What's this about?
Laude released Headlong, a deliberately minimal harness for agents that need to keep running over long periods rather than finishing a single task and exiting.
What exactly is a "persistent agent"?
Most agents are one-shot: you ask, they work, they stop. A persistent agent stays alive across hours or days, keeps its own notes, and picks work back up. The hard part is not the model — it is what the agent remembers, how it survives a restart, and how it avoids drifting off task over a long horizon.
What's the community saying?
HN 119 points, 53 comments
The interest centres on the "micro" part — a reaction against increasingly heavy agent frameworks, with commenters arguing that a small harness you fully understand beats a large one you do not.