Skip to content

METHODOLOGIE

Orchestration d'agents

The practice of directing multiple AI agents through a task: chaining them, tracking their state, and recovering when one fails. It requires knowing agents' failure modes to compose them reliably. Practitioners note the open-source layer for registry, lifecycle, permissions, and skills is still largely missing.

Origin

A practitioner term for directing and chaining multiple agents, noting the open-source layer for this remains incomplete.

definition
Declarative or coded agents, reusable, with routing between LLMs depending on the problem
gap
Missing open source building block (registry, lifecycle, permissions, skills)
role
Know agents' failure modes, chain them, catch an error before it propagates

Fiches (3)