Construction's AI problem still starts with workflow

A new April piece in Construction Management makes a point the industry keeps trying to skip past. Construction's AI bottleneck is not a shortage of tools. It's the way information still moves, or fails to move, between office and site.

The argument is straightforward. Firms are planning to invest more in technology, but only a small minority describe themselves as fully digital across all project phases. Usage remains stronger in design than it is close to the site. That matters because the weak point is usually the handover. Drawings are updated in one place, marked up in another, issued by email, downloaded locally, printed on site, and then quietly drift out of sync.

That is not a small administrative issue. It is exactly the kind of disorder that makes AI outputs less useful and less trustworthy. If naming is inconsistent, versions are unclear, and decisions are poorly recorded, no amount of summarisation or automation will rescue the process.

The same article points to training as a practical weakness. Many firms still spend very little of their technology budget on helping teams understand when to use systems, why they matter, and what good looks like in practice. So the old habits survive.

One of the more useful April updates came from Re-flow, which is preparing a PAS 2080 module designed to capture carbon data through normal field operations workflows rather than through a separate reporting exercise.

That matters because carbon reporting is increasingly expected to be accurate, traceable, and tied to what actually happened on site. Re-flow's pitch is that if the platform already handles quotes, rates, shifts, forms, materials, travel, resource assignments, and evidence trails, it can act as the structured capture layer for carbon reporting as well.

In other words, carbon data becomes a by-product of everyday delivery rather than an after-the-fact admin burden. That's a better direction than building yet another isolated compliance process.

The interesting part is not the module itself. It's what it signals. Construction data requirements are converging. Quality, safety, delivery, and carbon are all starting to depend on the same operational record.

Trimble's Tekla 2026 release brings the sort of AI features the market now expects: an embedded Trimble Assistant, early natural language support for modelling operations, and AI-assisted fabrication drawing generation with a human still in the loop.

Useful features, yes. But the stronger signal is elsewhere. Tekla is still framing the value around connected workflows, real-time project status, linked models and drawings, shared metadata, and fewer black boxes between office, shop, and site.

That is worth noting because it is a more mature way to talk about AI in construction software. The message is not that AI replaces the delivery process. The message is that AI becomes more useful once the surrounding data flow is coherent enough to support it.

The out-of-tolerance workflow is a good example. If the system can flag discrepancies between the design model and as-built conditions before steel arrives on site, that is not just a nice digital feature. It is rework reduction in a form people can actually value.

Today's UK trades research from Sage is still worth keeping in view because it adds real behavioural context to the software stories. 76% of UK tradespeople now use AI tools daily. Builders are at 73%, electricians 71%, and plumbers 63%.

The practical uses are revealing. Builders are using AI to calculate material costs, support record keeping, assess problems, chase payments, and generate project ideas. This is not frontier AI. It's everyday problem solving.

That is probably the most important adoption lesson. The workforce does not need to be convinced that AI exists. People are already using it where it saves time. The open question is whether construction businesses give them tools and workflows that improve delivery, or leave them to improvise with disconnected consumer systems.

The wider AI market is moving toward agentic systems and workflow orchestration rather than standalone tools. That matters for construction because the same dependency keeps showing up. Agents are only as useful as the systems, data, and permissions around them.

This is why the digital twin and robot-ready site conversation matters, even when the live deployment story is still early. The value is not in having an impressive demo. It's in having project information that is clean enough for automation to act on safely and consistently.

That is a less glamorous story than a model release. But, it's the one that will decide who gets beyond pilots.