A demo went viral on Twitter recently: an AI agent built on OpenClaw finds homes in the $500k–$1.2M range with no pool, renders a pool directly into the actual owner's backyard pulled from Google Maps, calculates the quote and estimated home value increase, then sends a physical postcard with a QR code. No human in the loop at any stage. The demo shows a specific address in Florida: 25,000 sq ft lot, $48,500 estimate, $37,500 projected value increase.
The clip is 11 seconds long, and the replies are debating whether it's staged. One skeptic found that the GitHub repo shown in the video doesn't actually exist. The author didn't respond to the criticism but did offer to build a "custom openclaw system for your business." Yeah, it's standard engagement bait. But!
I broke down every component of this system — and it turns out all of them exist and are available via API right now.
Breaking it down step by step:
- **Satellite imagery.** [Google Maps Static API](https://developers.google.com/maps/documentation/maps-static/overview). Pass an address, set maptype=satellite, get an overhead photo of the yard. $2 per thousand requests, first 10,000/month free.
- **Property data.** [PropStream](https://www.propstream.com/) ($99/mo) or [ATTOM Data API](https://www.attomdata.com/). 160M properties in the US, 165 filters: home value, lot size, purchase date, pool presence. Owner name and address from public records — no paid databases needed.
- **AI pool render.** [PoolSend](https://poolsend.com/) does this as a ready-made SaaS: drop in an address, get a photorealistic composite of a pool on the satellite image within a minute. $49/mo + $2.99 per postcard, 200+ construction companies in the US already use it. If you want to control the render yourself, you can use Flux via API with inpainting.
- **Print and mail.** [Lob API](https://www.lob.com/). 6×9 glossy postcard with QR code: under a dollar each, delivered in three business days. HTML templates, tracking, Zapier integration.
- **Orchestrator.** OpenClaw glues it all into an autonomous pipeline: the agent finds addresses, filters, triggers the render, sends to print. You pick the model: Claude Sonnet 4.6, GPT-5, Gemini 3.1 Pro. Any of them should handle it fine.
Cost per contact: $1–2. Potential deal size: $50k. Even if one postcard in five hundred converts into a sale, you spent $1,000 and made $50,000. 🤷🏻♂️
The same formula, different niches
While breaking all this down, a few more niches came to mind where the same "satellite + render + postcard" formula applies.
Solar panels. Rooftops without panels in sun-heavy regions. Render panels onto the real roof, calculate electricity savings. Aurora Solar already sells this as a SaaS, and Amazon showed how to detect roofs with and without panels using their Rekognition service.
Roofing. Satellites can see roof age, fading, debris. Render with a new roof + replacement quote and insurance discount. Deal size $10–30k.
Guest houses (ADUs). Find large lots with small homes, render an accessory dwelling unit in the yard, calculate rental income. Hot market in California.
Landscaping. A neglected yard turns into something nice with a single render. Deal size $5–50k, same wow factor.
EV charging. B2B play: large parking lots with no charging stations, render the stations, calculate charging revenue.
Same economics everywhere. A couple bucks per contact, and the potential deal is in the tens of thousands.
Not that I'm actually going to build any of this, of course.