Construction AI: the platforms start fighting over your data, and the EU AI Act clock hits August

For two years the agentic-AI story in construction was about openness. MCP, marketplaces, certified third-party agents you could call from inside Procore or Autodesk Forma - we covered that picture only last week. Here's the other half of it, and it's less comfortable.

On 12 June, Engineering News-Record ran a piece with a blunt headline: construction platforms are already fighting over data to train AI agents. The reporting is specific. Procore banned Trunk Tools - an agentic-AI provider used by large US general contractors including Gilbane and Suffolk - from access to its API. Trunk Tools lost its API connection last September and even had its booth at Procore's Groundbreak conference refunded, before Procore turned round and bought a rival agent platform, Datagrid, on 20 January. Procore's public line is that it's protecting "the integrity and security of all our customers' data". You can read that two ways, and ENR clearly reads it both. The prize, the article notes, is geometry - getting AI to actually understand the spatial relationships in a model - and whoever holds the most project data holds the best shot at training an agent that can.

This isn't a US-only curiosity. It's the same move SAP made in its API Policy v4/2026, which restricts autonomous AI systems from chaining SAP API calls - the enterprise software world's opening shot in the same war. If you run a cloud platform as your ISO 19650 common data environment, the data flowing through it is your project's golden thread. The question of whether your own AI agents - or a third party's - can read that data through the API is no longer a technical detail. It's a commercial term, and right now most of us haven't read it. There's a counter-move worth noting: on 4 June, US contractor McCarthy signed a multi-year deal with Palantir to build its own AI operating system, "Pulse", on top of its own data rather than depend on a platform's goodwill. That's the expensive answer. The cheap answer is to read your contract.

The procurement filter: Before you renew any CDE or platform agreement, find the two clauses that matter - who owns the data, and whether your own (or a third party's) AI agents can read it through the API. If the contract is silent, that silence is the vendor's option, not yours.

There's a date in the calendar that a lot of UK firms are treating as someone else's problem. On 2 August 2026, the EU AI Act's obligations for high-risk AI systems become generally applicable. That's the tier covering AI used in ways that affect safety, employment decisions, access to essential services and the like - and the Act applies extraterritorially. If an AI output your firm produces ends up being used in the EU, or you supply an AI-enabled service into an EU market, you can be in scope regardless of where your office is.

Browne Jacobson's 2026 construction horizon-scan, published as part of its sector series, makes the practical point that should worry every commercial director: standard-form construction contracts have not kept pace with AI adoption. The firm warns there's real uncertainty about how existing liability provisions apply to AI-generated outputs, and about the rights in the underlying data - who owns it, what licence terms attach, and who carries the can when an AI-generated design or document turns out to be wrong. JCT and NEC weren't written for a world where an agent drafts your RFI response or flags a clash. Most amendments haven't caught up. The honest position is that a lot of AI use on live UK projects right now sits in a contractual grey zone, and the Act's August deadline removes some of the room to wait and see.

For your board pack: Map where AI touches anything that could be called high-risk - safety-critical design, automated decisions about people, anything exported into the EU - and get the data and IP ownership clauses in your live contracts reviewed before August, not after the first dispute.

Google announced Gemini 3.5 Pro at I/O on 19 May, then told the audience to wait a month for access - a line that reportedly drew groans from a crowd that wanted it there and then. As of mid-June the model is in limited Vertex preview with general availability expected this month; Gemini 3.5 Flash, which went GA on 19 May, is the public 3.5 model until Pro lands. Reporting on 6 June put the headline specs at a 2-million-token context window and a "Deep Think" reasoning mode aimed at frontier multimodal work.

The number that matters for our world is the two million tokens. That's roughly enough to hold an entire O&M manual, a full set of specifications, or a stack of fire-safety documentation in a single context - and reason across all of it at once, rather than chunking it and hoping the retrieval picks the right page. That's precisely the shape of a golden-thread query: "show me every document that references this fire door's certification, and tell me what's missing." Whether Gemini 3.5 Pro is actually better at that than the models already shipping is unproven until it's in general release - and a big context window is not the same as good judgement about what's in it. Treat it as a candidate for your document-review workflows, run it against a closed-out pack where you already know the answer, and see if it earns a place. Don't point it at a live golden thread on day one.

A practical step: When Pro lands, benchmark it on a completed project's O&M pack with known gaps - if it finds the gaps you planted, it's ready for a real one.

Three stories, one spine. The platforms are competing to own your data; the regulator is about to start asking harder questions about how you use AI; and the models are getting good enough to reason across your whole document set. The order matters. If you don't control the data - if your project's golden thread lives inside a platform whose API terms can change at the vendor's discretion - then the cleverest long-context model in the world is reasoning over an asset you don't fully own.

Digital Construction Week at ExCeL on 3-4 June framed this year's theme as moving "from promises to proof points", and the UK adoption data backs that mood: around half of UK organisations now report at least moderate productivity gains from AI, which means the experiment is over and the operational reality has started. So the standing discipline doesn't change, it just gets more urgent. Know where your data lives. Read the clause that says who owns it. Keep a human signing off anything that carries liability. The tools will keep getting better; the contract you signed is the thing that decides whether that helps you or your vendor.

The takeaway: Pick one live platform contract this week and read its data-ownership and API clauses end to end. That single hour is worth more than any model upgrade you'll see this month.