Five Key Principles to Update CS Education in the Agentic Coding Era

If code becomes cheap, then understanding, judgment, and systems thinking become expensive. That is the shift we are living through.

Writing code has never been easier. Tools like Copilot and ChatGPT generate working code within seconds. But the real challenges of engineering have not disappeared. If anything, they have become the focus. This creates an educational risk. Students can become very productive very quickly, but without developing any depth. They can build systems they do not fully understand.

This blog post is not about how to use AI for coding. That part is trivial and will happen anyway. The real question is how we keep thinking, understanding, and engineering discipline alive. We present five principles for updating computer science education without losing its soul. They are based on our experience and expertise in industry and academia.

Specifically, Avi’s experience is in the chip design industry, where his team is currently achieving roughly a tenfold improvement in productivity compared with what was possible just a few years ago. In their day-to-day work, they also see that some engineers quickly adopt these new methodologies, grow with them, and become significantly more effective, while others struggle to keep up. In parallel, Orit’s experience is in academic research on computer science education, leading organizational change using the agile approach, and managing educational systems in times of crisis and instability. This combination of perspectives has led to the formulation of the following five principles, which recognize and reflect the importance of proactivity and agility.

Before using AI, students should write code by themselves: they should struggle to understand what is happening before using AI. This principle is derived from constructionism—a learning theory that emphasizes that people learn most effectively by actively constructing knowledge through the creation of meaningful artifacts. In our case, this principle is applied to the development of computer programs, a construct whose purpose is to support the learning of computer science concepts.

If it is Python, students should write Python. If it is a Web server, they should build it. If it is a data structure, they should implement it. Then they should do the same thing with AI.

This way, something interesting happens. Students can actually observe, not merely see, what the AI is doing, and so recognize patterns, spot mistakes, understand tradeoffs. Without trying first by themselves, AI is a black box; after experimenting on their own, it becomes a tool.

The purpose of this principle is to develop students’ program comprehension skills, which have become central in computer science education in the era of GenAI, as strong program comprehension opens up new pathways for teaching foundational programming concepts.4

2. Cultivating Intrinsic Motivation for Learning Through Engagement and Ownership

One of the most important skills today is learning things that nobody taught you, without instructions, whether it is a new language, a framework, or a domain. Students should practice this skill intentionally. The purpose of this principle is to foster intrinsic motivation, a crucial factor in an AI-driven learning environment.1 If students do not love what they do, they will always look for shortcuts.

According to Self-Determination Theory, meaningful learning depends on internal interest rather than merely on external rewards or task completion, and is fostered through curiosity and personal engagement.8

Fun is a powerful driver of motivation because it transforms learning from an obligation into an intrinsically engaging activity. When learning becomes playful, engagement and motivation increase. In computer science education, it can be fostered in many ways, including hackathons, challenges, competitions, building real-world solutions, and gamification. AI provides a suitable environment for such activities, since it supports the writing of solutions that are faster, more efficient, and more elegant.

Beyond being enjoyable, such learning environments support deeper learning and knowledge construction. Under competitive conditions, shallow understanding is quickly exposed, encouraging learners to deepen their understanding and develop habits of inquiry, through which understanding emerges from questioning, testing, and reflecting on their own thinking.

We can also engage students in coding experiments that demonstrate how small changes in code affect performance or behavior. Novices can change one condition in a game program (e.g., the speed of a moving object or the collision rule) and observe how this small modification changes the entire behavior of the game. This illustrates the relationship between code structure and system behavior.

Another example, more closely related to AI, is having students modify a machine learning model by changing a single parameter, such as the learning rate or the number of training examples, and observe how these changes affect prediction accuracy, convergence speed, or overfitting. Through such experiments, students observe how seemingly minor implementation decisions can significantly influence program performance and behavior.

All of the above can encourage students to move beyond passive consumption of knowledge toward active exploration and discovery, fostering agency, mastery, and intrinsic motivation, where learning becomes meaningful and real.

3. Apply Computer Science Across Disciplines

One significant impact of GenAI in the context of computer science education is the decline in enrollment in computer science programs.7 This observation is commonly attributed to AI’s capacity to perform many traditional tasks and solve problems historically addressed by computer scientists, a phenomenon associated with the “junior crisis.”3

Another, less popular way to interpret this decline in enrollment is that computer science alone is no longer attractive to many students, either because of lack of interest or due to practical considerations such as potential earnings. This observation can serve as an opportunity for computer science education by highlighting its connections to other domains, such as physics, agriculture, literature, music, economy, and ecology, in order to reflect its broader impact on the world. We can ask computer science students to build a tool for a transportation hub or a system for analyzing literary texts.

Among other approaches, we can teach the history of ideas to illustrate how concepts evolve over time, while connecting them to the backgrounds of the individuals who developed them, and the surrounding cultural context. Ada Lovelace’s work can be understood in the context of Victorian society, Alan Turing’s work in the context of World War II and early computing, and John McCarthy, who coined the term artificial intelligence, in the context of the Dartmouth Workshop.

In line with Principle #2, planning and developing games, as well as training them to succeed, also requires broader knowledge and skills across multiple domains.

This approach is called context-based learning.2 It is one way to foster meaningful learning. This approach demonstrates real value and forces students to expand their areas of interest beyond their main field.

4. Integrate Real Industry Experience into the Curriculum

The role of software engineers is changing dramatically in the era of agentic coding. It is no longer just about writing code in a professional manner. It is about managing systems and, more precisely, managing AI agents.

Therefore, students should develop essential contemporary industry skills, one of which is a managerial mindset that includes orchestrating workflows—specifically, decomposing tasks into subtasks, assigning them to appropriate agents based on their skills, monitoring execution, validating outputs, and implementing guardrails.

Again, we can play a game (See Principle #2): Assume that AI is an employee who may mislead you, not on purpose, but because this is how it works. AI can hallucinate, cut corners, and deliver with confidence even when it is wrong. The student’s role is to supervise, question, validate, and establish verification mechanisms. Indeed, such activities actually train students to think critically and reason about software development processes at a higher level of abstraction.

Authentic industry work can be integrated into the curriculum, creating a win-win situation for both sides: Students learn how things actually work and gain at least some of the foundational experience required by companies; companies get to know the students, who have now gained some experience, and the hiring risk decreases.

5. Continuously Reevaluate the Curriculum

We must admit that we do not yet fully understand what lies ahead or how to update the curriculum so that it remains meaningful, engaging, and relevant for students, industry, and academia. At the same time, there is a growing sense of urgency to rethink the curriculum, as tools evolve rapidly, professional roles shift, and long-standing assumptions are challenged almost daily.

It is therefore clear that the curriculum itself must be dynamic, organized in a way that leaves room for updates, experimentation, the addition of new elements not yet anticipated, and the removal of content that seemed relevant only a short time ago. Such updates should be carried out in close collaboration with industry to ensure that the gaps between academia and professional practice are minimized.

It’s worth repeating: If code becomes cheap, then understanding, judgment, and systems thinking become expensive. We should neither resist this shift nor be naïve about it. As with other disruptive technologies, a proactive approach is needed; otherwise, we risk becoming obsolete.5

Accordingly, in our previous blog post, “Should Universities Raise or Lower Admission Requirements for CS Programs in the Age of GenAI?,”6 we argue for a balanced stance: not necessarily raising, lowering, or maintaining admission requirements for computer science as they are, but rather reshaping them in different ways.

The five key principles presented in this post establish a foundation for redesigning computer science education, from which additional principles may be derived, informed by the cumulative experience we all gain as we navigate this transformative technological journey.

Ultimately, the goal is to educate engineers who understand what should be built, why it should be built, and how to verify that the product works properly and ethically. In this context, the fact that a significant portion of the curriculum is freed from code production should be viewed as an opportunity for computer science education.

1. Armony, Y. and Hazzan, O. (2024). Inevitability of AI in Education: Futurism Perspectives for Education for the Next Two Decades, Springer.

2. Bennett, J., Lubben, F. and Hogarth, S. (2007). Bringing science to life: A synthesis of the research evidence on the effects of context-based and STS approaches to science teaching. Science Education, 91(3), 347–370. https://doi.org/10.1002/sce.20186

3. Brynjolfsson, E., Chandar, B. and Chen, R. (2025). Canaries in the Coal Mine? Six Facts about the Recent Employment Effects of Artificial Intelligence. Stanford Digital Economy Lab, Stanford University. https://digitaleconomy.stanford.edu/app/uploads/2025/11/CanariesintheCoalMine_Nov25.pdf

4. Erez, Y. and Hazzan, O. (2025). Program Comprehension as a Central Skill in CS Education in the Era of Generative AI, BLOG@CACM. https://cacm.acm.org/blogcacm/program-comprehension-as-a-central-skill-in-cs-education-in-the-era-of-generative-ai/

5. Hazzan, O. and Erez, Y. (2023). ChatGPT (and Other Generative AI Applications) as a Disruptive Technology for Computer Science Education: Obsolescence or Reinvention, BLOG@CACM. https://cacm.acm.org/blogcacm/chatgpt-and-other-generative-ai-applications-as-a-disruptive-technology-for-computer-science-education-obsolescence-or-reinvention/

6. Hazzan, O. and Salmon, A. (2025). Should Universities Raise or Lower Admission Requirements for CS Programs in the Age of GenAI? BLOG@CACM. https://cacm.acm.org/blogcacm/should-universities-raise-or-lower-admission-requirements-for-cs-programs-in-the-age-of-genai/

7. National Student Clearinghouse (January 2026). Research Center: Postsecondary Enrollment Rises as Academic Trends Shift. National Student Clearinghouse https://www.studentclearinghouse.org/nscblog/postsecondary-enrollment-rises-as-academic-trends-shift/

8. Ryan, R. M. and Deci, E. L. (2000). Self-determination theory and the facilitation of intrinsic motivation, social development, and well-being. American Psychologist 55, 1: 68–78.

Avi Salmon is Innovation Lead at Intel Israel and a senior engineer dedicated to preparing the engineering workforce for the AI era. His work focuses on ramping up and upskilling teams, ensuring they have the cutting-edge methodologies and know-how needed to excel in this rapidly evolving field.

Orit Hazzan is a professor at the Technion’s Department of Education in Science and Technology. Her research focuses on computer science, software engineering, and data science education. For additional details, see https://orithazzan.net.technion.ac.il/ .