On StayOnMap today, owners create a listing themselves: photos, details and house rules. But many owners already describe their homes somewhere else first — in their own words, often in a few short messages with some photos in between.
A listing needs structure — rent, deposit, BHK, furnishing, location, amenities, rules — and a conversation rarely arrives in that shape. This experiment asks how much of the structuring an agent could do from such a message, and where it should stop.
The shape of the task
We treat it as one job owned end to end, which makes it a workflow agent rather than a single reply. The agent reads everything the owner has sent, plans what is still missing, asks only for those things, and assembles a draft.
Crucially, it does not publish. It hands the draft back to the owner with each field marked as either taken from their words or inferred. The owner confirms or corrects, and only then does the listing go live.
What we have noticed so far
The hard part is not extraction. It is knowing when a detail is genuinely absent rather than implied. An owner who does not mention a deposit may mean there isn't one, or may simply not have thought to say. Guessing either way produces a listing that looks complete and is wrong.
We have found it better for the agent to ask one clear question than to fill a field confidently. That makes the conversation slightly longer and the listing noticeably more honest.
Open questions
How many questions is too many before an owner gives up? Which fields are worth asking about, and which can safely stay blank? How should corrections from owners feed back into the next draft?
This is an experiment, not a finished feature. We will describe what ships when it ships.
