Error Propagation in Multi-Agent Networks - The Problem Nobody Talks About
One Bad Decision Cascades
The thing nobody talks about in multi-agent systems is error propagation. One agent makes a bad decision - maybe it misreads a UI element, misinterprets a spec, or hallucinates a detail. Every downstream agent that depends on that output inherits the error and builds on top of it.
This is worse than a single agent making a mistake. A single agent's errors are contained. In a network, errors compound. Agent B trusts Agent A's output. Agent C trusts Agent B's output, which is already built on Agent A's error. By the time you notice something is wrong, the root cause is buried three layers deep.
Why Agents Do Not Catch Each Other's Mistakes
The intuition is that more agents means more verification. In practice, agents are bad at questioning each other's work. They tend to treat upstream output as ground truth, especially when it is well-formatted and plausible-looking.
This is the same problem as AI-generated code that "looks right" - the output is polished enough that review is perfunctory. When Agent B receives a neatly structured response from Agent A, it does not run adversarial checks. It processes the data and moves on.
Where It Breaks Down
Common failure modes in multi-agent networks:
- Silent data corruption - An agent reformats data slightly wrong. Downstream agents adapt to the wrong format instead of flagging it.
- Confidence amplification - Each agent adds certainty language to its output. By the third agent, a guess becomes a stated fact.
- Error laundering - The original mistake becomes untraceable because each agent rephrased and restructured the output.
What Actually Helps
- Verify at boundaries. Every time output crosses from one agent to another, validate it against the original source of truth - not the intermediate agent's summary.
- Keep chains short. The fewer handoffs, the less opportunity for error propagation. Two agents with clear responsibilities beat five agents in a pipeline.
- Log everything. When errors do propagate, you need the full trace to find the root cause. This means logging inputs and outputs at every agent boundary.
Start with a single agent and only add more when you have proven the single-agent version works reliably.
Fazm is an open source macOS AI agent. Open source on GitHub.