Reliable Agentic Systems need Durable Execution 🤖
summary
Claim: durable execution via Dapr Workflow lets long‑running agentic pipelines resume from checkpoints and avoid re‑calling LLMs. Live demos run locally with a Dapr sidecar, Redis state store, .NET 9 apps and a Llama 3.2 model; the five‑stage spatial anomaly workflow, routing to a temporal‑rift handler, and parallel starship diagnostics (with voting) all complete and return specific outputs like “wormhole”, “temporal rift”, urgent repairs and a 2,550‑day construction estimate. Patterns shown: Prompt Chaining, Routing, Parallelization, Orchestrator‑Workers, Evaluator‑Optimizer.
AI-written summary · about
chapters (19)
- 0:00Reliable agentic systems intro
- 2:15What defines agentic systems
- 4:22Durable execution importance
- 6:50Deer project overview
- 9:17Deer workflow engine deep dive
- 14:59Deer conversation API
- 16:44Workflows with LLMs
- 18:05Demo setup and tools
- 20:15Prompt chaining demo
- 26:35Prompt chaining code walkthrough
- 30:53Routing pattern use cases
- 32:38Galactic anomaly classifier
- 36:29Parallelization for agents
- 39:19Starship diagnostics workflow
- 43:19Orchestrator worker pattern
- 47:13Space colony planner
- 51:27Evaluator optimizer pattern
- 54:57Alien translator workflow
- 58:29Key takeaways
related talks
48mFailure is not an option: durable execution + Dapr = 🚀
20mEnhancing Workflow Reliability with Dapr
33mFrom Prototype to Production: Making Agentic AI Systems Reliable
39mDapr in 2025: Durable Execution for AgenticAI Systems and Becoming the Backbone of App Platforms
45mDapr Workflows: How to Write Reliable, Fault-Tolerant Applications...
24mDurable Agents: long running AI workflows in a flakey world
20mBuilding Deterministic Infrastructure for Non Deterministic AI Agents
25mDapr Workflows: The Key to Simplifying Business Process Orchestration