At a glance
~90%
less manual research work
2
countries live with developers, Canada and Japan
1,000s
of bylaw documents across hundreds of jurisdictions, ingested by agents
- Customer
- Architects Studio Japan (ASJ)
- Industry
- Real estate
- Challenge
- Knowing what a site can become meant assembling parcel, zoning, and bylaw records by hand, from sources that disagree
- Solution
- Permits AI: a multilingual, geo-aware AI agent, backed by research agents that find, ingest, and reason over government records
The challenge: knowing what a site can become before committing to it
Every development decision starts with the same question: what can be built here? For a developer weighing two sites for a six-storey rental building, the answer is spread across parcel records, zoning maps, property assessments, and municipal bylaws.
Height, density, setbacks, and rezoning requirements each have to be checked. A height limit on one page can carry an exception several pages later, with conditions that change what is possible. Then the whole exercise is repeated for the second site, and the findings compared.
That research is manual and slow, and it is repeated for every site. As a benchmark, one US architecture firm quotes 2 to 4 hours just to research the zoning on a single property, and 20 to 30 hours once a 3D study of the buildable envelope for an apartment building is added. It also doesn’t travel. Every city publishes its records differently, and every country more so.
For ASJ, that last point mattered most. ASJ saw AI as the way to transform how its network researches sites and to take its platform beyond Japan, and Japanese interest in Canadian AI expertise shaped where it looked for a partner. It needed one product that could do rezoning research in Canada and Japan, in English and Japanese, without rebuilding it for each place.
Why Wizard Labs
ASJ came to Wizard Labs through a trusted referral. One team of five people took the project from scoping to production, led by a Wizard Labs solutions architect.
The same team that modelled bylaws across Canada went on to tackle Japan’s land records, so nothing was lost in a handoff between strategy, data, and engineering.
The solution: AI agents that do the research, on a map
Permits AI puts parcels, zoning districts, and property values on an interactive map with an AI agent built in. A developer selects a site and asks, in plain language, what they could build there. The agent knows which parcel is on screen, searches the relevant bylaws and records, and answers with citations to the source documents.
Comparing sites
Comparing sites works the same way. Select a second parcel, and the agent weighs both against the developer’s goal, recommends one, and explains why. It can pan the map, open panels, and highlight parcels as it works, and it turns its findings into a PDF report the developer can share.
Research agents keep the records current
Behind the chat, a team of research agents does the work an analyst would otherwise do by hand. A supervisor agent directs one agent that searches government sources and another that verifies what it finds, then ingests the documents that pass. Across Canada and Japan, they have collected thousands of bylaw documents from hundreds of jurisdictions.
Sources can disagree. The team tested cases where an official zoning bylaw should outrank an outdated webpage, and tuned the agent to favour the authoritative source.
Japan: zoning maps drawn before mapping standards
The hardest problem came with Japan. Much of its zoning information predates international mapping standards. In major city cores, many maps are hand-drawn, each to its own scale, so they can’t simply be laid over a modern map.
Wizard Labs built a recurring ingestion process that fans out across hundreds of AI sub-agents. Each takes a region, retrieves its records, converts them into the platform’s coordinate system, and loads the result. That process now covers Japan nationwide, alongside land prices, transaction records, and lookups of national law, with the interface and agent working fully in Japanese.
See it in action
The results: rezoning research, automated in two countries
In production in Canada and Japan
Permits AI is in production. Property developers in Canada and Japan use it through ASJ to research sites before they commit to them.
Rezoning research automated end to end
Agents find, verify, and ingest the regulatory documents, and the chat agent reads them and answers with citations a developer can check. Based on interviews with real estate developers, Wizard Labs estimates this removes roughly 90% of the manual research work.
A second country, nationwide
Coverage spans all of Japan, built on new data models, four new government data sources, Japanese law lookup, localized maps and currency, and a Japanese-speaking agent.
Thousands of bylaw documents across hundreds of jurisdictions
Collected and ingested by research agents rather than by hand.
A full product, not a prototype
A team of about five delivered a multi-organization platform with access controls, API access, admin tooling, and production monitoring.
What’s next
ASJ plans to take Permits AI global. The platform was built for that. Each government data source plugs into the ingestion pipeline as its own module, and the agent and interface already work in more than one language. A new region means adding new sources, not building a new product.
