What an AI assistant for real estate should actually do

Every real estate platform now ships an assistant. Most of them are the same object: a chat window, a language model, and a thin pipe to the product's database. Ask a question, get a paragraph. It demos beautifully. Then Tuesday happens — you're double-parked outside a walkthrough with a seller on the phone — and the chat window is exactly as useful as a browser tab always was, which is to say: not now.

We think the bar is higher, and that it's worth writing down. An assistant for this business has to clear five requirements.

1. Reach — it meets you where the work happens

Real estate is practiced in cars, driveways, kitchens, and hallways. An assistant that only exists inside an app is an assistant for your desk, and your desk is where you spend the least valuable hours of your day. The test we use is the driveway test: you've just signed a listing, you're standing next to your car, you have ninety seconds. Can you hand off real work — not a reminder, the work itself — from right there? Texting should count. A phone call should count.

2. Memory — one thread, not twenty sessions

If you told it something in March, it should know it in August. If you told it over text, it should know it when you call. Assistants that start every conversation blank make you the integration layer — you carry the context between sessions, which was the job you were trying to give away. Memory also has to be honest about its source: what a client actually said, not what a model supposes clients like them tend to say.

3. Hands — it finishes, it doesn't advise

"Here are some tips for launching your listing" is content, not assistance. The assistant should hold the same tools you do: draft the listing, stage the page, compose the brochure, match the buyers, book the showing against your real availability, file the lead with the source that brought it. If its answer to a request is a list of steps for you to take, it's a search engine wearing a name tag.

4. Taste — the output is client-ready, in your brand

In luxury, an almost-right brochure is a wrong brochure. Generated marketing has a look — the stock gradient, the template grid, the font that ships with everything — and your clients can see it too. The assistant has to compose in your typography, your palette, your voice, and it must never dress a specific home in a stock photograph. Where the facts matter — prices, comparables — they should come from real computation it can show, not fluent improvisation.

5. Restraint — your name never moves without you

The more capable the hands, the more this matters. Client messages should stage as drafts; the send belongs to you. Pages should wait for approval before publishing. Anything consequential should be confirmed, not assumed. An assistant that invents a testimonial or fires off an unreviewed reply under your signature isn't saving you time; it's spending your reputation.

We built Elle against these five, and the order matters: reach first, because an assistant you can't reach is an assistant you won't use. She has her own phone line — text her photos from the driveway, call her from the car — and one memory across text, voice, chat, and the iPhone app. She holds every tool the platform has, she composes in your brand, and she is structurally incapable of some mistakes: she cannot send a client email, publish your page without approval, or produce a valuation a deterministic engine didn't compute.

Whatever platform you choose, keep the test. Stand in the driveway, give it ninety seconds, and see what's finished when you walk back in the door.