MIT researchers have developed an AI copilot for structural engineers. While the algorithm generates the design, the system presents regions for the designer to evaluate and adjust — instead of showing only the final result. The work was published in the journal Communications Engineering on November 10, 2026, and highlighted by MIT News on November 13.
The black-box problem in structural design
The technique used is topology optimization, which iteratively fills a design space to meet defined goals, such as carrying a load with as little material as possible. It is already used in the automotive and aerospace industries, but its black-box nature has limited adoption for buildings and bridges: the engineer only sees the finished design and, to change something, often has to start over.
The copilot uses machine learning to quickly predict the regions of a design a human would most likely want to change. To train the model, the team created a synthetic dataset of thousands of likely human feedback entries, assessing aesthetics, performance and manufacturability. During execution, the system highlights those regions and the user can thicken or thin the structures there.
39% higher buckling load
In demonstrations, the human-in-the-loop approach improved manufacturability and linear buckling load — the force a part carries before it buckles — by 39%, adding only 15 seconds to the total design time compared with fully automated methods.
“Right now, the human doesn't have any agency or control over the design process as it's playing out,” explains lead author Dat Ha, who earned his PhD at MIT in 2026 and is now a structural designer at LeMessurier. “For these really complex designs, we wanted to also tap into our brain, which is the most complex machine there is, to solve these design problems in a way that's more practical to industry.”
For now, the system only allows users to thicken or thin structures in the identified regions; the team hopes to add other modification options in future versions. The paper's authors include Ferdous Alam, now an assistant professor at Georgia Tech; Markus Buehler, MIT professor; Faez Ahmed, MIT associate professor of mechanical engineering; and senior author Josephine Carstensen, MIT associate professor of civil and environmental engineering.
Image credit: Communications Engineering (2026). The image compares classic, human-informed and AI-assisted topology optimization workflows.
Sources: MIT News · Tech Xplore · Communications Engineering — DOI 10.1038/s44172-026-00818-x
Transparency: this article was produced with the assistance of artificial intelligence, cross-referencing public posts and internet sources. The facts were verified against the original sources cited above; consult them for more details.
By Muse