How AI Can Support Ireland's Housing Crisis
17,112 people are homeless. Rents are €2,200. Ireland needs 33,000 homes a year and isn't building them. AI doesn't fix that. But it does fix one specific blockage that has been hiding in plain sight.
By Faith Olopade
The numbers don't get easier the longer you stare at them.
17,112 people in emergency accommodation. Average rent in Dublin around €2,200 a month. The country needs roughly 33,000 new homes a year and is consistently delivering closer to 25-30,000. Planning permissions granted but unbuilt sit at multiples of the annual delivery target. Every party agrees there is a crisis. Nobody agrees on the cause.
We are not going to claim AI fixes any of this. AI does not pour concrete. AI does not own land. AI does not vote on a county development plan. The crisis is structural, political, and material, and the fixes are the same.
But there is a smaller, technical claim that we believe is true: a large fraction of the people whose job it is to build homes in Ireland cannot easily access the data they need to do that job well. Not because the data is secret. Because the data isn't usable.
Where AI actually helps
Every new home in Ireland starts with a planning application. The application has to argue, with reference to specific past decisions and specific policy text, that the proposed development should be granted. The people who write these applications — architects, planning consultants, agents, sometimes solicitors — spend a real fraction of their billable time doing one specific task: looking up similar past applications to see what was granted, what was refused, on what grounds.
This task is information retrieval. It is the task computers are best at. It is also, until very recently, a task that has been done in Ireland by hand, one council portal at a time, by professionals charging hundreds of euros an hour to do it.
That is the small specific thing AI can fix. Not the housing crisis. The research bottleneck inside the housing crisis.
When a planning consultant in Cork is preparing an application for a 12-unit residential infill site, the relevant precedent is not just other Cork applications. It is every comparable infill application in the country, the conditions attached to grants, the grounds cited in refusals, the appeal outcomes when refusals were challenged. The consultant needs to know the local development plan in detail, but also how policy has been interpreted in other authorities with similar zoning language. None of that information is hidden. All of it is technically public. None of it has been retrievable in under a few hours.
A retrieval system that has read every planning record across all 31 authorities, plus the development plan, can answer that consultant's research question in under thirty seconds. The consultant still does the planning work. The consultant still writes the application. The consultant still applies professional judgement about what is relevant to argue and what isn't. The thirty seconds vs the four hours is the difference between marginal applications being made well and marginal applications being made hastily because the research budget ran out.
If even five percent of marginal applications get prepared more thoroughly because the research is faster, the rate at which good applications get granted goes up. Not by much. But the housing crisis is a problem you only solve by stacking dozens of "not by much" wins.
The real blindspot
The framing we keep coming back to is that the data was never the problem. The problem was that the data didn't work.
A 2,000-page county development plan PDF is "available" in exactly the same sense that a manual for a piece of industrial machinery you don't own is available — yes, you could read it, but the cost of reading it for the one paragraph you need is so high that nobody does. So instead, every consultant in the country has internalised partial summaries of partial sections of partial plans, and the quality of any given application depends heavily on which partial summary the consultant happened to internalise.
This is the actual texture of the housing crisis at the document level. It is not that nobody is trying. It is that the people trying are doing knowledge work without a search index over their own knowledge base. AI changes that not by being smarter than the consultants but by being faster at the parts of the consultant's job that have always been mechanical look-up.
We have made the planning corpus and the Dublin City Development Plan searchable in this sense. Searching them is now a thirty-second task. Other development plans, the Bord Pleanála inspector reports, the building-control records, the derelict-sites registers — all of these have the same shape and will get the same treatment.
What we deliberately are not doing
We have been very explicit, internally and in the product, about the limits of this:
We do not write planning applications. The AI summarises retrieved records and quotes development-plan text. The application itself has to be written by a human who is professionally responsible for what it says.
We do not predict whether a given application will be granted or refused. The model is grounded only in past decisions; planning law is contextual and political; we have no business making forward predictions about future committee decisions, and we will not.
We do not anonymise or rewrite public records. Planning applications are public for a reason and we publish them as the council publishes them.
We do not replace planning consultants, architects, or solicitors. The product is a research tool. The professional judgement is theirs.
If anyone tells you AI is going to fix the housing crisis, they are selling something. The crisis is structural. The crisis will be fixed by structural change, not by software.
But the part of the crisis that lives at the document level — the friction caused by professionals being unable to access the data they need — is fixable, with software, today, and that work is worth doing on its own terms.
That is what we are doing. That is the only thing we are claiming to do.