“A data scientist agent, an engineer agent, a product manager agent”—this kind of multi-agent design, which simply copies human organizational structures, may be flawed from the outset.
This featured clip from AI on Air is excerpted from a recent episode of The Cognitive Revolution Podcast, hosted by Flo Crivello, founder and CEO of Lindy. He argues that in most cases, capabilities should be consolidated into a single agent as much as possible, rather than split across multiple agents.
▷ Human organizations need division of labor because each person has only 24 hours a day and limited capacity for context; agents don’t have these constraints. They can branch off and be copied at any time, and do as much as they want in a day.
▷ So “division of labor” isn’t a good reason to have multiple agents. People often anthropomorphize too much and, by intuition, build multi-agent systems with far more agents than optimal.
▷ He also adds that this isn’t an absolute rule: there are plenty of good reasons for multi-agent systems, but division of labor isn’t one of them.🪁
This featured clip from AI on Air is excerpted from a recent episode of The Cognitive Revolution Podcast, hosted by Flo Crivello, founder and CEO of Lindy. He argues that in most cases, capabilities should be consolidated into a single agent as much as possible, rather than split across multiple agents.
▷ Human organizations need division of labor because each person has only 24 hours a day and limited capacity for context; agents don’t have these constraints. They can branch off and be copied at any time, and do as much as they want in a day.
▷ So “division of labor” isn’t a good reason to have multiple agents. People often anthropomorphize too much and, by intuition, build multi-agent systems with far more agents than optimal.
▷ He also adds that this isn’t an absolute rule: there are plenty of good reasons for multi-agent systems, but division of labor isn’t one of them.🪁