Imagine an AI that could not only take your programming exams but also grade them. That's the tantalizing possibility explored in a recent research paper examining ChatGPT's performance on real-world, Spanish-language programming tests. The results? A mixed bag. While ChatGPT managed to pass the exam, its performance was comparable to a first-year student, not a seasoned expert. It excelled at basic coding tasks but stumbled on complex data structures and algorithmic reasoning, much like a student still learning the ropes. Interestingly, adding more context and instructions to the prompts – a technique often used to boost AI performance – actually hindered ChatGPT in this case. It consistently scored lower with complex prompts compared to simpler ones, suggesting that too much information can sometimes confuse the model. Even more surprising were the AI's grading skills (or lack thereof). When tasked with evaluating student-written answers, ChatGPT consistently overestimated their quality. Even poorly written answers received inflated scores, making it unreliable as an automated grading assistant. This raises intriguing questions about how AI understands code quality and whether it can be trained to evaluate solutions with the same nuance as a human instructor. While ChatGPT's ability to generate basic code is impressive, this research highlights its limitations in tackling the more complex aspects of programming. It suggests that human instructors won't be replaced by AI anytime soon. Instead, the future likely lies in collaborative learning environments where humans and AI work together, leveraging each other's strengths to improve educational outcomes.