Artificial intelligence is beginning to change the daily work of civil and structural engineers. But where do AI capabilities actually stand in the industry?
More than 1,000 engineers from around the world recently heard from experts at an event hosted by Bentley Systems, the global infrastructure engineering software company, about how to gauge the advances AI is making in their sector. The gathering, a webinar hosted online in late May, covered two distinct but connected ideas: How engineers who don’t code can use AI tools to start automating their work, and how the new Model Context Protocol (MCP) server technology could eventually make the coding step unnecessary altogether. (Bentley engineers, for example, have already used AI agents to redesign a steel structure and cut its weight by 40%.)
The hour-long session offered the clearest look yet at how AI is starting to shift the day-to-day work of civil and structural engineers, from Python and AI-enabled coding tools to MCP, which lets AI assistants like Claude or Copilot talk directly to engineering software. Think of MCP like a translator for AI agents: Instead of engineers spending hours clicking through menus to run calculations, they can type a plain English sentence—”optimize this steel frame”—and the AI agent does it automatically inside the STAAD software, in full compliance with the built-in engineering codes.
Bentley Chief Technology Officer Julien Moutte has been blunt about why this matters: The world needs more infrastructure than it has engineers to build it. AI agents, he argues, are an answer to that gap. By handling the repetitive calculation work, AI lets engineers focus on the decisions that better utilize their specialized skills and training. But there is a catch: AI that is only 90% accurate still leaves the engineer with 100% liability. That’s why the human must remain in the loop. “Would you drive across a bridge that is ‘hopefully’ designed right?” Moutte recently wrote for Bentley.
Timo Harboe Zollner, a Danish structural engineer with a massive following who teaches Python programming to fellow engineers, joined the webinar to give an honest, unvarnished picture of where things stand in the industry with AI. He is a newcomer to STAAD, yet Harboe Zollner showed how he used AI coding tools to write Python scripts that connected to the software, automatically pulled reaction forces from two versions of a structural model, and plotted the differences. He said that when done manually, the task takes “a really long time,” but with the right AI script, that timeframe can be reduced to just minutes.
Getting there with AI, though, wasn’t a five-minute miracle that many viral LinkedIn posts promise. “So six hours later, I was done debugging and actually getting it to work,” Harboe Zollner said. What does that means for trust? He said he would never use an AI-generated script on a real project without fully understanding every line. “If you know Python and you know STAAD, this can make you work 10 times faster,” he said. “But it can be really dangerous if you don’t know what you’re doing.”
For an industry routinely facing aggressive project deadlines, the automation promised by software APIs is a desperately needed lifeline. Yet a formidable barrier to entry remains: Civil engineers are trained to build physical infrastructure, not software architecture. As Harboe Zollner demonstrated, AI tools like Cursor and Codex can help engineers stitch together Python scripts to lower that threshold, but they still demand a baseline of programming fluency and involve lengthy reviews.
The second presenter, Biswatosh Purkayastha, is a solutions engineering manager at Bentley. He used Python and a mathematical algorithm called Delaunay triangulation—a method that generates cleaner, more accurate geometric meshes—to build a tool that automatically models circular water tanks, a notoriously finicky task. Then he showed the Bentley MCP server in action: Claude, the AI assistant, connected to STAAD and optimized a steel structure, piece by piece, automatically checking each one against building codes. “After decades of continuous research and innovation, Bentley has made remarkable progress in AI, and [we’re] now providing the technological bridge toward the future of intelligent engineering workflow.” Purkayastha said. “Our approach in the engineering first, meaning our AI is anchored in the real-world physics, structural logics, engineering context, delivering precise, trustworthy automation that you can rely on.”
The real breakthrough, Purkayastha’s and Harboe Zollner’s presentations made, lies not in writing code faster, but in skipping it altogether. That is the promise of MCP. Instead of scripting commands line by line, engineers type what they need in plain English and the AI handles the rest, inside the same validated software they already use. It is a paradigm shift that would transform the AI from a mere coding assistant into an integrated digital apprentice. It would also liberate engineers from tedious modeling tasks and the steep learning curve of programming. Ultimately, engineers could shift focus to other areas of the engineering workflow that would not have been given full attention otherwise, given schedule constraints such as quality review and design coordination with other disciplines.
The message from both presenters was the same: AI agents are genuinely useful in civil and infrastructure engineering, but only when an engineer stays in the loop. The AI handles the repetitive, time-consuming work. The engineer handles the judgment calls.
