The use of AI-powered tools in software engineering (SWE) has increased significantly, primarily due to advancements in large language models (LLMs). LLMs can generate code from natural language prompts and even produce complete software artifacts. Along with these changes comes a new class of people working in the software industry: citizen developers. Citizen developers generally have no technical background but can produce technical applications or artifacts with the aid of LLMs. Moreover, LLM-powered AI agents are being used across many application areas, and there is a trend towards using such agents to solve all problems. These changes merit consideration of how, by whom, and when SWE tasks should be automated. Throughout this paper, we argue that some problems should be solved with LLMs, while others should not. We point out that the developers' backgrounds matter as much as the problems to be solved in this regard. The perspectives we offer in this paper suggest that future research should consider the limitations of both the users' knowledge and the technology behind the tools.