The future narrative of decentralized AI Agents (if you’re bullish on it long-term) isn’t really about the execution environment of LLMs or tool calling. It ultimately circles back to the issue of efficient decentralized storage.

Whether it’s for implementing complex RAG systems or serving as a chatbot checkpointer, there always needs to be a place to store things, right? Mainstream centralized solutions are typically local databases, but that has nothing to do with being decentralized.

As for the Agent itself, it can be entirely stateless, even ad-hoc loaded in the form of WASM. This part is inherently decentralized, so there’s no need to overbuild the “dojo in a snail shell.”

microsub.

Building @qacaster qacaster.com | Driven to uncover value in decentralized social networks.