I expanded HomeScout's rental coverage and it changed how the product feels
I have been rebuilding HomeScout from "better Dublin rental search" into more of an AI rental agent.
The latest useful update is coverage. The product is now tracking 900+ active rental listings across multiple sources, instead of feeling like a thin wrapper around one search site.
That sounds like a boring backend/data update, but it changes the product quite a lot.
When coverage is weak, AI search is mostly a nice interface. When coverage is broad enough, the agent can actually start doing useful work:
- compare more listings against the same renter brief
- surface better-value places that would otherwise be missed
- avoid making users repeat the same search across multiple sites
- draft inquiry emails from a more complete shortlist
- make the area/rent guidance feel less theoretical
The product direction is still deliberately conservative. I do not want an AI rental product that blindly applies to homes or sends messages without approval. The better version is: monitor the market, rank the options, explain the tradeoffs, draft the admin, then let the renter decide.
Current stack is boring on purpose: Next.js, Convex, FastAPI, Better Auth, Resend, Mollie, and a lot of glue code around listing normalization.
The unsexy work is the hard part: stale listings, duplicate properties, inconsistent addresses, price normalization, and deciding how much confidence to show users when the underlying market data is messy.
Product for context: https://homescout.io
Curious how other founders handle this kind of product boundary. When you add AI to a workflow with real-world consequences, do you push toward full automation, or keep the product approval-based for longer?
This is a strong direction because the approval boundary is probably the trust layer, not a limitation.
For rentals, full automation can sound impressive technically, but the user risk is high. Nobody wants an AI applying to the wrong place, messaging landlords too aggressively, or making decisions from messy listing data.
The stronger frame may be less “AI rental agent” and more “rental search co-pilot that watches the market, ranks the tradeoffs, and drafts the admin while the renter stays in control.”
That keeps the AI useful without making it feel reckless.
I’d also test this first with renters who feel the pain daily: people relocating, busy professionals, or students searching across too many sites. They already understand the cost of stale listings, duplicate properties, and missed opportunities.
If useful, I can put together a short written breakdown on the positioning, first renter segment, trust angle, and a simple 7-day user acquisition plan for HomeScout.