Generative AI is changing software engineering from a practice constrained by scarce implementation effort to one shaped by abundant, low-cost code. This talk examines the engineering challenge that follows: when a fleet of coding agents will build almost anything you describe — quickly, but only what you describe! — how do you ensure that what they build will be auditable, correctable, and maintainable? To explore that question, I undertook a 20-week case study. I built a production-grade system for automated document accessibility remediation by managing coding agents. From this experience, I propose Model-Based Agentic Software Engineering (MAGE), a methodology organized around two moves. First, bind engineering intent to structured models through which agents can reason. Second, govern implementation through mechanisms enforced by the engineering environment. MAGE provides a method for developing software rapidly while preserving quality, and for converting recurring agent failures into durable engineering controls.
About James Davis
James C. Davis is an assistant professor of Electrical and Computer Engineering at Purdue University. He worked for IBM from 2012-2015 (IBM Spectrum Scale/GPFS, Poughkeepsie, NY) and received his PhD degree from Virginia Tech in 2020. His research is in software engineering, with applications in systems and cybersecurity. His work appears at venues such as ICSE and IEEE S&P, and has been recognized with three ACM SIGSOFT distinguished paper awards. His lab is supported by the NSF, Google, Rolls Royce, Cisco, Socket, and OpenAI. He received the NSF CAREER award in 2026. His work has been integrated into products and platforms such as AWS FreeRTOS, Node.js, and Ruby.