Written by Olivia Nielsen, Principal, Miyamoto International
Over the holidays, I did something that would have felt reckless not long ago.
Instead of carefully planning a data architecture, writing specifications, and lining up weeks of cleaning and validation, I vibe-coded a global housing database. I used Claude Code to pull together housing data from UN-Habitat, the World Bank, OECD, CAHF, and several other sources and let the structure emerge as the data came together.
It worked.
What normally takes weeks, sometimes months, of assembling spreadsheets, reconciling definitions, and rebuilding tables happened in a fraction of the time. Out of that experiment came the Global Housing Database: an interactive, visual platform covering 91 countries and 19 housing metrics, now live as a 3D website.
This isn’t a story about AI replacing expertise. It’s a story about AI removing friction.
For anyone who works in housing, too much time is spent doing the same thing over and over again: hunting for data that already exists, translating formats, rebuilding the same tables, and debating whose numbers are “right.” Getting the data in the first place is still hard. But once it’s there, AI changes the balance. It lets us stop re-assembling the puzzle and start asking what the picture actually shows.
And once the data was finally in one place, some patterns became impossible to ignore.
When “Just Build More” Stops Making Sense
Housing debates often default to a simple answer: just build more. But the data shows how quickly that logic breaks down when real constraints are extreme.
It’s hard to “build more” when:
- Land costs 99 months of income per square meter,
- Cement costs five days of income per bag, or
- Building codes haven’t been meaningfully updated since 1925.
In these contexts, supply doesn’t respond the way theory predicts. Markets don’t self-correct. Informal settlements expand not because people prefer them, but because formal housing is structurally out of reach. Households adapt rationally to systems that are misaligned with reality.
AI didn’t invent these constraints. But it made them visible—quickly, and across countries—without waiting months for a bespoke study.
What Actually Moves the Needle
When you step back from slogans, the data starts to tell a more nuanced—but still consistent—story about what tends to work.
Countries with modern building codes, updated every one to three years, tend to have much lower rates of informal housing—often under ten percent. This doesn’t mean that updating codes automatically eliminates informality, or that codes alone drive outcomes. Housing systems are complex, and many forces move at once. But across regions and income levels, countries that keep codes current rarely see informality spiral out of control.
Similarly, where construction materials are affordable, the housing sector is more likely to function as a job engine, generating more than 70 construction jobs per 1,000 people. Affordable materials don’t guarantee employment growth, but when basic inputs like cement are priced far out of reach, construction activity almost always stalls. The relationship appears too consistently to ignore.
The same pattern shows up with social housing. Maintaining a significant social or rental housing sector—above roughly twenty percent of the stock—does not by itself “solve” housing shortages. But countries that sustain this balance are far less likely to experience extreme housing mismatch. And where housing policy performance is strong, deficits don’t just slow; they often begin to reverse over time.
None of these relationships should be read as simple cause-and-effect formulas. They are correlations, not proofs. But when the same patterns repeat across dozens of countries, they stop looking like noise. They act as directional signals, pointing to the kinds of policy choices that make certain outcomes more likely, and others far harder to avoid.
The Housing Death Spiral
One pattern stood out more than the rest.
In countries that consistently prioritize luxury housing over affordable housing, several outcomes move together with striking regularity. Housing deficits increase. Informal settlements expand. Cost burdens deepen. The correlations are extremely high.
This does not mean that luxury housing alone causes informality or cost stress. But it does suggest a reinforcing dynamic. Once land, finance, and regulation are aligned toward the top of the market, affordability doesn’t just fall behind, it becomes structurally difficult to reintroduce. The system begins to feed on itself.
It’s what I’ve started calling a housing death spiral. And while the term is provocative, the implication is practical: the longer corrective policies are delayed, the harder and more expensive it becomes to change course.
The hopeful counterpoint is that the data also shows the spiral is not inevitable. Countries that intervene earlier (with land policy, updated codes, rental and social housing, and materials policy) are far more likely to avoid the worst outcomes. In that sense, the housing crisis is rarely an accident. It is usually the cumulative result of policy choices made, or deferred, over time.
Imperfect on Purpose
It’s important to be clear about what the Global Housing Database is and what it is not.
The website is still a work in progress. Some of the indicators, such as housing mismatch or policy achievement, are ones I developed myself. They involve judgment. They are, by definition, somewhat subjective. But housing systems themselves are messy. Waiting for perfectly objective indicators before engaging seriously with these questions is one reason the debate stays stuck.
The goal of these metrics is not to provide final answers. It’s to make patterns visible, invite challenge, and improve the conversation. AI helps not by eliminating subjectivity, but by making it easier to test ideas, refine assumptions, and iterate quickly. What once would have taken years to prototype can now be built, debated, and improved in weeks.
Countries Aren’t Housing Markets, Cities Are
Working at the country level also highlights the database’s biggest limitation: countries are not housing markets, cities are.
National averages hide enormous variation. Manila is not rural Philippines. Lagos is not Nigeria. São Paulo is not Brazil. Ulaanbaatar is not Mongolia. Housing outcomes are shaped at the city and neighborhood level, where land prices bind, zoning matters, and informality grows.
The next step is to move downward and inward to cities, metropolitan regions, and neighborhoods. That’s harder. But for the first time, it’s realistic. AI makes it possible to integrate municipal data, satellite imagery, local regulations, and subnational finance at scale.
That’s where housing policy becomes real.
What Vibe Coding Really Changed
AI didn’t solve housing. It didn’t tell me what to believe. And it certainly didn’t remove the need for judgment.
What it did was give me time back.
Time to stop rebuilding the same datasets. Time to question assumptions. Time to focus on policy choices instead of data archaeology. Time to spend on the hard questions, the ones that actually determine whether people can afford a safe, decent home.
If we use AI well (if we vibe code responsibly) it won’t give us easy answers. But it will make it much harder to pretend we don’t know what works.
Explore the (still evolving) Global Housing Database and 3D visualization here.
