Every Nordic engineering giant built an in-house AI. Here's what none of them built.

Every Nordic giant built an in-house assistant

Sweco moved first and fastest. A first version of SwecoGPT reached hundreds of employees in one afternoon; the full tool took two days to build on Azure and was deployed, complete, within two weeks of the initial proof of concept. Microsoft's customer story from May 2024 reports that nearly half of Sweco's roughly 22,000 consultants use it regularly, for document analysis, search and translation.

Tyréns built Sven AI, named AI application of the year (Årets AI-tillämpning) at the AI Arena Awards 2025. The firm's own claim, made in the trade paper Byggindustrin: three weeks of work can be done in minutes. Ramboll went furthest organisationally, gathering software and AI development in a unit of its own, Ramboll Tech, described in its recruitment listings as over 300 people directing all AI initiatives in the group. Norconsult presented itself to investors at its November 2025 capital markets day as combining engineering, architecture and digital expertise.

AFRY wrote its version into the annual report, in the language of governance rather than pilot project:

AFRY has established an AI hub to facilitate innovation and disseminate tools, best practice and guidance. … All divisions have appointed coordinators to promote the use of AI and drive initiatives to embed it in sales processes and project delivery.

The assistants solved access

Access. A consultancy's working knowledge lives in project archives, method handbooks, standards and the heads of senior engineers. For decades the only interface to it was asking a colleague or reading for hours. The in-house assistants made that knowledge askable: retrieve the relevant section, summarise the report, translate the memo, draft a first paragraph. A team manager at Sweco puts the saving at up to two hours a day.

Adoption backs it up. Getting nearly half of a 22,000-person company to use an internal tool regularly, by choice, is a better outcome than most enterprise software ever sees. The generous read is also the correct one: the giants identified a real bottleneck, access to their own knowledge, and removed it in under three years.

None of them built the deliverable

The deliverable. An infrastructure project does not ship answers, it ships documents: technical descriptions, investigation reports, review statements, tender annexes. Each one must pull in the right requirements, stay consistent across hundreds of pages and survive formal review. The distance between answering a question about a standard and producing the document that passes review is the distance between a search engine and an engineer. Every tool in the table above stands on the near side of it.

The assistant model has a second, quieter problem: someone has to keep feeding it. A development lead at one Nordic consultancy described the life cycle of a bot built with real ambition:

Vi la riktiga resurser på en egen AI-bot. Sen hade ingen tid att mata den — den kan fortfarande bara det den kunde dag ett. Vi glömde sista frågan: hur kommer informationen in?

The two problems are the same problem. An assistant is a destination: you go to it, ask, and carry the answer back into the document, and keeping it current is extra work that belongs to no one. A system that works in the deliverable itself sits where the information already flows. The drawings, the requirements and the review comments reach it because handling them is the job, not a side task.

The giants could still win from here

Quite possibly the giants themselves. This is not a story of incumbents who slept through a shift: they moved early, spent real money, got real adoption and were right about the first bottleneck. Their domain knowledge and their clients' trust are not moving anywhere. What the survey shows is where the frontier now runs: one layer down from the assistant, at the deliverable itself.

That layer is what we build. Yesper is the AI civil engineer for construction and infrastructure: it completes written deliverables and acts on drawings, inside the review discipline the consultancies already own. The firms that pair their knowledge with that layer set the pace for the rest.