The tech press is excited about AI transforming construction. New companies, new capital, new tools that will finally fix the industry’s productivity problem.

I’ve heard this before. I was there for the last one.

The BIM promise

Starting around 2007, the AEC industry made a serious, coordinated push into Building Information Modeling. Not just Revit as a drafting tool. BIM as a methodology. The promise: model the building in three dimensions, connect all the systems, coordinate across disciplines in a shared data environment, and the waste comes out.

The investment was real. Firms trained people, rebuilt workflows, migrated project libraries. Owners started mandating BIM deliverables. The federal government issued requirements. The UK mandated Level 2 BIM for all public sector projects by 2016. Billions of dollars went into adoption.

The technology worked. Models got richer. Clash detection improved. Coordination inside individual firms got measurably better.

The 85 percent budget overrun rate didn’t move.

The rework costs didn’t move. The coordination failures didn’t move. The same numbers that were circulating in 2010 at the height of BIM optimism are still circulating today, adjusted for inflation. More sophisticated software. Same project outcomes.

Why better tools didn’t fix outcomes

This is the question that matters: if coordination improved, if models got richer, if software got better, why did project-level failure rates stay stubbornly constant?

The answer runs against how technology investors are trained to think about the world.

The architect’s fee was fixed at schematic design. More coordination effort was pure cost to them, not margin improvement. The general contractor’s profit margin came from managing change orders, not eliminating them. Rework was a revenue event, not a failure event. The MEP consultant was billing hours. Speed and accuracy worked against their invoice.

Every party in the delivery chain was economically structured to optimize for their own position, not for project outcome. Better software gave each of them a better tool for doing exactly that. Architects produced richer documentation of a design scope that owners may not have fully understood. GCs used more sophisticated scheduling tools to build better paper trails for their claims.

The technology was neutral. The business model was not.

What’s different, and what isn’t

AI changes the calculus on specific problems that BIM-era tools couldn’t address. The ability to parse unstructured building data, reason across documents, and connect design decisions to downstream consequences is genuinely new. The processing of RFIs, submittals, and contract language at scale is real capability.

But AI doesn’t change the incentive structure.

If the MEP firm is still hired on a fixed fee, still protected by limitation-of-liability clauses, still delivering drawings that become the GC’s problem at bid, then automating their design workflow creates efficiency that gets captured by them, not by you. The technology is neutral. The business model is not.

What you should actually look for

The firms using AI in ways that will change your outcomes are building something specific: pricing models tied to outcomes, not hours. Accountability structures where their upside is connected to your project’s performance. Delivery arrangements where the information flows to the owner in real time, not filtered through the parties whose financial interests don’t always align with yours.

The technology observations in the current wave of AEC AI enthusiasm are accurate. The capability is real.

The lesson from BIM is that real capability plus misaligned incentives produces sophisticated tools used in service of existing business models. The software improves. The outcomes don’t.

Ask the AI companies, and the AEC firms using AI, how their economics change when your project performs better. The answer tells you everything.