Juan Andrés Núñez

The industry has changed. AI agents write in seconds what took us hours. The market demands a new engineer profile, yet education lags behind, still teaching with 2022 methods for a world that no longer exists. If how we work has fundamentally shifted, how we learn must shift too. For the first time, what was once science fiction is now possible. Cognitive science has known for decades how humans learn best: personalized tutoring (Bloom's 2 Sigma Problem), practice at the edge of ability (Ericsson's Deliberate Practice), immediate feedback loops (Hattie & Timperley), retrieval over re-reading (Roediger & Karpicke), among many others. These principles were impossible to scale in remote education, until now. As a senior software engineer specializing in generative AI and a professional educator who taught web development at university for years, this intersection has become an obsession: adapting how we teach technology to match how we now build it. In this talk, I'll show live how Mentat (an AI orchestrator) can replicate the characteristics of in-person, one-on-one tutoring at scale: Socratic conversation, deliberate practice, active recall, comprehension gating, adaptive scaffolding, and real-time validation. We'll see live how science-backed learning principles translate into working code, and why teaching fundamentals matter more than ever in an era where the execution barrier has collapsed.