the public site aggregating those feeds. Our inference: a builder selling its own inventory may not be bound by it unless homes are MLS-listed.
Cooling-off period
A set window in which a buyer can cancel without penalty. It's 10 days for new Alberta condos, and we found none for other new homes.