Shortcut or Setback? GenAI in Coding Courses

Students often use GenAI to study complex subjects. However, frequent AI users generally become exhausted trying to get good outcomes (Simkute et al., 2025). And, users with high confidence in AI tend to think less critically (Lee et al., 2025). Similarly, what if the use of GenAI prevents students from attaining learning objectives and makes concepts more difficult to understand? E.g., when first year business school students use GenAI to practice problem-solving skills to analyze a problem and learn to create simple Python code to solve it, they must ideally 1) derive the steps (and draw a flow diagram) to solve the problem; and 2) use the steps and flow to develop basic Python code. With generative AI tools increasingly integrated into compilers, teachers soon realized that students, to do exercises in a perceived ‘easier’ and ‘faster’ way, use the built-in GenAI tool in the compiler (Google Colaboratory) and then try to complete the rest of the work in the reversed order—by trying to decipher the steps and information flow from the GenAI generated code. However, it means that 1) students do not achieve the learning objective to analyze and deconstruct a problem; and 2) since GenAI code is often more complex than expected, and the course does not aim to teach advanced coding, but basic coding structures, it confuses them rather than let them practice basic coding. In this illustrated example the aim is to derive a sequence of steps to implement in Python: “Ask a user if they want to generate a random number. While the user says ‘yes’, generate and print a random number between 1 and 100 to the screen.” The expected (basic) code is on the left in Figure 1. In giving the instructions to Google Colab’s GenAI assistant, it creates the code to the right of Figure 1. This overly complex code uses redundant steps and, in using it to retrospectively analyze the problem, the students become: 1) confused as the code shows unknown elements; and 2) unable to derive the sequenced steps (since they do not deconstruct the problem but instead decipher the code).