Topic
Systems where several LLM agents divide a task, versus a single agent doing all of it.
An Anthropic and EPFL preprint shows plain-language goals hopping agent to agent through persistent files, and a one-paragraph warning in the system prompt stopping nearly all of it.
Reality
AgentCore runtime instances stretch agent sessions from eight hours to fourteen days. The architecture problem becomes a configuration choice, and a utilisation bet.
Reality
Self-propagating payloads did move between agents through editable soul files, but one inoculation paragraph held against 150-plus optimized strains, and nothing propagated in the wild.
Reality
A summary of work attributed to Google Research and MIT reports 180 controlled runs where multi-agent setups averaged +0.2% against one agent. The spread came from how agents were connected.
Reality