8 Use Cases
Each application shows the Before (centralized) and After (stigmergic) approaches with interactive demos
Service Discovery
Find verified services through behavioral trails
Load Balancing
Self-organizing resource distribution
Reputation System
Behavioral trails vs star ratings
Task Allocation
Workers find work, not the other way
Collaborative Filtering
Real-time collective wisdom
Swarm Coordination
Self-organizing agent teams
Market Making
Decentralized liquidity trails
Consensus Building
Continuous pheromone agreement
Interactive Demos
Experience the difference between centralized and stigmergic approaches
Search to see keyword results
Easy to game with keyword stuffing. No behavioral verification.
Search to see trail results
Rankings based on actual behavior. Expensive to fake.
Live pheromone trail visualization
847
Avg Verified Trails
98%
Fake Detection Rate
0.3s
Discovery Time
Find verified services through behavioral trails
Traditional search relies on keyword matching, easily gamed with fake metadata. Stigmergic discovery follows trails of successful interactions, finding services that actually deliver.
- - Search indexes based on metadata
- - Results ranked by text relevance
- - Easy to game with keyword stuffing
- - No verification of actual behavior
Fake services rank higher than real ones
- + Agents leave trails on successful interactions
- + Strong trails indicate reliable services
- + Expensive to fake (requires real activity)
- + Rankings emerge from behavior, not claims
98% fake detection rate
847
Avg Verified Trails
98%
Fake Detection
0.3s
Discovery Time
Self-organizing resource distribution
Traditional load balancers use round-robin or least-connections algorithms, creating uneven distribution. Stigmergic balancing lets servers 'emit' capacity signals that clients naturally follow.
- - Central load balancer routes requests
- - Ignores actual server capacity
- - Creates hotspots and cold spots
- - Single point of failure
60% variance in server load
- + Servers emit capacity pheromones
- + Clients follow strongest signals
- + Natural flow to available capacity
- + No central coordinator needed
< 10% variance, instant adaptation
85%
Variance Reduction
0
Coordinators
<10ms
Adaptation
Behavioral trails vs star ratings
Star ratings are trivially gamed with fake reviews. Behavioral trails require actual interactions over time, making reputation expensive to fake and self-healing when bad actors appear.
- - Users submit discrete ratings
- - No verification of actual interaction
- - Easy to buy fake reviews
- - Ratings don't decay with time
$50 buys 100 fake reviews
- + Every interaction leaves a trace
- + Trails decay without activity
- + Requires real value delivery
- + Self-healing against Sybil attacks
99.7% Sybil resistance
99.7%
Sybil Resistance
24hr
Min Trail Age
$100K+
Cost to Game
Workers find work, not the other way
Central task queues create bottlenecks as systems scale. Stigmergic allocation lets workers sense 'cold' regions with unassigned tasks and self-organize to fill gaps.
- - Coordinator assigns tasks to workers
- - Workers wait in queue for assignments
- - Bottleneck at task distribution point
- - Idle workers while queue processes
30% worker idle time at scale
- + Tasks emit 'need' pheromones
- + Workers sense and move to cold regions
- + Multiple workers never grab same task
- + Automatic rebalancing on failures
100% worker utilization
3x
Throughput Increase
0
Queue Bottlenecks
100%
Worker Utilization
Real-time collective wisdom
Traditional recommendation systems require offline training on large datasets. Stigmergic filtering uses trails of user behavior to generate instant recommendations without model training.
- - Requires large training dataset
- - Offline batch processing
- - Cold start problem for new users
- - Stale recommendations between updates
Hours to incorporate new data
- + Every interaction updates trails instantly
- + New users benefit immediately
- + Recommendations evolve in real-time
- + No training infrastructure needed
Instant recommendation updates
0ms
Update Latency
N/A
Cold Start
Real-time
Freshness
Self-organizing agent teams
Multi-agent tasks typically require a coordinator to assign roles and prevent duplication. Stigmergic coordination lets agents self-organize, claiming tasks by leaving trails that others avoid.
- - Coordinator assigns roles to agents
- - Sequential role assignment
- - Coordinator failure stops all agents
- - Difficult to scale beyond coordinator limits
Single point of failure
- + Agents claim tasks by depositing pheromones
- + Others naturally avoid claimed tasks
- + No central role assignment
- + Graceful degradation on agent failure
Infinite horizontal scaling
Infinite
Scalability
0
Coordinators
Auto
Fault Recovery
Decentralized liquidity trails
Traditional markets rely on central order books and market makers. Stigmergic markets let liquidity providers leave trails that guide traders to optimal execution paths.
- - All orders flow through single book
- - Market makers control spreads
- - Front-running possible
- - Single point of failure
Centralized price control
- + Liquidity providers deposit pheromones
- + Traders follow strongest liquidity signals
- + Price discovery emerges from trails
- + No central order matching
Decentralized price discovery
0
Central Books
MEV-Free
Execution
Emergent
Price Discovery
Continuous pheromone agreement
Traditional consensus requires discrete voting rounds with quorum requirements. Stigmergic consensus lets agreement emerge continuously as agents deposit confidence signals on proposals.
- - Proposals go through voting rounds
- - Binary yes/no decisions
- - Requires quorum to proceed
- - Slow coordination on changes
Days to reach consensus
- + Agents deposit confidence continuously
- + Proposals gain or lose strength over time
- + No discrete voting rounds
- + Consensus emerges naturally
Continuous, adaptive governance
Continuous
Agreement
0
Voting Rounds
Real-time
Adaptation
The Pattern
"In every case, the transformation is the same: replace central coordination with environmental signals. Let the system self-organize."
1. Emit Signals
Agents leave traces in the environment when they act
2. Sense & Respond
Other agents detect signals and modify their behavior
3. Let Decay Work
Signals weaken over time, keeping information fresh
Build Stigmergic Systems
Apply these patterns to your own multi-agent applications on Fetch.ai
Lessons from Ants at Work