Imagine a world where teachers no longer spend hours crafting coding exercises, but instead, have an AI assistant generate them on demand. This is the promise of Large Language Models (LLMs) like ChatGPT and Codex, explored in a recent research survey. The study dives into the current state of using LLMs to automatically create programming exercises, examining their strengths and weaknesses. Turns out, these AI models can already generate functional and novel exercises, potentially saving educators valuable time and enabling personalized learning. However, there are challenges. One key issue is ensuring the generated exercises are actually challenging for students, as LLMs can also easily solve them! The research also proposes an evaluation matrix to help educators choose the right LLM for their needs, considering factors like cost, data privacy, and the quality of generated code. This matrix, along with a proposed benchmark called the Programming Exercise Generation Benchmark (PEGB), aims to provide a standardized way to assess the effectiveness of different LLMs for this task. The future of coding education might involve AI assistants generating personalized exercises, but ensuring these exercises are robust, challenging, and effectively contribute to learning remains a key focus.