CS education must shift from coding to AI verification, 23-author paper argues

A landmark white paper from two NUS-Google Workshops in Singapore, authored by 23 faculty, industry practitioners, and students, calls for a fundamental overhaul of undergraduate computer science education in the age of generative AI. The paper argues that traditional CS curricula have long centered on practical software engineering skills—implementation-level programming, debugging, testing, and documentation. However, generative AI now automates many of these tasks, making them less tenable as the core of a CS degree. The authors propose a strategic pivot toward system design, abstraction, and critical evaluation of AI-generated artifacts, emphasizing the need for graduates to verify and manage co-created outputs rather than write code from scratch.

The paper introduces the concept of 'breadcrumbs'—small, engaging exercises and nudges integrated into existing courses to gradually shift focus without overhauling the entire curriculum. Key recommendations include fostering AI-native competencies (e.g., prompt engineering, model evaluation), recentering fundamental education (e.g., algorithms, data structures) but with an AI lens, enhancing advanced pathways (e.g., AI safety, ethics), and embracing new pedagogies like project-based learning with AI tools. The authors also call for institutional support, such as updated accreditation standards and faculty training. The goal is to prepare graduates who can create, solve problems, and co-create with AI, not just write code that AI can now generate. The paper is available on arXiv under cs.CY.