Big change, value to prove.

What the evidence covers: gains on single tasks; the effect on whole firms is still unclear.

My evidence read. Selected cases report faster work on specific tasks. Broad surveys of firms find much less evidence that productivity or jobs have actually changed. Faster drafting does not prove that finished, checked work takes less total effort.

Ask management. Which outputs are improving after checking and rework, and how much of the time saved turns into more delivery, better quality or lower cost?

What would change my view. Repeated gains in checked output per paid hour would strengthen this view. Widespread, sustained use without those gains would weaken it.

What the case evidence establishes →

What the evidence covers: little evidence yet of clients paying for it.

My evidence read. Reusable workflows and knowledge systems offer a plausible route. Evidence that clients repeatedly buy these as services, and that providers retain the benefit, is still thin.

Ask management. Which capability could we sell repeatedly, what would the client pay for, and does the offer make money after delivery and support?

What would change my view. Repeat sales, clear client value and sustained provider margins would support the hypothesis. Continued reliance on hours, or unprofitable offers, would challenge it.

Where reported value still falls short →

What the evidence covers: it is still unclear who keeps the savings.

My evidence read. Lower production costs could benefit the firm, its clients or its suppliers. The evidence does not yet show who will be able to hold on to prices. Owning data alone is no guarantee of an advantage.

Ask management. If production becomes cheaper, what protects our fee, client relationship and bargaining power? What happens if our supplier changes its terms?

What would change my view. Lasting margins and customers who find it hard to switch would support the view. Easy substitution, or competitors cutting prices until the gain is gone, would weaken it.

The limits of the current value evidence →

What the evidence covers: real examples, but in specific workflows.

My evidence read. Named cases describe reusable engineering checks and connected company information, alongside surveys showing wider use. Arcadis engineers build and check technical-review workflows; Alston's supplier describes questions across six connected systems. These support the direction of change. Counts of users and queries do not show that most firms have made the change, or what it is worth.

Ask management. Where have individual experiments become a repeatable way of working across teams, with an owner, common information and accepted outputs?

What would change my view. Sustained use across projects would strengthen the view. Repeated failure to move beyond individuals using their own AI assistants would weaken it.

Use, depth and organisational readiness →

What the evidence covers: firms are adopting at different speeds; a lasting advantage is unproven.

My evidence read. Readiness and adoption differ between firms. That is not yet evidence that AI has given some firms a lasting lead. Standard tools could also help slower adopters catch up.

Ask management. What should improve relative to comparable firms: delivery, quality, client retention or margin? Could competitors reproduce the gain easily?

What would change my view. Sustained outperformance linked to changed workflows would support the view. Rapid imitation or no retained benefit would challenge it.

What adoption differences can show →

What the evidence covers: a real risk, but little evidence yet of the outcome.

My evidence read. Routine work may disappear before new ways of learning are in place. Other accounts describe shared checklists and AI-assisted access to project knowledge helping staff learn, including Hill & Wilkinson's supplier case. Neither shows the long-term effect on skills or careers in the built environment. The effect depends on how firms redesign supervised learning; junior job losses are not a settled outcome.

Ask management. Which tasks teach the judgement we will need in five years? If AI takes them over, where will people practise, receive feedback and earn responsibility?

What would change my view. Juniors becoming less ready to take responsibility would strengthen the concern. Better skills and careers under redesigned training would weaken it.

How people and responsibilities change →

What the evidence covers: how roles change is clearer than whether they are better rewarded.

My evidence read. In the cases, professionals still check the work and stay responsible for it. Some also describe people configuring reusable checks, choosing schedule-recovery actions and teaching or supervising machines. This shows how roles might change, but not that those roles are in higher demand or better paid. Parts of expert judgement can also be automated.

Ask management. Which decisions need accountable expertise, which tasks can move to AI, and what skills and authority will our people need?

What would change my view. Sustained changes in responsibilities and client demand would support the view. Little role change, or wider automation of those activities, would challenge it.

How work and roles fit together →

What the evidence covers: firms adopt despite obstacles; the pace is less certain.

My evidence read. Surveys show use growing alongside readiness gaps; cases show adoption in particular workflows. Data, liability and physical delivery help explain the uneven pattern. This does not establish that every built-environment activity will lag other sectors.

Ask management. Which constraints actually bind in our work, and which could we remove? Are we using genuine delivery risks to guide investment or to defer it?

What would change my view. Repeated use despite friction supports the view. Rapid, widespread adoption without those constraints, or persistent failure to adopt, would challenge it.

Why the market is not moving as one →

What the evidence covers: the direct effects are not yet tested.

My evidence read. Concern and corporate responses are visible in the wider AI debate. The evidence reviewed here is much weaker on whether they change built-environment deployment. Public concern alone does not stop a project.

Ask management. Whose agreement do we depend on? Where could workforce concerns, client rules, information rights or infrastructure consent change the plan?

What would change my view. Documented restrictions, delays or changed deployments would support the view. Repeated adoption with concerns resolved would narrow it.

Start with the challenge and constraints →