Engineer Builds AI Agent to Master Software, Saving Hundreds of Hours and Hinting at the Future of Infrastructure Design

• 10 October 2025
• Uncategorized
• 919 Views

An innovative application of generative AI, unveiled at the 2025 Bentley Illuminate conference, is poised to reshape how engineering firms train and deploy their talent.

Kyle Rosenmeyer, a model-based design leader at VHB, has built a custom AI agent using Microsoft’s Copilot Studio that rapidly teaches engineers how to use complex Bentley software, effectively eliminating the need for engineers to spend time “learning CAD.”

Rosenmeyer’s AI agent acts as an instant software tutor and troubleshooter, allowing engineers to ask natural language questions about Bentley products like OpenRoads or ProjectWise and receive precise, step-by-step answers.

He claims the tool has saved him over 100 hours in the past year and, based on a company survey, saves other users an estimated one to two hours per week.

“Engineers aren’t paid to learn software. They’re paid to design roads and bridges,” Rosenmeyer said. His central philosophy is that software should “get out of the way” so engineers can focus on core design tasks.

The custom agent was built on Microsoft’s low-code Copilot Studio, a platform designed for creating internal AI assistants.

Rosenmeyer attributes the agent’s high performance to a meticulously crafted “system prompt” that defines the AI’s personality and expertise, combined with access to five distinct knowledge sources: Bentley’s main site, blog, official documentation, YouTube channel, and, most critically, the user-generated content from the Communities forums.

“User-generated content… is where real engineers are asking real questions and sharing real insights. It’s messy but gold,” Rosenmeyer noted, emphasizing that the AI’s ability to “reason” its way to solutions for vague questions is a testament to the quality and accessibility of Bentley’s deep, well-structured public documentation.

The VHB agent is currently deployed within Microsoft Teams, making it instantly accessible and secure for all employees. Rosenmeyer stressed that building such a tool is “stupid easy,” requiring only basic skills to input the system prompt and relevant URLs.

He is now leading workshops to share the blueprint with others, advocating for a widespread adoption of similar Copilots across the industry.

Looking ahead five years, Rosenmeyer envisions the entire learning curve for professional software disappearing. “You won’t ‘learn CAD.’ You’ll just talk to your assistant,” he predicted.

Furthermore, he argues this shift in the design office is critical preparation for a larger transformation: the rise of “Physical AI” and robotics on construction sites. As autonomous equipment becomes a reality, the engineering and design workflows must be fully digitized and integrated with AI tools to manage the infrastructure data layer, ensuring the industry is not “caught unprepared” for the coming automation wave.

• iTwin
• AI
• Bentley Systems
• OpenRoads