coordinationmediumemerging

Consensus-Based Decision Pattern

Multi-agent collective decision-making with deliberation or voting

Overview

The Challenge

Multi-agent systems need to make collective decisions, but single-agent decisions can be biased or incomplete. Direct voting can be brittle, and debate-based approaches do not scale well.

The Solution

Implement structured consensus mechanisms where multiple agents independently generate solutions, then reach agreement through voting, deliberation, or hybrid approaches based on task type.

When to Use
  • High-stakes decisions requiring multiple perspectives
  • Tasks where individual agent errors are common
  • Situations requiring democratic or fair outcomes
  • Knowledge-intensive tasks (use deliberation)
When NOT to Use
  • Time-critical, low-latency requirements
  • Simple factual queries with clear answers
  • When agent diversity is low (similar training/biases)

Trade-offs

Advantages
  • +Reduces individual agent biases
  • +Improves accuracy on complex tasks
  • +13.2% improvement on reasoning tasks (voting)
  • +Transparent decision-making process
Considerations
  • −Higher latency and cost (multiple agents)
  • −Requires tie-breaking mechanisms
  • −Can amplify shared biases
  • −Coordination overhead
Implement this pattern with our SDK
Get RepKit

Deep Dive

Overview

Consensus mechanisms allow multiple agents to collectively reach decisions that are more robust than individual agent outputs.

Approaches

Voting

  • Each agent proposes a solution independently
  • Majority vote determines outcome
  • Best for: Reasoning tasks, factual questions
  • Performance: +13.2% on reasoning benchmarks

Deliberation

  • Agents engage in structured debate
  • Each round: exchange arguments, revise positions
  • Converges when unanimous or iteration limit reached
  • Best for: Knowledge-intensive tasks
  • Performance: +2.8% on knowledge benchmarks

Hybrid

  • Start with independent proposals
  • Deliberate only on contested decisions
  • Vote with weighted confidence scores

Byzantine Tolerance

For mission-critical systems, use Byzantine fault-tolerant consensus:

  • Requires 3f+1 agents to tolerate f failures
  • Ensures agreement even with malicious/faulty agents

References

Example Scenarios

Medical Diagnosis Support

Three specialized medical agents analyze patient symptoms independently, then vote on likely diagnoses. Unanimous agreement triggers high-confidence output; split votes escalate to human review.

OutcomeReduced misdiagnosis rate by 23% compared to single-agent approach
Want to learn more patterns?
Explore Learning Paths
Considerations

Agent diversity is critical - agents with similar training will have correlated errors, reducing the benefit of consensus.

PreviousMarket-Based Coordination Pattern
NextA2A Protocol Pattern
Dimension Scores
Safety
4/5
Accuracy
5/5
Cost
2/5
Speed
2/5
Implementation
Complexitymedium
Implementation Checklist
Multiple diverse agents
Voting/consensus protocol
Tie-breaking strategy
0/3 complete
Tags
consensusvotingdeliberationmulti-agentdecision-making

Was this pattern helpful?

Suggest an improvement