Everyone believes. Almost nobody has shipped.
Only 27 percent of AEC firms use AI for automation, problem-solving, or decision-making, while 94 percent of the firms already using it plan to increase their investment. Skepticism would show up as abandonment, and the adopters are doubling down. If your firm sits in the other 73 percent, you have plenty of company. You are the curve.
The usual explanations are conservatism and thin margins, and I do not buy either one. This industry adopted laser levels, GPS grade control, and prefab quickly once each one paid for itself. Contractors adopt what works.
So here is what I think is really going on. Bad data was estimated to cost global construction $1.85 trillion in a single year, including nearly $89 billion in avoidable rework. Job site reality lives in photos nobody labeled, a superintendent's memory, and a text thread with forty people in it. You cannot point a model at that and expect an answer.
Your last AI pilot failed because your site was never captured in a form a machine could read. The model was fine.
That is the problem I work on
At Yembo I built ScanMyHome: walk a space with an ordinary phone and get back a measurement-grade 3D reconstruction, floor plan, dimensions, wall cutouts, and the objects in it identified. Capturing an uncontrolled space in bad light with someone's belongings in the way is the same engineering problem on a job site as in a living room, and I have spent a decade losing arguments with it and eventually winning some. Insurance carriers settle claims against those numbers.
That is the capture itself rather than a dashboard sitting on top of data somebody else collected, and it is the step your industry keeps skipping and then blaming the model for.
You are also short people. The industry needs something like 349,000 net new workers this year, and most firms cannot find them. That makes every avoidable site visit and every rework hour more expensive than it was two years ago, which is the business case, and notice it does not require anyone on your payroll to be replaced.
What does not carry over is everything downstream of the capture: means and methods, sequencing, code, and the judgment your superintendents apply when a wall opens up and the drawing was wrong. I am not going to pretend a model helps you there.
So the value I bring your team is calibration. When a vendor tells your firm what their computer vision will do next quarter, I can tell you from having built it whether that is a roadmap or a wish.
If your team wants to work rather than listen, the agentic workflows workshop runs leadership teams through their own processes, and my work on AI for interior design covers the same capture problem on the design side of a remodel.
