Log in
Imagine holding the line for another generation, banning the use of GenAI in computing education, and avoiding the hard work of reinventing objectives, assignments, assessments, courses, and curricula. Students in 2040 arrive in our CS1 lectures having relied on these tools in the same way as today’s “digital natives” have relied on ubiquitous Internet access and communication tools. Continuing with longstanding tradition, we ask them to average rainfalls or check whether a string is a palindrome.
Do we have any belief that our students will learn anything of value in such a world?
Ignoring GenAI is not an option. Bans cannot work unless we can provide secure environments even for large project-based programming assignments, and students educated in such a restrictive regime will be at a disadvantage relative to learners who interact with GenAI tutors. We cannot prepare students for careers in either industry or academia by pretending that GenAI does not exist.
Indeed, the norm is already shifting to students and professional engineers using these tools. None of us call ourselves compiler-assisted programmers today, and future software engineers will not refer to themselves as AI-assisted.
It is our responsibility as educators to adapt to this rapidly changing technical context. Indeed, many of these changes we suggest here have been discussed for years before the advent of GenAI. While none of us is happy with the timing or urgency of the GenAI (r)evolution, as a discipline we should be wise enough to not let the crisis go to waste.
Writing Code Is No Longer the Challenge
Most current curricula have been predicated on the notion that writing code is the challenging task. However, with GenAI, the production of code is no longer the bottleneck for students or professional engineers. Instead, the key challenges lie both before and after the production of code. In the realm of specification and design, we are challenged by our capacity to express our desired outcomes with sufficient precision. After the production of code, we are challenged by the difficulties of comprehending code, evaluating code quality, and validating overall correctness and trustworthiness. Embracing these challenges will bring our introductory courses to their full potential, where students can work with complex, meaningful projects that integrate all these competencies. Leaving our course outcomes, activities, and assessments unchanged in the face of GenAI poorly serves our students.
We have known and discussed aspects of these problems for many years. For instance, Harlan Mills wrote scathingly about the ways we teach syntax (while ignoring composition) as far back as 1988.5 GenAI is both enabling and forcing that long-overdue shift in focus.
Rethinking the Core of Computing Education
Changes to introductory programming (CS1) courses are the most pressing; they should be instituted with as much haste as our institutional policies can muster. But the scope of those changes should not end with introductory programming. As we shift our focus from producing code toward higher-level skills in specification, design, and verification, we will find opportunities to focus on higher-level concepts throughout the curriculum. We believe the learning outcomes of a core “computing education” must focus on comprehension, specification, usability, design, quality, verification, and validation. We expect that this core computing education would be applicable across the entire family of related computational disciplines. Specification and verification are just as valuable in data science, software engineering, and computational biology as in a general CS program.
Much of the focus of this core computing education sequence can be described as “leveling up” the competencies of our students and our discipline. In order to effectively use GenAI, succeed in industry, communicate with academic colleagues, or understand changing technology, our students need a deeper understanding of core competencies and more precise use of computing vocabulary. Let us teach our students how to effectively, critically, and ethically use GenAI for lower-level code writing under human guidance. This frees the human (student or professional) to focus more on higher-level competencies such as design (specify and prompt), orchestrate (problem solving and decomposition), and evaluate (verification and validation), as shown in the accompanying figure.
CS curricula requirements have generally only evolved through accretion. We may wish to use this reimagining of the first half of computing programs as a once-in-a-generation opportunity to reconsider the rest of our curriculum requirements, including those that are less essential in this new era.
Student Behavior
The ubiquity of GenAI has already shifted student behavior, but that behavior will continue to evolve. We must adapt our education mechanisms to address new metacognitive skills that are needed in a world with GenAI; curiosity, independence of thought, reading critically, and evaluating quality are paramount. Students who use GenAI as an ever-present and infinitely-patient tutor could be empowered in ways that even our best students from a few years ago were not. Enhancing students’ metacognitive awareness and control requires aligning learning outcomes with appropriate assessment tasks (providing learning incentives) and scaffolded teaching/learning activities (providing learning support); otherwise, students will succumb to the temptation of unreflectively “outsourcing” all effort to GenAI and submitting GenAI slop.
Again, the need to reconsider the educational system and curriculum we provide for our students is not novel. The Computing Education Research community has been discussing the value of metacognition in introductory CS education for many years. Curriculum designers and accreditation bodies have looked for this sort of behavioral uplift through the lens of “Student Dispositions,”4 arguing that these behavioral traits are essential factors to consider when defining success for a computing degree. GenAI is forcing us to confront those needs urgently, and revise our understanding of educational success.
Program Outcomes
We are not calling for a full “burn it all down and start from scratch” approach. From a sufficiently elevated perspective, many things remain the same. Good teaching is still good teaching, although the intended learning outcomes, mechanics of assessment tasks, and teaching/learning activities have changed. Good engineering is still good engineering, although the tools have changed. Good learning is still good learning, although the tools to support learning have changed.
For example, ABET Computing Commission program outcomes for accredited CS and SE programs remain just as applicable.1 When describing proficiency in the field, the outcomes use the phrase “in the context of the program’s discipline.” These program outcomes already account for some evolution of the discipline of computing. As such, the high-level competencies and dispositions of good graduates are still generally the same, although tools and context have changed.
Catalyze Change for Quality Learning
Leading change can be challenging for individuals and institutions, but our discipline will suffocate unless we act. A renewed approach to computing education has the potential to better prepare students for successful futures in the computing profession. We envisage changes in Learning Outcomes, Assessment Tasks, and Teaching/Learning Activities.2 The following provides initial guidelines organized around three levels of ambition: Quickfix, Renovation, and Visionary.
Quickfix
Intended learning outcomes. Course learning goals must be updated to explicitly address the responsible use of GenAI, and its implications. This will impact the relative importance of topics, with code writing de-emphasised in early courses.
Assessment tasks. Create an explicit policy on GenAI use in each course. Constructing unsecured assessments predicated on the invalid assumption that students are not using AI undermines such assessments and their value. Assessments should clearly delineate between competencies students are expected to demonstrate without GenAI and with GenAI. Summative assessments should often occur in secured environments (oral or proctored written exams) and formative assessments in unsecured environments (take home assignments or projects). Unsecured assessments should aim to assess the software design process more holistically, including the full range of artifacts students produce towards this end; including, for example, design and specification documents, prompt logs, code quality, and test cases.
Teaching/learning activities: Students should be given the opportunity to learn how to use GenAI in a way that develops their awareness of their role in learning processes involving GenAI. Metacognitive skills and effective study techniques should be emphasized.
Renovation
Intended learning outcomes. Revisit the learning goals of individual courses to ensure they align with GenAI-age software design practices. An example would be modifying an introductory programming course to incorporate GenAI.6 Also consider whether any existing courses should be removed from the required curriculum or new courses should be added.
Assessment tasks. Ensure consistency in policy within groups of courses. Create open-ended and/or authentic projects for students that allow them to practice using GenAI tools.
Teaching/learning activities: Incorporate an AI Tutor (CodeHelp, CodeAid, UIUC Chat) that is designed to assist students in acquiring the intended competencies. Provide students with guidance on how to use that tool and modify instructional staff practices to provide more support; for example, 1-1 tutoring sessions, discussions of community. To encourage student engagement and development of communication skills, active learning and flipped classrooms should be considered.
Visionary
Intended learning outcomes. Reimagine the curriculum from the ground up to both address long-standing curricular issues and to embrace how to effectively, critically, and ethically use GenAI as a fundamental way of programming. For example, system design, requirements specification, and testing might be taught with increasing levels of sophistication across multiple software courses.
Assessment tasks. Create a department- or school-level policy for what students are permitted or expected to do with GenAI. For example, the University of Sydney has a policy that GenAI is permitted in all take-home assignments, and variants of their approach are being considered for adoption by several Australasian universities.
Teaching/learning activities: Create a strategy as a department to provide students with consistent learning experiences. For example, all programming courses might use the same AI tutor calibrated toward the particular course.
You Are Not Alone: There Are Resources Available to Help
Computing departments everywhere are grappling with the same challenges all at once. Cross-institutional collaborations are likely to be more effective and less time-consuming than each institution charting a path on its own. For example, revisions to accreditation standards and curricular guidelines can serve as a blueprint for everyone. Schools that successfully integrate GenAI into courses (or the whole curriculum) should share their lessons learned with one another. Indeed, example courses that can serve as a starting point for many of the early courses in the curriculum are already available.3 Educational institutions and educators should work together to ensure that students graduate more intellectually empowered, socially connected, and capable of succeeding in their studies and careers than ever before, only now with the power of GenAI.
Emerging Agentic AI Systems. Since our Dagstuhl seminar,a GenAI tools have rapidly evolved from reactive code generators into increasingly proactive, so-called agentic systems (that is, systems exhibiting limited operational agency) capable of planning multi-step solutions, autonomously invoking external tools, maintaining state across sessions, and incrementally refining outputs. This shift toward greater autonomy further reinforces our central claim: the bottleneck is no longer code production, but human-in-the-loop judgment in problem framing, orchestration, and evaluation. As AI systems take on larger portions of decomposition and implementation, students and professionals must learn to effectively supervise, constrain, and critically and ethically audit semi-autonomous computational agents. In the short term, this strengthens the need for explicit GenAI policies and secured assessments. In the longer term, it suggests that curricula should explicitly address human–AI collaboration, including oversight, accountability, and system-level design in environments where AI acts with partial autonomy.