Construction in the age of AI: where do we really stand?

I start with the figures, because they are clear. In 2025, just under one European company in five uses AI (19.95%). In construction, it is one in ten (10.79%): the lowest score of all sectors measured by Eurostat. And among companies that do not use AI yet, construction firms are also the ones that have thought about it the least.

Switzerland is moving faster than the average. According to a UBS survey of around 2'500 companies in 2026, six in ten already use AI. But here too, construction is among the latecomers. And it is one of the sectors that least expect AI to affect their jobs.

I do not think this delay is laziness. A building site changes every day, teams rotate, some of the workers are temporary, and much of the information still lives on paper, in notebooks or in WhatsApp messages. AI needs clean data. Building sites, as I see every week, produce very little of it.

Even today, in 2026, Swiss building sites are buried in time sheets, just like in 1950. Since the Federal Labour Act of 1964, every employer must record the hours of every worker, day by day and week by week (Labour Act, art. 46). Today, those records must be kept for five years (Ordinance 1 to the Labour Act, art. 73). The rule is sound. But on many sites, the way it is done has barely changed in sixty years: a sheet filled in by hand, brought to the office on Friday, then typed up again. Do the maths for a team of 30 workers: 150 working days to read and re-enter every week. One hour misread or mistyped, and the client gets a wrong invoice and the worker a wrong wage. These mistakes come back, sometimes again and again.

For a long time, the sector could do without these tools. I see three reasons that make the question pressing.

Productivity is flat. Worldwide, labour productivity in construction grew by about 1% a year over twenty years, against 2.8% for the economy as a whole (McKinsey Global Institute, 2017). The figure is old and global, but nobody has really contradicted it since.

Hands are missing where it matters. In 2025, the skills shortage eased in Switzerland in most occupations. Not on building sites: site managers, foremen and production managers remain in 2nd place among the hardest jobs to fill, out of 32 groups (Adecco and University of Zurich, November 2025).

Temporary work carries part of the sector. In 2025, temporary work hours fell by 7.1% in Switzerland. Construction is one of the few sectors where they grew (swissstaffing, January 2026). Every temporary worker means a file, documents, hours to approve and a wage to calculate, often for just a few weeks.

AI in construction is not a humanoid robot building walls. It is mostly discreet tools.

First in the lab, then on site. In Dübendorf, the DFAB HOUSE built by ETH Zurich and Empa was the first building to take six digital construction processes from research to a real site, including a robot working on site. An ETH Zurich spin-off, Gravis Robotics, now fits excavators so they can work on their own; it raised 200 million dollars in 2026.

In the office. This is where AI is moving fastest: reading a document, checking that a permit is valid, checking a time sheet, catching a payroll error before it goes out. Nothing spectacular, but this is where I see the most value: hours saved every week, and fewer mistakes for people paid by the hour.

A lead: safety. According to Suva, more than 50'000 accidents still happen every year on Swiss building sites. Suva also points out that temporary workers face a higher risk, especially when they start at a company they do not know. Computer vision tools promise to spot a forgotten helmet or a dangerous area. It is a serious lead, but it immediately raises the question of monitoring workers (see below).

AI does not build formwork for a slab, does not judge whether the ground will hold after three days of rain, and does not replace the eye of a foreman who knows his team.

The studies point the same way. For the World Economic Forum, frontline jobs, including construction workers, are the ones expected to create the most jobs by 2030. For the International Labour Organization, clerical jobs are the most exposed to generative AI, and jobs are more likely to be transformed than to disappear.

So for me, the real question is not "will AI replace construction workers?". It is: who will do, tomorrow, all the admin work around them, and how much time does it still take today?

For the ecosystem we are building, AI is arriving today as a blessing. It may be the key to our repetitive administrative problems.

Who owns the site data? Hours, location, photos, identity documents: this is personal data. In Switzerland, an employer may only process what relates to the job (Code of Obligations, art. 328b), and systems designed to monitor workers' behaviour are prohibited (Ordinance 3 to the Labour Act, art. 26). My position is simple: a tool that helps is welcome, a tool that monitors is not.

Which rules will apply? The European Union has adopted its law on AI (the AI Act). The heaviest obligations for tools that touch employment and the management of workers have been postponed to 2 December 2027. Switzerland has no AI law of its own: the Federal Council is preparing changes to Swiss law in order to ratify the Council of Europe convention on AI. In my view, a Swiss company that works with Europe has every interest in getting ahead.

Who decides, and who is responsible? If a tool suggests rejecting an hour, blocking a file or flagging a risk, I want a human to make the call, and to be able to understand why.

Where to start? I would not start with a robot. Rather with what wastes everyone's time every week: paper, re-keying, the back and forth between the site, the agency and the office.