If you sit in a room of 50 contractors and ask them the first adjective that comes to mind when someone says “AI on the jobsite,” you’d probably expect most of the answers to be somewhat negative. But during an education session last week at the Associated General Contractors of America’s annual convention, the responses were pretty evenly split between “excited,” “confused” and “hopeful.”
One of the biggest barriers to increased adoption of AI in the industry has been a lack of case studies. Luckily, during the conference, contractors had an opportunity to see a high-profile case study: A progressive design-build project at San Francisco International Airport that has entailed building a six-story office structure in the core of the airport terminal. The building houses the airport’s executive leadership and functions as the Airport Integrated Operations Center, which is essentially the brain center of airport operations. Because of its role, the building contains advanced technology systems, including a large data center, a massive generator and fuel infrastructure.
The project embodies the challenges that come with fast-paced, change-heavy construction environments, said Garin Anderson, a superintendent at Hensel Phelps. As Anderson explained, the nature of progressive design-build means constant design evolution and real-time adaptation. To meet the challenges of the project, Hensel Phelps has been using a platform from a company called Track3D, which calls its technology an “AI-first Reality Intelligence Platform” that “serves as a central hub for all visual data, transforming reality capture data from a system of record to a system of intelligence.”
Track3D’s AI system was implemented to capture job site conditions and progress using frequent 360-degree video walks. At first, the goal was to simply maintain a reliable visual record, mostly for quality control and insurance purposes, but Hensel Phelps quickly saw the potential to extract much more. Over time, the contractor began using the data to track installation progress of various systems, such as framing, insulation, drywall, ductwork, cable trays, fire protection and doors. The AI system color-codes progress stages and overlays them on floor plans, making it easy to visualize completion percentages and spot gaps or errors.
One of the standout examples from the SFO project involved identifying a misinstalled duct that had been placed in the wrong room. The AI flagged the discrepancy on the floor plan, prompting a double-check that revealed the mistake—not with the AI, but with the actual field installation. Anderson said that prevented further downstream issues, like framing around the mislocated duct. Another success he pointed to was using the AI tool to assist with weekly schedule updates, owner meetings and commissioning coordination.
However, AI isn’t without its limitations. Anderson and NK Chaitanya, CEO and co-founder of Track3D, noted that AI struggles to track elements not visible on 2D floor plans and has difficulty identifying in-wall devices without highly detailed BIM models, which is something many owners don’t consistently provide. Additional issues included initial connectivity challenges during data uploads, and the camera’s sensitivity to lighting conditions and walking speed. On the practical side, successful implementation requires consistent and carefully executed 360-degree captures, as well as training field teams to walk job sites in a way that ensures complete visual coverage. There’s a learning curve involved in using both the camera and the capture process effectively. While data processing time is improving and approaching a six-hour turnaround, there’s still a dependency on visual data, meaning the system can only track what it can actually see. From a performance standpoint, AI cannot yet automatically verify the precise placement of devices compared to BIM models and still requires human oversight and feedback for quality control. Despite these challenges, Anderson and Chaitanya emphasized that these are not fixed limitations; they are evolving issues being actively addressed through continued technological development and real-world application.
AI in Estimation
Something that has been repeated in many industries is that AI won’t replace humans, but humans who use AI will replace humans who don’t. In a separate session at the AGC convention, Jennifer Johnson from ConstructConnect echoed that sentiment.
Last month, ConstructConnect updated its AI-powered tool, Takeoff Boost, to better align with wall centers as shown in plans and reduce errors by minimizing areas outside of buildings. These enhancements benefit walls, framing and ceiling contractors by speeding up estimate creation.
Some of those improvements were made possible by ConstructConnect’s relationship with its super users, such as Pete Malmberg, a senior lead estimator at C&C Drywall. During the session, Malmberg shared his journey with Takeoff Boost, beginning with small, familiar projects to test how well it worked.
“Start small, and then it fails small,” Malmberg said.
Now that he’s past the failing small phase, Malmberg has found AI helpful on big jobs with repeated layouts, such as apartment buildings with dozens of slightly different unit types. Instead of doing each one by hand, he could have AI do the bulk of the work and just make quick edits. However, he also noted that the AI doesn’t catch every detail, like small walls or special materials, but knowing what to look for makes it easier to spot and fix those issues.
When asked what would be on his AI “wish list,” Malmberg would like to have AI reading spec books, naming materials automatically and even helping generate bid proposals. Pete joked that his goal is to be “strategically lazy” by letting AI handle the repetitive work so he can focus on ensuring accurate estimates and winning jobs.
What’s next?
The emergence of AI models like DeepSeek is poised to dramatically reshape construction by slashing the cost of model development and retraining, says Patrick Murphy, CEO of Togal.ai, What once cost tens of thousands of dollars can now be done for under $5,000, he told SmartBrief in late January, at the height of DeepSeek’s early rise. This shift allows companies to retrain proprietary models more frequently, accelerating improvements in accuracy and relevance for construction-specific tasks.
Hamzah Shanbari, director of innovation at Haskell, noted that AI is already transforming workflows by speeding up access to project data and automating routine tasks. His team uses tools that let workers text queries and instantly retrieve project-specific information.
“We are still functioning the same way, but we are getting the information much, much faster,” Shanbari says.
AI is also being used to draft RFIs, analyze safety photos, and manage knowledge across projects, he adds.
Still, both Murphy and Shanbari have raised concerns about the security implications of open-source and foreign-developed AI models. Shanbari warned against using platforms like DeepSeek for sensitive or NDA-protected data, while Murphy emphasized that global competition in AI is accelerating.
“The floodgates are going to open pretty fast,” he said, calling this a “Sputnik moment” for American innovation.
Looking ahead, both agree AI will soon enable generative design, more complete drawings and near real-time design validation. As tools become cheaper, smarter, and easier to use, Murphy believes adoption will surge, especially if the user experience is seamless.
_____________________________________
If you liked this article, sign up for SmartBrief’s free email newsletter on construction technology from the Associated General Contractors of America.