It shouldn't just be a chat box. Each direction below is a different answer to "what is the home screen of an AI buyer's agent?" — built to express what the agent can truly do with our data: appraise a home against sold comps (the moat), shortlist against your brief, compare candidates, read a suburb, surface early agency-direct finds, prep an auction. Real Eastern-Suburbs Sydney content throughout.
Opens on what your agent did for you — a personal analyst's morning note. Early agency-direct finds, price drops, "the comps moved," auction reminders. Pushes rationed to strong fits. Chat demoted to a follow-up.
The agent renders interactive widgets, never a wall of text — appraisal cards, a compare matrix, suburb panels, a map — that you pin, rearrange and share. A command bar drives it. Artifacts you keep, not a transcript that scrolls away.
Your brief is a standing mission. Watch the agent hunt the whole market, see what it did ("scanned 214 → shortlisted 8 → culled 30 as overpriced"), approve its proposals, and trace why. An autonomous-analyst dashboard.
Lead with the moat. The hero is The Appraisal Card: a verdict chip ("~10% over comps"), the fair-value band, a confidence badge, "see the comps." Each home is a deal you judge — Buy / Pass / Ask. Numbers computed, agent narrates.
Talk like a phone call — but voice is multimodal, not voice-only. As the agent speaks, tappable artifacts surface in sync (appraisal card, comp chart, map) so you can audit later. Built for the moment outside an open home.
Each variant is a self-contained HTML file in a ~390×844 phone frame, with a design-notes toggle (real vs faked), an "agent working" state, an honest-gap element, and a provenance/citation surface. Built by 5 parallel Opus subagents from a shared brief + research pass. Placeholder imagery via picsum.photos.