Suppose several agents assess the same case and return different answers. Majority voting selects the answer that receives the most votes. In a voting/best-of-N ensemble, the candidates can come from several agents or repeated runs of one agent. section 4.1.3
Voting differs from iterative debate: candidates need not discuss and revise their answers before selection. The selection rule can also change: a weighted vote gives some agents more influence, while a judge agent chooses an answer. section 4.1.3
A diagnostic decision
Several agents assess one input in a diagnostic system. If most return the same answer, a majority vote selects it rather than relying on one agent’s judgment. This can outvote an isolated error, but agreement is not proof: agents with the same mistaken information may vote for the same wrong answer. section 4.1.3
Before adding a vote, check the trade-offs:
- Use it when a problem is ambiguous or multiple assessments could improve reliability. section 4.1.3 section 4.1.3
- Seek genuinely different viewpoints; repeated, near-identical agents can repeat the same mistake. section 4.1.3
- Skip it when one agent can easily produce the correct answer, or when extra runs cost too much time or computation. section 4.1.3
Table 4.4 Architectural patterns overview
| Pattern |
When to use |
When not to use |
|---|---|---|
| Sequential pipeline, agent to agent |
Well-defined, ordered tasks need to be executed in sequence. |
Not all agents need to be executed every time. |
| Parallel delegation |
Multiple tasks or agents can be executed in parallel to improve performance. |
|
| Hierarchical coordination |
The user interface agent delegates to multiple agents on behalf of the user. |
No need for complex decision-making or command of agents |
| Iterative debate and refinement |
You want to review and double- or triple-check responses. |
The problem is trivial or aligns well with simple execution. |
| Voting/best-of-N (ensemble) |
Ambiguous problems that may have multiple interpretations or facets |
Unambiguous problems |
| Role-playing collaboration |
High creativity and novel outputs are desired and align well with the principal goals. |
Real-time systems or cases with low desired creativity |
| Conditional routing (branching) |
Customer support systems that may require human approvals or intervention |
|
| Peer-to-peer network |
Decentralized command increases system robustness and reduces or eliminates complete failure. |
The architectural patterns table places voting/best-of-N alongside other coordination options and contrasts ambiguous problems with unambiguous ones. section 4.1.3
