Agentic systems are beginning to reshape scholarly work, e.g., they can now design experiments, develop mathematical arguments, generate full academic papers, peer reviews, and rebuttals. This raises a natural question: to what extent can research itself be automated? The recent history of research automation can be read as a progression from retrieval and indexing, to citation linking, to summarization, and finally to end-to-end agentic systems. This survey examines the agentic systems most directly relevant to that question, namely those whose primary outputs are scholarly artifacts. First, we distill the recurring engineering principles and architectural patterns in current systems. Second, we propose a compact conceptual framework for identifying how to map the systems within the landscape as the field evolves. For the latter, we organize these systems along three dimensions: the phases of the research process a system covers, the kind of artifacts it produces, and the verification regime that governs its principal claims. Because this area is evolving faster than traditional publication cycles can accommodate, we treat the survey as a living document, updated regularly to