A series of small manageable chunks becomes a long running task :)

If LLMs are going to act as agents, they need to maintain context across these chunks.