Comment by devonkelley

This is the framing most people in this thread are missing. The difference between "agent that does stuff for a company" and "agent that does stuff for me" isn't technical, it's about who defines the tool chain and who the agent is optimizing for. Every multi-agent system I've built has this tension buried in it. The agent's behavior changes dramatically based on whose goals are baked into the system prompt vs whose goals come in at runtime.