AI can improve the work and challenge the housing thesis

Ask about demand as well as delivery

For a housing investor, AI creates two different questions. Could better tools reduce errors, improve coordination or shorten parts of the development process? And could changes in employment and wages alter the households that can afford to live in the finished homes?

One effect does not cancel the other. A more efficient delivery process can coexist with weaker tenant demand. A technology-positive outlook therefore needs a demand scenario as well as an operating thesis.

The evidence supports bounded conclusions

A 2025 study in The Quarterly Journal of Economics followed 5,172 customer-support agents at one software company. AI assistance increased issues resolved per hour by 15% on average, with gains varying across workers. That finding concerns a particular task and setting; it does not establish a construction saving, a rent increase or an economy-wide employment outcome.

The ILO’s 2025 occupational index measures exposure to generative AI, rather than observed jobs lost. Many occupations combine automatable tasks with work that still requires people. Exposure may bring transformation and productivity opportunities, while displacement remains a risk.

Evidence: [1] [2]

A job count is not enough to describe local demand

OECD research finds regional exposure varies with industry and occupational mix. A local economy with several employers can still have correlated income risk if those employers depend on similar work. Conversely, a high-exposure occupation may use AI to become more productive rather than disappear.

Our interpretation is that useful housing diligence should examine the employers, occupations and income sources supporting the expected resident group. It should also consider alternative employment access and how a weaker hiring environment would affect leasing or purchases. This is a framework for investigation, not a completed AI-risk ranking of Burlington, Manor Way or Friday Harbor.

Evidence: [3]

Separate infrastructure announcements from durable housing demand

AI infrastructure is another demand hypothesis to investigate. The analytical distinction is between temporary construction accommodation and households that remain after a facility opens. An announced investment amount does not tell us either group’s size, housing preferences or willingness to pay.

Before using that thesis to justify permanent homes, we would seek local evidence of workforce timing, commuting patterns, operating employment, competing accommodation and demand after construction. A housing plan should also test what happens when the temporary workforce leaves. This is a diligence framework, not a measured demand estimate or a claim about the three Washington project markets.

Test useful workflows without assuming the answer is correct

Document comparison, source-linked research and exception detection are promising workflows to test. Their value would come from verified outputs and reduced rework, not the volume of generated text. The right pilot compares accuracy, review time and errors with an existing process.

NIST’s generative-AI guidance identifies confidently incorrect output as a risk and calls for source verification and oversight. Consequential facts, calculations and decisions need accountable review. We describe this as an approach to evaluate, not a claim that a particular system is deployed or that errors can be eliminated.

Evidence: [4]