The author discusses the challenge of learning new skills in the age of AI agents that can write code. The core concept is the importance of "friction" in the learning process. AI, like slot machines, offers the path of least resistance, potentially leading to an abdication of thinking. Consequently, the author intentionally removes AI assistance from their coding platforms for learning. True learning requires going through the process yourself before delegating to an agent. This is why AI is more useful for explaining concepts than for performing tasks one intends to learn. Silent errors can occur in AI-generated code, especially for beginners who haven't developed a deep understanding. Learning by building, with tests as a guide, is crucial for retaining ownership of the code. This involves reading enough to grasp concepts, choosing a real project, and writing tests first. The author emphasizes that AI should be an accelerator, not a compass, and a high bar should be maintained for accepting AI-generated code. Ultimately, AI can enhance software engineering, but it does not change the fundamental nature of the discipline.