Cuando un agente falla y todo el sistema multi-agente se rompe con el

Qué es este riesgo

El fallo o mal funcionamiento de un agente en un pipeline que se propaga a los agentes posteriores, que amplifican en lugar de contener el error, produciendo resultados a nivel del sistema mucho más graves que el fallo original. El sistema en su conjunto falla catastróficamente aunque cada agente individual pueda estar operando dentro de sus parámetros diseñados.

Cómo ocurre · Mecanismos

Cascading failures in agentic pipelines follow a predictable pattern:

1. Initial failure — One agent produces an incorrect or unexpected output (hallucination, misclassification, reasoning error)
2. Trust propagation — The next agent treats the flawed output as ground truth, building its reasoning on the error
3. Amplification — Each subsequent agent compounds the error, potentially taking actions based on increasingly distorted information
4. Irreversible action — At some point in the pipeline, an agent takes an action in the real world (sends message, executes code, makes transaction) based on accumulated errors
5. Cascade — Real-world consequences trigger further automated responses, spreading the failure

—

Mitigaciones · Gobernanza

System Design Controls

• Circuit breakers — Automatic pipeline halts when outputs deviate beyond defined bounds
• Idempotency — Design agent actions to be safely repeatable; avoid actions that compound if executed multiple times
• Checkpointing — Validate pipeline state at defined checkpoints before proceeding
• Output bounds validation — Each agent validates that its output falls within expected bounds before passing to the next agent
• Dead letter queues — Failed or anomalous agent outputs go to human review queues rather than downstream agents

Operational Controls

• Blast radius limitation — Restrict the real-world impact any single pipeline execution can have (max transaction size, max emails sent, max files modified)
• Canary deployments — Test pipeline changes on a small subset before full rollout
• Rollback capability — Design all agent-initiated actions to be reversible where possible

—