In one line: AI has moved from novelty to every coffee break in construction. The decisive question is no longer whether the tools work, it's where your project data goes once you use them. For most BIM teams, that answer just moved from IT's desk to theirs.
“Everyone in the room wanted AI. Nobody wanted to upload to it.”, BIM Coordinator, Stuttgart
A year ago, AI wasn't really on the agenda in our rooms. This past month it ate every coffee break, and the question that follows about ten seconds after any demo isn't whether the tools work. It's where your data ends up once you start using them. For most teams, that answer just moved from IT's desk to yours.
This past month the Catenda team has been in Paris, Vilnius and Stuttgart: on the show floor at BIM World, in a packed room in Lithuania, and 150 metres up on the deck of the Stuttgart TV Tower. The rooms and the drinks were different at each event, but one conversation kept circling back wherever we went.
Welcome back to Common Ground. As before, this newsletter exists to share what we hear from BIM practitioners across Europe, rather than to promote what we do.
People weren't against AI a year ago; most just hadn't seen anything concrete enough to form a real opinion either way. Now the conversation has split.
Some teams, usually those with strong digital maturity, are asking sharp questions:
The other teams are in a different place. They've seen the demos and know something is coming, but they don't yet know how it touches their week, and so they're unsure what to ask.
Where does the data go? A year ago that question went in a helpdesk ticket. Today it sits on the BIM Manager's desk, the answer is yours to know, not IT's.
For general contractors there's a second layer: the real question isn't where you can put project data, it's where the client will allow it to go.
When evaluating any AI integration, four questions come first:
The first three are table stakes. The fourth is the one that usually decides whether a pilot ships. For general contractors under strict confidentiality terms, it often settles what's possible before any technical evaluation begins.
Most contracts don't yet address AI processing at all, in which case the safer default is to assume nothing is permitted unless explicitly allowed.
This part rarely shows up in the demo, but it's what decides whether a pilot ever leaves the sandbox. If a vendor can't answer any of those four questions clearly, treat that as a no until you have a clearer answer.
The pattern we're seeing teams settle on is Model Context Protocol (MCP), at least for now.
What is MCP? Model Context Protocol is an open standard that lets you connect AI tools to your existing systems, like your CDE. It means you can pick the model already approved by IT and by clients, whether that's Copilot, Claude, Gemini or something else. It sits naturally alongside how we think about BIM & data coordination and the open standards behind the open way.
The point is control: MCP separates the model from the connection, so the model becomes a swappable component rather than a lock-in decision.
Last year it was OpenAI, this year it's all about Claude. Whichever model tops the benchmarks this quarter might not next quarter. That's not a reason to wait, it's a reason to stop shopping.
The leading models are all improving quickly, but the gaps between them are narrower than the gap between having IT approval and not having it. If your organisation has already cleared an enterprise agreement with one provider, use it. Six months lobbying for a second approval because it scores two percent higher is six months you could have spent learning what AI actually does to your workflows.
The tools are being demoed and piloted widely, but adoption is gated less by capability than by data governance. Across recent BIM events in Paris, Vilnius and Stuttgart, the recurring blocker wasn't whether AI works, it was uncertainty about where project data goes once it's used.
Four questions come first: where the data is processed, who controls the model, what happens to your inputs after the session ends, and what the client's contract permits. If a vendor can't answer any of these clearly, treat it as a no until they can.
It has shifted from IT to the BIM Manager. A year ago it was a helpdesk ticket; today it sits with the person running the models. For general contractors, the binding constraint is often the client's contract rather than internal policy.
MCP is an open standard for connecting AI tools to existing systems such as a common data environment (CDE). It lets teams plug in whichever model is already approved by IT and clients, Copilot, Claude, Gemini or another, without rebuilding the integration each time the model changes.
No. The gaps between leading models are narrower than the gap between having IT approval and not. If your organisation has already approved one provider, use it, and spend the time learning what AI does to your workflows instead of lobbying for a marginally higher-scoring alternative.
BIM rollouts rarely fail at the start. They stall later, in the dip where the work has increased but the payoff hasn't landed yet. Why adoption stalls, and two field-tested moves to get past it: protect your champions and engineer the 'aha' moments.
Read the articleIn March 2022 Catenda (Norway), Ingecid (Spain) and Createc (UK) were awarded the large and long-term deal to provide software and services for decommissioning and waste management to Norwegian Nuclear Decommissioning (NND). Together the three companies created the consortium ICCircle.
Read the article