Design Notes on Memory, Context, Orchestration, and Self-Directed Evolution in AI and AGI Systems

This paper presents opinionated design notes on memory, context management, orchestration, and self-directed evolution in AI and AGI systems. In this context, AI and AGI are primarily discussed with current LLM-based systems in mind. As interest in Artificial General Intelligence (AGI) continues to grow, it becomes increasingly important to examine not only model scaling or training methods, but also how systems manage memory over long time horizons, revise their own internal structures, and coordinate complex internal processes. Rather than proposing a finalized architecture or implementation, this work explores a set of aggressive design hypotheses, including memory maps, selective context retrieval, controlled and extensible orchestration, modular skill representations, multi-model utilization, and self-directed internal updates. Related links and updates are available at: https://hajimetwi3.github.io/AGI-Systems/