The AI tools are ready. But is our data? .

I’ve been exploring the latest wave of AI tools for construction, and honestly, the capabilities are impressive. We’re seeing platforms that can instantly spin up “Lessons Learned” applications from a simple email, or audit complex construction contracts in seconds to flag critical risks. On the surface, it looks like the future is already here.

But having spent most of my career on construction sites — building across mining, residential multifamily, and institutional projects — I know that “capabilities” in a demo don’t always translate to “value” in the field.

Construction isn’t Fintech. We don’t move at the speed of software; we move at the speed of concrete.

A former boss of mine used to say something that has stuck with me: “We build things meant to last 50 years, without major interventions. We can’t rely on technology that hasn’t been validated for at least five.”

It’s a smart, necessary conservatism. When your product has a 50-year lifecycle, you can’t treat innovation like a software update. You need to know — without a doubt — that a new method won’t generate unexpected warranty issues or maintenance costs a decade down the line.

That same logic applies to AI. The barrier right now isn’t the technology; it’s the data. In construction, data governance is often a structural mess: fragmented systems, inconsistent reporting, and a reliance on unstructured information. Until we fix the “garbage in,” even the most brilliant AI will only churn out “garbage out.”

I expect it will be at least another two years before we see AI delivering real, widespread returns in our sector. Not because the tools aren’t smart, but because we need time to get our house in order first.

Innovation is coming, but in construction, it pays to pour a solid foundation before you start framing the walls.

Related thinking on this site: AI ⨯ Project Management — Native Integration makes the same argument from the framework side; How Employee Behavior Could Shape Future AI Policies looks at the policy half of the same problem.

Some demos that prompted the post: The Thinking Game · DeepMind documentary · Construction AI demo