Construction has a divide that few other sectors have as sharply: the work in the office and the work on the construction site. In the drawing room, in calculation and in project management, AI use is heading in the same direction as in other office work: text, calculations, checking drawings, drawing up quotations. On the construction site itself, things are different. There, less work is done with language and text and more with images, planning and logistics — progress photos being analysed, material planning being optimised, safety checks being supported by cameras. The balance between these two worlds determines where the risk lies: not evenly distributed, but concentrated in the places where decisions are made that cannot later be undone — an incorrect dimension calculation, a misinterpreted soil report, a drawing that is passed on to execution without being checked.
Construction companies work with many subcontractors, self-employed draughtsmen and project teams that come together per job. That makes it harder to see what AI is being used, because part of that use does not take place on company equipment or within company accounts. A calculator using a calculation tool on their own laptop, a draughtsman using a sketching program with an AI function, a project manager consulting a chatbot about a contract clause — none of that use appears on a licence overview, and no one has ever explicitly approved or rejected it. That is not a matter of negligence; it is a consequence of how the sector is organised. Anyone who wants to know what is really happening has to ask the people who do the work, and that only works if asking questions does not immediately lead to a correction.
Not every application of AI in construction carries the same weight. A tool that summarises emails is different from a tool that checks a structural calculation or a system that autonomously reorders material based on inventory data. Governance in this sector begins with distinguishing these roles: where is AI used as an aid in text work, where is it used to produce or check technical content, and where does it make a decision with financial or physical consequences without a human in between. That classification determines what level of oversight is needed, and that level differs greatly between a planning assistant and a tool that assesses drawings for structural consistency.
Construction works with main contractors, subcontractors, architects, consultancy firms and suppliers, and AI use can occur at any of these levels. When an error arises from an AI-generated calculation or an automatically generated drawing, it is not always immediately clear who is responsible for what: the party that used the model, the party that approved the output, or the party that accepted the final result. Governance here does not mean excluding AI use from the chain, but recording who carries out which control step and how that step can be demonstrated, so that in the event of a dispute or incident it can be reconstructed what happened and on what information a decision was based.
The question "do you use AI in your work" often produces a negative answer in construction, even when the answer should actually be affirmative. This is not due to unwillingness but to uncertainty: people are not sure whether it counts as 'AI', or they fear that an honest answer will have consequences for their position or for how their work is assessed. A governance approach that works therefore does not begin with a ban but with an inventory that is separate from assessment — only once it is clear what is going on can it be determined which policy logically fits and which risk classification belongs to each application.
Construction already has extensive quality and safety systems, from certification to oversight of structural safety. AI governance does not need to become a separate circuit alongside that structure; it works better when it connects to existing risk categories and reporting lines, so that an AI application with an impact on structural safety is assessed along the same lines as other risks with comparable impact, rather than in a new, separate framework.
The combination of office work and hands-on work, and of a chain with many separate parties, is not unique to construction. Similar patterns occur in the installation sector, where technology and administration likewise overlap, in manufacturing, where production planning and engineering deploy AI in their own way, and in the transport sector, where planning and logistics play a role similar to that on the construction site. Anyone working in one of these sectors will likely recognise some of the same questions.
An inventory of AI use and a classification by risk answer the question of what is going on and where oversight is needed. They do not answer the question of how much of the work itself, task by task, is suitable to be handed over to AI. For construction companies that want to know where in the process — from calculation to planning to administration — AI can make the most difference, that is a different question than governance alone can answer. The work scan from FTE TO AI calculates per task which part of the work can be taken over, as a supplement to the overview provided by the Responsible AI Scan.
Vraag maar. Governance begint bij weten wat er draait — ook wat niemand heeft goedgekeurd.
Answers come from this site’s knowledge base. Not tailored advice, and not a scan of your company.