Arcadis announced on 28 September an expanded collaboration with Autodesk. The release is datelined Amsterdam and runs to roughly 670 words of strategy language, most of which could be lifted into any enterprise AI partnership announcement of the past two years. Two sentences in it are not interchangeable, and they are the reason this is worth reading closely.
The first is a number. “In one example,” the release says, “automation has reduced quality reviews from five days to approximately half a day”. That is Arcadis’s own figure, for existing AI and automation initiatives in North America and Europe. Outside the boilerplate paragraph describing the firm, it is the only number in the release.
The second is an architectural statement: “Arcadis is connecting its own Model Context Protocol (MCP) capabilities with Autodesk Assistant, a flexible approach that lets Arcadis’ knowledge, standards, and guidance be applied within AI-powered workflows while retaining full ownership of what it builds.”
Why That Sentence Matters
What follows is analysis. For most of the past three years, the enterprise AI deal in this industry has mostly run one way. A platform vendor builds an assistant, trains it on what it can see inside its own products, and sells the result to firms whose standards and judgment then live inside the vendor’s system. The firm gets capability; the vendor gets the firm’s institutional knowledge as training and configuration data, and the switching cost that comes with it.
What Arcadis describes is the inverse. MCP — the open protocol for connecting models to external tools and data sources — lets a company expose its own knowledge as a server that any compliant client can call. Build the server once and it can be addressed by Autodesk Assistant today and by something else later. The standards, the review rules, the client-specific guidance and the engineering judgment encoded in them stay on Arcadis’s side of the interface. “Retaining full ownership of what it builds” is not throat-clearing; it is the commercial point of choosing that architecture.
The release is also unusually direct about the multi-vendor intent. Arcadis says its “wider digital ecosystem includes partnerships with other leading organizations”, and that it creates value “by combining a focused group of world-class digital technology partners, including Autodesk, with fast-moving startup solution partners and Arcadis’ engineering expertise.” Autodesk is named as one of a group. That is a different posture from a firm standardising on a platform.
It Follows a Pattern Arcadis Set Earlier This Year
This is the second time this year that Arcadis has moved on AI in a way that buys it leverage rather than licences. The firm took a stake in Nomic and entered a long-term commercial partnership after a six-month trial across roughly 150 engineers in 12 countries, with AEC Magazine reporting that the arrangement also gives Arcadis a role in shaping Nomic’s product roadmap. Neither that investment’s size nor the partnership’s financial terms were disclosed.
That announcement contained its own tell about vendor strategy. Per AEC Magazine’s report, “Nomic’s agents integrate with tools Arcadis teams already use, including Autodesk Forma and Bentley ProjectWise, with further platform integrations in development” — two rival platforms named in the same breath, as destinations for the same agent layer.
The Autodesk announcement discloses no terms either. The release puts no figure, no duration and no exclusivity on the arrangement, and names no client. Searches of the coverage turned up no reported financial detail; this reads as a commercial and technical collaboration rather than a transaction, and nothing in the document suggests otherwise.
Stack the two moves and a consistent strategy appears. Equity and roadmap influence in a startup whose agents do drawing review, submittal review, RFI research and code compliance. An MCP layer of Arcadis’s own, pointed at the incumbent platform its teams already draw in. And a stated preference for a portfolio of partners over a single stack. That reading is this newsroom’s analysis, not a claim Arcadis has made — the release describes an ecosystem, not a bargaining position.
What Autodesk Gets
Andrew Anagnost’s quote is the clearest statement of the vendor-side logic: “Arcadis and its strong leadership team are clearly organizing for the future, not simply adopting new technology. They understand that AI creates the most value when grounded in deep industry expertise, trusted data, and real-world context. That’s how AI becomes project intelligence, helping people make better decisions and deliver better outcomes for clients.”
The release describes Autodesk Assistant as understanding “geometry, engineering intent and project history” — which is the asset a firm’s MCP server cannot supply for itself. Arcadis has the standards and the judgment; Autodesk has the model of the building. The argument for connecting them is strong on both sides, and it is notable that the connection is being made through an open protocol rather than a proprietary integration. A reasonable reading, though neither company has said it: for Autodesk, a reference deployment at a firm of Arcadis’s size is worth more than the configuration data it forgoes. Arcadis reports around 34,000 staff in over 30 countries and €4.9 billion in gross revenues for 2025.
Heather Polinsky’s quote holds the human-expertise line the release returns to repeatedly: “Our clients choose Arcadis because of the knowledge, judgment and experience of our people. That expertise remains at the heart of how we create value.” The release makes the same point in its own voice, in a sentence worth quoting because so few AI announcements include it: “Technology alone does not create value. Adoption does.”
The Project Named, and the One Number
The release gives one concrete joint initiative: See Through Walls, which it describes as “a joint project using AI, sensor data and predictive modeling to infer the condition of what is hidden inside existing buildings, supporting material reuse, circularity and decarbonization.” No stage, timeline, pilot site or result is attached to it in the document. On retrofit work — where what is hidden behind the plasterboard is a well-known source of cost risk — inference rather than destructive investigation is a real prize, and it is also among the hardest kinds of claim to validate. Treat it as a stated direction until there is something to measure.
Which leaves the five-days-to-half-a-day figure as the only quantified thing in the announcement a reader can weigh. It is self-reported, it describes “one example” rather than a programme average, and quality review is exactly the kind of document-heavy, rules-based task where current models genuinely do well. It is not a platform claim — it comes from initiatives already running — and that makes it more useful than most of what gets published in this category. It is still one number from an interested party, and it should be held that way.
The rest is architecture. The architecture is the part that will still matter in three years.
