Something I’ve updated on with Grok Bot: I think having a team of agents for work is the future. Initially, I yearned for a single super agent and resisted Bot’s multi-agent approach. It’s hard to figure out the “right” setup: the composition of the team, the scope of each agent, etc. But there’s something better about multi-agent I can’t quite articulate yet. One factor is that multi-agent lends itself better to an eventual multiplayer experience with agents. Two is more vague in my mind but the word I keep coming back to is “expressiveness”. Work is increasingly shifting towards building systems vs doing the work by hand, and multi-agent is a much richer medium for expressing the design of that system vs a single agent. Even in the limit if a single agent had infinite context and perfect attention, I feel a multi-agent system would be easier to reason about and debug and thus a better interface long term. Will be writing a longer form post on this with examples once I have more clarity.