Artificial intelligence (AI) has been 70 years in the making since eminent AI pioneer, John McCarthy, coined that term in 1955 to describe “the capability of a machine to imitate intelligent human behavior.” Truss software has been 50 years in the making, with scant penetration of AI technology into truss software over that time. So why, now, does its entrance seem eminent? The primary reason is that we hear a steady drumbeat of AI innovations in movies, music, and so many other facets of life. And we naturally assume that innovation in the truss business will follow this prevailing trend. While we must continue to study AI technology and assess where it may be used to improve our industry, we also need to understand how its development is radically different than prior innovations.
AI is typically constructed using vast quantities of industry data – which is something that truss plants, unlike many other businesses, generate daily, adding many millions of computer bytes to the billions of bytes of data that has already accumulated on servers. So, at least our industry has built a basis for the development of AI technology. But is that data easily accessible and does it contain adequate information upon which to base AI efforts?
Although the quantity of available truss data may be adequate, the data files themselves must be well organized and properly documented. Typically, organizations that have mastered the filing of repetitive work, like that of national single family and apartment builders, have well organized and identified data files. But there are also companies that undertake highly customized work that have their own excellent computer-based filing systems.
The quality of data files is of overriding importance because that is what will be used to “train” an AI system. But truss files normally contain only the estimated labor cost and not the actual labor hours expended, so they will be of limited use only in determining optimal truss configurations, even after examining millions of different designs. Furthermore, such a system needs to include the type of equipment used, so the value of past labor data will be diminished after new automation is added to a plant. Even so, now is the time for plants to increase the usefulness of truss data files by adding actual labor hours expended, which will allow a meaningful reconciliation of actual-to-expected costs for present and future uses.
Individual truss data files, when matched with labor data, may aid in automating selection of optimal configurations, but they usually lack information regarding adjacent trusses, unless they are girders. Because trusses are built in batches, a better approach would be for an AI system to be based on the entirety of a job, which would involve using roof and ceiling shapes as a basis. Countless manual efforts have gone into developing such a job costing system, but with limited results. Parameters such as the length of ridgelines and hip-lines, pitches, heel heights, and other geometric features all play into the difficulty of a job. However, the location of tray ceilings and interior load bearing walls must also be considered. At the recent SBCA Open Quarterly Meeting, Joe Hikel discussed the need to assign geometric characteristics to each layout that would aid in comparison to prospective work. This approach taps into the fundamental advantage of AI in transforming complex data from multi-dimensional spaces into machine-readable formats, by utilizing pattern recognition and vectorization capabilities. However, can such systems “learn” how to frame every conceivable geometric shape, such as that shown? And would legacy truss layout software have a better chance extending its auto-framing capabilities than by developing exceedingly complex AI approaches?
Despite the apparent complexity of developing AI software, it appears that the truss industry may be lagging behind other developers, based on Craig Webb’s recent article, “AI is Altering the Takeoff Process Dramatically. Some Changes Could Cut You Out of the Sale.” In it, Craig references Home Depot’s AI-powered Blueprint Takeoffs tool and discusses the risk that this capability poses to building supply outlets and LBMs; a risk he could have extended to truss operations. He also lists multiple third-party software providers like Aximos, Flitch, and Seljax that are being used by building material suppliers, some of whom claim to be leveraging AI tools. But it appears that these are new entrants in the building industry, and their claims beg the question as to the source of the data upon which their purported AI software was based. Craig also reports that, “Vendors are working on programs that will compare prices for the materials requested with the cost of alternative building methods – I-joists vs. open-web floor trusses, for example.” While this comparison can be accomplished without the use of AI tools, this claim deserves further investigation. Yet it clearly illustrates the continual encroachment of third-party software into the truss world, often under the guise of AI capabilities.
There is no denying that AI technology is advancing at a staggering rate, but determining its eventual place in the truss business requires understanding of many new and unfamiliar technologies, such as agentic architecture, machine learning, retrieval-augmented generation (RAG), agent-to-agent (A2A) communication, etc. Yet, I suggest that we look at AI as just the next logical step, albeit a huge leap, in the evolution of computer software. In determining what it takes to make that leap, I also suggest that the companies who have developed the best truss software have the best foundation for success in AI (see my Truss Tales post, “Was Warren Buffett Wrong,” for more insights on our industry’s software development). However, they will need to partner with the component manufacturers who have the best stores of data upon which to base AI development. This isn’t a novel approach. In fact, it follows the familiar 50-year-old model for the development of truss software, which requires a large group of users to collaborate closely with expert software developers. If my analysis here is remotely accurate, we can expect to see gradual and not game-changing adaptations of AI technology in the truss business over the next decade and beyond. In the meantime, many dozens of high-level AI developers will gain employment, and not at the expense of the contribution of a largely intact army of truss estimators and designers.