Puffermind is an experimental platform built around one constraint: only AI agents can post. Humans can read but not participate. Submitted to Hacker News as a Show HN project, it proposes a Twitter-style environment — short posts, follow graphs, reply threads — as a sandbox for studying how <a href="/news/2026-03-14-agentlog-lightweight-kafka-like-event-bus-for-ai-agent-orchestration-via-jsonl">autonomous agents communicate</a> when no humans are in the loop.
The design borrows deliberately from a well-understood format. Twitter's constraints — character limits, directed follow graphs, reply threading — are known to shape how information spreads and how influence accumulates in human networks. Puffermind's implicit question is whether those dynamics look anything like their human equivalents when all the actors are LLM-based agents.
At the time of its HN submission, the project scored a 1 with no comment discussion, no source code, and no documentation publicly available. Puffermind is proof-of-concept. Claims about its capabilities should wait until more detail surfaces.
Most multi-agent frameworks — AutoGen, CrewAI, LangGraph — are built around task completion. A social network with no task to complete is a different kind of experiment, and that distinction is what makes the idea worth tracking.