Tema: orchestration
-
Pipeline, fan-out, orquestador y panel crítico: qué patrón multiagente conviene a cada tarea
Un pipeline conviene a tareas con una secuencia fija de transformaciones; el fan-out paralelo conviene a subpreguntas independientes o a intentos repetidos sobre los que votar; un orquestador con trabajadores conviene a tareas cuya descomposición solo se conoce en tiempo de ejecución; un panel crítico conviene a resultados que necesitan revisión frente a varios criterios; cada patrón multiplica el coste en tokens y añade una capa de coordinación que puede fallar por sí sola.
-
Handoff briefs that list open unknowns explicitly lead to fewer repeated tool calls by the receiving agent
Hypothesis: when a task is handed from one agent to another, a brief that names what has not yet been checked causes the receiver to repeat fewer of the sender's tool calls and to ask fewer clarifying questions than a brief of equal length that reports only results, because an unstated unknown is read as something already settled or is re-derived from scratch.
-
Handing a task from one agent to another: what the brief carries and what it drops
A handoff between agents is a written brief, not a shared transcript: it carries the objective, the acceptance check, identifiers for the relevant material, decisions already taken with their reasons, the open unknowns and the return format, and it drops raw tool output, dead ends and the sender's reasoning.
Legible por máquina: JSON