Yoders
Technology Viewpoint
Construction Platforms Are Already Fighting Over Data to Train AI Agents
Procore has fired the first salvos for the battle to control construction data to train agents for construction tasks
Firefly AI rendering promted by Dylan Schutter/ENR
This rendering, created by the artificial intelligence-enabled image generator Firefly, was created by ENR Assistant Art Director Dylan Schutter using prompts describing the future of construction such as robotic welder arms and AI-assisted crane picks.
Congress recently forwarded a House proposal to create a federal framework governing artificial intelligence development. Yet, in construction a different fight over construction data is brewing over the need to train agentic AIs to be able to do everything from reviewing project documents to preparing a bid or even inspecting welds in steel fabricators' facilities.
OpenAi, the maker of ChatGPT, recently signaled it would comply with a recent Trump administration executive order that would let federal regulators assess the impacts of foundation models like GPT or the family of frontier foundation large language models behind Anthropic's Claude AI. While governments are increasingly invested in reviewing how these foundation models will affect users and even jobs, the AIs emerging in construction aren't based on general data scraped from the web.
Rather, they are based on construction-specific data to help the AI agents understand dunnage, mechanics' liens, geometric symbols on 2D plans and all of the work processes design and construction have managed to preserve for themselves through, first, the CAD revolution and even 3D building information modeling.
While agentic AI has already been a boon for preconstruction processes such as estimating and document review, making AIs understand geometry remains a moving target. This has made construction data to train AI agents a highly desired commodity for tech companies with construction management platforms. Who has access to project data and uses it to train their AIs is now being stingily protected by those platforms.
Last year, Procore banned Trunk Tools—a popular provider of AI agents used by large general contractors such as GIlbane Building Co. and Suffolk Construction—from access to its application programming interface, making it much harder for contractors who use Procore as their construction management system to use Trunk Tools and its agents.
Procore said it was merely protecting "the integrity and security of all our customers' data as AI quickly reshapes the technology lands," but the popular platform quickly thereafter acquired DataGrid, another provider of agentic AI services to contractors that even boasts its own agent builder for GCs to use their own data to build their own agents. The integration of DataGrid as the natural language AI agent companion in Procore's platform was swift.
Construction is unique because so many contractors and even architects and engineers have closely guarded project information to give themselves a competitive advantage for decades. Indeed, the science and engineering behind moving construction forward is valuable and not inexpensive, especially when making designs such as the Sydney Opera House buildable.
Looking for quick answers on construction and engineering topics?
Try Ask ENR, our new smart AI search tool.
Ask ENR →
Historical construction data becomes more valuable the more of it you have, but these platforms closing guarding and hoarding construction data, with the consent and agreement of their customers, threatens to undo the progress in open information sharing made thanks to the application programming interfaces and standards that came with 3D building information modeling in the late '90s and early '00s.
Sadly, the promise of BIM never came to full fruition for similar privacy and competitive concerns. The need to preserve processes that limit risk and minimize the sharing of project information to a need-to-know basis stopped BIM from ever being the collaborative, model-based platform once promised. Between 2005 and 2012 BIM shifted from an ideal of all stakeholders working together on one shared model to cloud platforms to automate RFIs and change orders, processes that are less than collaborative and open.
One construction technologist who has built an agentic AI platform once told me "if we're just making a better change order platform, then we've lost the script from making models that don't require change orders."
Like many industries, construction stands at the precipice of agentic AI changing the processes it once depended on and automating repetitive tasks that have long held back best delivery practices and efficiency.
It will be a huge missed opportunity if the vendors that serve construction choose the process and risk-mitigation methods of the past rather than recognizing the possibility and opportunity presented by open standards, data sharing and truly looking at every architect, engineer, contractor, specialty contractor and even owners and their reps.
This may sound like a utopian vision of building teams working together in harmony with none of the problems inherent to construction or moving design forward. Indeed, I may be a dreamer, but I'm not the only one. The dream of coordinated BIM and teamwork in the model is only enhanced by agentic Ai and automation of repetitive tasks in both design and construction. Let's not repeat the mistakes of the past when it comes to sharing construction data.
