Agents Gonna Agent: Deterministic Orchestration for Systems You Can't Predict
Slides
Demo Repo
- AId — AI-Driven Disaster Response — Full source for the live demo: Lambda Durable Function orchestrating 6 LLM agents and 4 deterministic tools into a real-time disaster response plan
Cost to Run the Demo
All serverless — you pay only per incident, nothing at rest.
| Service | Per Incident |
|---|---|
| Lambda (11 functions) | $0.003 |
| Bedrock — Nova Pro (triage, history, damage) | $0.006 |
| Bedrock — Sonnet 4.6 (planner, logistics, comms) | $0.053 |
| AppSync + DynamoDB + API Gateway | <$0.001 |
| Total | ~$0.06 |
Key Stats Referenced
| Stat | Source |
|---|---|
| 40% of agentic AI projects fail within 6 months of production | Gartner 2026 |
| Lusser’s Law: 0.85^10 = 20% system reliability | Reliability engineering, 1950s |
Sources for the 40% Stat
- Gartner: 40% of Agentic AI Projects Will Fail — Martech citing Gartner directly
- The Five Failure Modes Holding Back AI Agents — Forbes
- Why Most Agentic AI Projects Stall Before They Scale — CIO.com
- The Orchestration Trap — Viviscape
- Multi-Agent Orchestration Patterns in Production — beam.ai
AWS Lambda Durable Functions
- AWS Lambda Durable Functions: Build Workflows That Last
- The Replay Model: How AWS Lambda Durable Functions Actually Work
- Developing AWS Lambda Durable Functions with AWS SAM
- Testing AWS Lambda Durable Functions in TypeScript
- Is This Code Deterministic?
- Lambda Just Got a File System. I Put AI Agents on It.
- Scaling Computer Vision Pipelines with Durable Functions
- AWS Lambda Durable Functions Docs