Arcadis, the Amsterdam-headquartered design and engineering firm, has made a strategic financial investment in Nomic and entered a long-term commercial partnership with it. The investment amount was not disclosed, and neither were the partnership’s financial terms.
That is the announcement. The reason it is worth more than a line in a funding roundup is what preceded it: a six-month trial involving roughly 150 Arcadis engineers across 12 countries, with results the company was willing to put in writing.
The Trial, and Its Numbers
Per Arcadis’s own announcement, 86% of participating engineers said Nomic changed how they approached their work, and more than 25% described the shift as fundamental. One workflow, the company says, took several days of manual document cataloguing down to hours.
These are self-reported figures from a firm that has just become a shareholder, which is the obvious caveat and should be applied firmly. “Changed how I approach my work” is also a soft measure — it captures novelty as readily as productivity, and a six-month pilot is exactly the window in which novelty is strongest.
But the shape of the disclosure is unusual enough to note. Most enterprise AI announcements in this industry describe a partnership and a direction of travel. This one describes a population, a duration, a geographic spread and a percentage, and it does so before claiming a result rather than instead of claiming one. Arcadis reported €4.87 billion in gross revenues in 2025 with more than 34,000 staff in over 30 countries, according to its about page. A 150-engineer trial is roughly half a percent of headcount — a real pilot, not a proof of concept, and not yet a rollout.
AEC Magazine reported that the arrangement also gives Arcadis a role in shaping Nomic’s product roadmap, and that the trial exercised the platform on drawing review against firm standards, code compliance checks, submittal review, RFI research and BIM coordination.
What Nomic Is, and Where It Came From
Nomic describes itself as a domain-specific AI platform purpose-built for architecture, engineering and construction. Its site lists two products: a platform with out-of-the-box agents for drawing review, submittal review, RFI research, code compliance checks and firm-wide knowledge search, and an Agent API offering drawing parsing, structured data extraction and multimodal file embedding for teams building their own agents. It integrates with SharePoint, Box, Egnyte and Autodesk rather than requiring firms to migrate their data.
The company’s lineage is the part most AEC readers will not know. Nomic’s GitHub organisation hosts GPT4All — an open-source runtime for running language models locally, with more than 77,000 stars — alongside its Python SDK, the deepscatter visualisation library and contrastors, a contrastive model training library. It also now hosts aec-bench, described as a multimodal benchmark for evaluating agentic systems in architecture, engineering and construction.
That public code history is an AI-infrastructure history, not an AEC one — a local model runtime, an embedding SDK, a plotting library — with the construction work sitting on top of it. The benchmark repository is the tell: building your own evaluation set is what you do when you have concluded the generic ones do not measure the thing your customers care about.
Nomic CEO Andriy Mulyar’s line in the Arcadis release is aimed squarely at that distinction: “Arcadis has the scale and technical depth to put AI where it belongs: inside real engineering work, not next to it.” Arcadis CEO Heather Polinsky’s is more conventional: “AI is already changing how our industry designs, coordinates and delivers complex projects.”
Why This Shape of Deal Keeps Appearing
An incumbent taking equity in its own AI vendor is becoming a pattern rather than an exception. McCarthy and Palantir are co-building an operations platform under a multi-year partnership. DPR’s venture arm and Suffolk Technologies invested in Skillit while also being customers. The logic is the same each time: the buyer wants roadmap influence and price certainty on something it is about to become dependent on, and the vendor wants a reference account large enough to be a distribution channel.
It also solves a specific problem for AEC AI vendors. The hard part of selling drawing-review agents is not the demo — it is that every firm’s standards, templates and review conventions differ, and a tool that cannot absorb those is a toy. A partner willing to run 150 engineers at it for six months across 12 countries is supplying training conditions that no amount of capital buys directly.
The risk sits on the other side of that trade. A vendor whose roadmap is shaped by one very large customer can end up building that customer’s internal tooling with a logo on it, which is a good business and a bad venture outcome. Nomic now has to serve Arcadis’s specific workflows without narrowing to them.
What to Watch
Three things would turn this from an encouraging pilot into evidence.
The first is whether Arcadis expands beyond the trial population, and says so with a number. Six months and 150 engineers is a serious test; 3,000 engineers and a renewal is a verdict.
The second is whether the agents survive contact with liability. Drawing review against firm standards and code compliance checking are not summarisation tasks — they produce outputs a professional stamps. Every AEC firm deploying this has to decide where the engineer’s review sits in the loop, and the honest answer today is that nobody has published one.
The third is aec-bench. A public benchmark for agentic systems in AEC, maintained by a vendor that sells agentic systems in AEC, is either a genuine contribution to a field that badly needs shared evaluation or a marketing surface. Which one it turns out to be is checkable, and worth checking.
