AI in Construction Now Hinges on Who Controls the Data

Who Owns the Project Data Now Decides Which Construction AI You Can Run

For most of the past decade, construction software sold itself on features: better scheduling, cleaner document control, faster RFIs. As of mid-June 2026, the contest has moved somewhere harder to see and harder to swap out of. What platforms now compete over is data, specifically whether they can use a contractor's own project data to train their AI agents, and which outside agents are allowed to touch that data through an API. That decision, written into a developer-terms document most buyers never read, increasingly determines which AI a contractor can actually run on its own information.

The shift shows up in adoption. Engineering News-Record reported in mid-June that the share of contractors seeing measurable business results from AI climbed to 38 percent, up from 17 percent a year earlier. The jump matters less as a milestone than as a trigger. Once agents start producing returns a CFO can point to, the project data that makes those agents good stops looking like exhaust and starts looking like an asset worth fencing off. Adoption crossing from pilots into production is what turned data rights into a procurement fight.

The clearest example is already on the table. According to Engineering News-Record, Procore, one of the largest construction management platforms, updated its developer policy in 2025 to restrict bulk data downloads used to train commercial language models, and pulled Trunk Tools, an AI-agent vendor used by large general contractors, from API access. Procore later acquired Datagrid, an agentic-AI company of its own. Procore's stated reason is the defensible one any platform would give: protecting the integrity and security of the customer data it holds. The agent vendors on the other side make an equally reasonable case, that they train on construction data with the customer's consent, and that the customer, not the platform, should decide where its data goes. Both positions can be sincere. What is new is that the disagreement now decides which tools a contractor can deploy.

It helps to be concrete about why data control is the whole game. An AI agent in construction is only as capable as the project data it can read and learn from. A model that has never seen your drawings, your submittal history, or your change orders cannot reason about your job. Put plainly, artificial intelligence in construction management depends entirely on the messy, project-specific record it is allowed to reach. So whoever controls access to that record, and the right to train on it, sets a ceiling on how good any agent built on top of it can be. The API is the choke point. Cut access there, and a capable third-party agent collapses into a demo.

That is why a horizontal idea is worth raising here, as context rather than as a pitch. A model- and agent-agnostic integration layer such as AgentPMT comes at the problem from the buyer's side: it lets any MCP-capable agent connect to the tools and data a contractor already controls, with credentials held in an encrypted vault and injected only at the moment of use, never exposed to the agent itself, every action the agent takes written to an audit feed, and spending held under a budget. (MCP, the Model Context Protocol, is the open standard agents use to call outside tools.) No single product settles the Procore question. The useful point is narrower: the lock-in this episode exposes, where a platform choice quietly narrows a contractor's AI choices, is the exact constraint a portable, buyer-controlled approach is built to avoid.

Why the data fight is happening now

Three forces converged to make 2026 the year data became the prize. The first is the adoption curve just described: results make the underlying data valuable, and valuable data gets fenced. The second is labor. The industry is staring at a worker shortfall in the hundreds of thousands this year, with a large share of the skilled trades nearing retirement, which keeps automation near the top of every capital plan even as budgets tighten elsewhere. The third is the size of the waste these tools are aimed at. Rework, the cost of building something wrong and building it again, runs into the trillions across the industry each year, and the gap in aging infrastructure is larger still. When software can credibly chip at numbers that big, the data that powers the software becomes strategic.

Put those together and the logic is close to mechanical. Measurable results make project data worth training on. Data worth training on gets restricted. Restriction reshapes who can build a competitive agent. The more useful AI in construction becomes on an actual jobsite, the more the question of who holds the data decides who captures the value. For a contractor, that collapses two purchases into one. The platform decision and the AI strategy are no longer separate. Standardize everything inside a single vendor's walled garden and you have also, quietly, chosen the limits of what AI in construction management you are allowed to run on your own records.

Trunk Tools' answer: read the drawings, ride on top

The vendor on the wrong end of Procore's API change did not retreat. On June 17, Trunk Tools launched Cortex, which it describes as an intelligence layer trained over several years on real projects with large U.S. general contractors. It goes after the task that has blocked construction AI longer than almost any other: reading the drawings. Plans, specs, RFIs, submittals, schedules, and change orders are where a project actually lives, and most of it sits in formats that defeated earlier tools. Cortex reads and structures that material, then powers a set of named workflow agents for drawing review, document browsing, RFI handling, bid analysis, and submittal registers. It connects to the systems where project data already sits, including Procore and Autodesk Forma.

Notice the move. A company that lost one platform's API access shipped a layer explicitly built to sit on top of many platforms and the customer's own documents at once. Founder and CEO Dr. Sarah Buchner frames construction's need as a "system of action" rather than another single-app feature, the company's way of saying the value lives in reading and acting across a contractor's whole document set, not inside any one product. The performance figures Trunk Tools cites are its own and read best as vendor claims rather than independent results: it reports submittal cycles running 71 percent faster for teams on Cortex-backed review. Treat the exact number with the usual caution. The durable point underneath it holds regardless: reading drawings across every platform a contractor uses, whoever owns the front end, is where the leverage sits. A company that can do that well has something no single platform can easily wall off.

For buyers, the practical translation is a question about location. Where does an agent's intelligence actually live, and does it survive a platform changing its terms? An agent whose capability is tied to one vendor's data access is one policy update away from a downgrade. That portability concern is what makes interoperability worth treating as part of construction workflow management itself rather than a nice extra: wiring a tool up once through an open protocol so any agent can use it, and turning an internal or custom API into a governed tool where the buyer decides exactly which endpoints and methods an agent may reach. The aim is mundane and load-bearing, keep the intelligence answerable to the contractor, not to whichever platform happens to host the data this quarter.

The questions a buyer should be asking

The same logic is playing out across the wider real estate and construction market this month, which makes the construction-data fight look less like an isolated spat and more like a preview. Y Combinator's latest real estate and construction cohort is dominated by agents that do work rather than dashboards that display it. On the buildings side, automation and sensor data are turning into a measurable slice of asset value as smart-building deployments scale. On the residential side, the quality of any consumer or agent-facing tool tracks straight back to the proprietary data behind it. CoStar's new conversational apartment search leans on one of the largest housing datasets in the country; Opendoor just installed its first chief AI officer to put pricing and transactions under AI judgment. In each case a real estate AI agent rises or falls on the data it is allowed to see, and artificial intelligence in commercial real estate runs on the same rule the construction platforms just made explicit: own the data, or rent your intelligence from whoever does.

So the useful response is not a forecast. It is a short list of questions to put to any vendor before signing, the kind of due diligence that used to be about uptime and now has to be about data. Who is allowed to read your project data, and who is allowed to train on it? If you adopt an agent today, does it keep working when the platform beneath it revises its API terms next year? Can you see, in an auditable record, what every agent did on your behalf, cap what it spends, and require a human to approve consequential actions? Those answers decide whether the AI construction industry's next wave leaves a contractor with portable, owned capability, or with a powerful tool that belongs, in the end, to someone else's platform.

---

Sources

• Construction Platforms Are Already Fighting Over Data to Train AI Agents, Engineering News-Record
• Construction's AI Fight Moves to Data as 38% of Contractors Report Measurable Impact, MarketScale
• Trunk Tools Launches Cortex to Tackle Construction's Hardest AI Problem: Drawings, Trunk Tools
• Trunk Tools Launches Cortex AI Platform to Interpret Construction Drawings, Engineering News-Record
• Y Combinator's 2026 Real Estate and Construction Cohort Bets Big on AI Agents, MarketScale
• Smart Buildings Become a Financial No-Brainer as the Market Races Toward $554 Billion, MarketScale
• CoStar Group Launches Apartments.com Ai, Business Wire
• Opendoor Names Chief AI Officer; BiggerPockets Has a New CEO, Real Estate News