Over the past two years, working with AI agents has quietly rewritten how Sebastian builds ChatTutor, an edtech product grounded in self-regulated learning research from the University of Copenhagen. As tools change monthly and no workflow has time to become routine, he found himself running the same plan-monitor-reflect loop on himself that the product scaffolds for students. This candid talk walks through what that shift looks like in practice — prototyping in working code, directing an agent instead of authoring every line, and confronting failure modes like "metacognitive laziness" — and asks whether the frameworks that make human learning robust are also what make AI agents trustworthy.

Key Takeaways: Prototyping in working code is collapsing the traditional design-to-development pipeline. "Vibe engineering" speeds up real product work — but can just as easily hide risk behind false confidence. Self-regulated learning (plan, monitor, reflect) is emerging as the practitioner's best defense when tools outpace habits. The designer-engineer's job is shifting from author to director and taste-keeper of agentic output. The open question for every organisation: if this loop makes human learning robust, does it also make AI agents trustworthy?

This session is part of DTU's Industrial AI Program: From Artificial Intelligence to Industrial Intelligence at Digital Tech Summit 2026. Through a series of connected sessions, DTU brings together leading research and industry perspectives to explore how AI is reshaping products, processes, and operations across the industrial landscape.

Partners (the session): ChatTutor

Partners (the overall Industrial AI program): AI Denmark and AI Boost