AI, Inventory, and Infinite Interfaces
I've been experimenting with real MLS listings inside Claude through early connectors, exploring what's possible when listings are no longer locked inside a portal.
Why it matters: For 20 years the portal has been both where listings live and how people search them, and AI is the biggest opportunity yet to separate inventory from the interface.
AI opens the door to an infinite number of customized interfaces – a portal knows your clicks, but Claude knows your life.
The prototype below is a working proof of concept: it runs on early, non-public connectors into a few MLSs – a plug that lets an AI tool read listing data directly.
For example, Claude already knows what kind of bike rides I like, the types of coffee shops I prefer, that I keep booking Airbnbs on the water, and what I can afford, so my search starts there with no extra input.
Agents can hand each buyer their own custom portal: a bespoke, disposable site that only AI coding makes possible.
A real estate agent meets with a buyer, learns their specific needs, and by that afternoon has a custom portal built for them – using it until they buy a home and never opening it again.
This is another real, working example, built for a family moving from Los Angeles: neighborhoods ranked by drive time to the lake, tennis, and their daughter's riding barns, plus lot sizes for the garden, school-run times at 7:45 a.m., and the LA neighborhood each one feels like.
In this prototype, if you ask Claude for the next step on a property you're interested in, it puts the listing agent front and center – the your listing, your lead fight in a new place.
This is what happens when the MLS supplies the listings directly, and it cuts against business models built on selling that buyer lead to someone else.
It appears we're moving toward a world where AI is like the Star Trek communicator, with AI assistants starting to run more of the day.
People start with one interface, tell it what they want, and it tries to deliver.
The key insight is that there is one communicator and there are no apps: no one on the Enterprise opens a transporter, calendar, or CRM app – it all just happens, and a home search is just another request.
Portals now have to contend with AI, and Zillow made the first move this month by putting its own listing display inside Claude.
Every search routes back to the portal's tour, agent, or loan officer, so it keeps the lead even though the search now happens inside Claude.
Most people just want to see houses, not build apps, and that's the portal's edge: one portal connector shows every market, while each MLS connector covers only its own, for now.
A small group – including Bright MLS, Unlock MLS in Austin, RealPlus in New York, and Flexmls – is actively working on this, starting with the less exciting but essential foundation: data protection, licensing, and governance, all of which have to be solved before agents get access, let alone consumers.
The bottom line: This is an experiment, and it's meant to show what's possible.
It's early, but the ideas are radically new – infinite interfaces, disposable portals, and a search that already knows you – all rolling up into one AI assistant that becomes the place you start.
And finally: MLSs and agents get the second chance they missed in the early 2000s, because whoever controls the inventory decides what gets built on top of it, and by whom.