The end of checkbox accessibility
A new data layer is taking shape beneath the broken checkbox. And AI agents will make the gap impossible to ignore.
There's a field on Google Maps called "Wheelchair Accessible." It's a yes or a no. Sometimes it's filled in. Most of the time, it's wrong.
A friend of mine, a wheelchair user, told me he once drove forty minutes to a restaurant that Google marked as wheelchair accessible. When he got there, there were three steps at the entrance. No ramp. The host offered to "help carry him in."
That's not accessibility. That's a checkbox someone clicked in 2019 and never looked at again.
And here's the thing nobody in tech wants to admit: almost every accessibility solution on the market today works the same way. Binary. Yes or no. Accessible or not. A single field in a database that flattens a deeply physical, deeply personal experience into one bit of information. (For the real numbers on how broken the status quo is, see our piece on accessibility facts every diner should know.)
That's not a data problem. That's an intelligence problem.
And we're about to find out just how badly it's broken.
The Agent Age Is Coming. Accessibility Data Isn't Ready.
In the next two years, the way humans interact with information will fundamentally change. AI agents, running on frameworks like Claude's MCP protocol, embedded in apps and assistants, will increasingly make decisions on our behalf. They'll book restaurants, plan trips, recommend venues, and coordinate logistics.
When your AI assistant books dinner for you, it's going to pull from whatever data is available. Right now, the best it can do is read a binary checkbox from a mapping API and hope it's accurate.
For most people, a wrong restaurant recommendation means a mildly disappointing evening. For a wheelchair user, it means getting to the door and not being able to get inside.
The gap isn't that AI agents don't care about accessibility. The gap is that accessibility intelligence doesn't exist in a form agents can consume. There's no structured, scored accessibility data layer that an AI can query and trust. There's no API that returns "this restaurant has level entry, an accessible restroom, wide aisles, and a score of 84 out of 100 based on multi-source verification."
Until now.
Accessibility Is Not an App. It's a Data Layer.
I built ROLLIN because my father has FSHD, facioscapulohumeral muscular dystrophy. Taking him out to dinner shouldn't require a phone call, a Google Street View investigation, and a prayer. But it does. Every time.
Accessibility needs to become infrastructure. Not an app you download. A data layer.
The insight that changed everything for me wasn't "someone should build an accessibility app." It was this: accessibility needs to become infrastructure. Not an app you download. Not a directory you browse. A data layer that lives underneath every app, every agent, every platform that touches the physical world.
Think about how weather data works. You don't go to one weather app. Weather data is an intelligence layer, structured, scored, constantly updated, that powers thousands of applications. Your calendar uses it. Your airline uses it. Your smart home uses it.
Physical accessibility data should work the same way. Structured. Scored. Verified. Available everywhere.
That's what ROLLIN is.
What We Built
ROLLIN is an accessibility intelligence platform. We score 105,000+ restaurants across 16 states on a 0 to 100 scale using weighted physical factors: wheelchair entry, accessible restrooms, level pathways, wide aisles, parking, elevators, and more.
But the score is only the surface. Underneath it is a signal-processing engine that combines multiple data sources, applies trust-weighted community verification, and uses a graduated confidence system that refuses to inflate scores when data is incomplete. We'd rather show honest uncertainty than false confidence. (We wrote about the details of this approach in how ROLLIN learns from every visit.)
The platform operates on three layers:
Consumer. A free web app and an iOS app with an agentic concierge. The concierge adapts to how you move: three modes from proactive discovery to quiet planning. On-device photo AI analyzes entrance ramps, door widths, restroom features, lighting conditions, and surface quality using Apple Vision, entirely on the user's phone. No cloud processing. No privacy trade-off.
Developer. A REST API with six endpoints, a Python SDK auto-generated by Stainless, and an MCP server published on npm. AI assistants can query ROLLIN's accessibility data natively through the Model Context Protocol. When an agent searches for "wheelchair accessible Italian restaurant near Midtown with good lighting," it gets structured JSON back: scores, features, confidence levels, evidence sources.
Business. Embeddable widgets that let hotels show nearby accessible restaurants for their guests, and restaurants prove their own accessibility with a verified score badge. The data is the same data that powers the consumer app and the API, one source of truth.
How It Actually Works
Let me walk you through a real scenario.
A wheelchair user in Manhattan opens the ROLLIN app and types: "Accessible Italian near Times Square, no steps, good lighting."
Results come back scored and ranked, each with a 0 to 100 accessibility score, individual feature breakdowns, confidence levels, and the sources behind the data. She taps a result, sees exactly what to expect at the entrance, restroom, and interior, and books a table.
After the visit, she submits feedback, confirms the features are accurate, adds a photo of the entrance. That feedback enters the system weighted by her trust score. A new user's confirmation counts modestly. A verified contributor with dozens of accurate past submissions carries significantly more weight. When enough trusted signals agree, the venue's score recalculates automatically.
Now here's where it gets interesting.
A developer building a travel app for users with mobility needs calls the same API. They get the same scored data, the same feature-level detail, the same confidence metrics. They don't need to build their own accessibility dataset. They don't need to verify anything. They query ROLLIN and get intelligence back.
An AI agent planning a trip for someone does the same thing, through the MCP server. It queries accessibility scores, checks feature requirements against the user's stated needs, and makes a recommendation it can actually trust. Not a checkbox. Not a guess. Verified, weighted, scored intelligence.
The venue itself can embed a ROLLIN widget showing its score on its own website. A hotel can show guests which nearby restaurants are genuinely accessible. A delivery platform can surface accessibility data alongside menu items.
One data layer. Every use case.
Why Now
Three things are converging that make this the inflection point.
First, AI agents are becoming the primary interface between humans and information. The Model Context Protocol, function calling, tool use, these aren't experimental anymore. They're shipping in production. Every major AI lab is building agent frameworks. Those agents need structured data to make good decisions, and accessibility data doesn't exist in structured form anywhere else.
Second, the ADA is 35 years old and physical accessibility is still undocumented. There are 61 million Americans with disabilities. The accessible tourism market represents tens of billions in lost annual revenue. Every restaurant, hotel, and venue in America has physical accessibility characteristics. Almost none of them are documented in a way that's structured, verified, or queryable. That's not a niche problem. That's an infrastructure gap. (ROLLIN currently covers 16 states and 77 regions with 105,000+ scored locations.)
Third, community verification at scale is now possible. Trust-weighted feedback systems, on-device photo AI, and signal processing mean you don't need an army of inspectors to build an accurate accessibility dataset. You need a platform that turns every user interaction into a signal, weights it by trust, and compounds accuracy over time.
The data moat isn't the data itself. It's the verification layer on top of it.
The Invitation
If you're a developer building anything that touches the physical world (travel, dining, hospitality, navigation, local search), you can query ROLLIN's accessibility data today. Free API keys are available at joinrollin.com/portal. Documentation is at joinrollin.com/developers. The MCP server is on npm.
If you're a business owner, you can show your accessibility score on your own site. If it's high, that's a competitive advantage you're probably not using. If it's not, that's information that helps you improve.
If you're a disability advocate or accessibility practitioner, this data is yours. It was built for your community. Every score, every feature flag, every piece of community feedback exists because someone believed that wheelchair users deserve better than a checkbox.
And if you're building AI agents, this matters more than you think. The decisions your agents make will only be as good as the data they have access to. Right now, accessibility is a blind spot in every major AI system. ROLLIN is the fix.
We didn't build an app. We built the accessibility intelligence layer.
It's live. It's scored. It's verified.
And it's open for business.
