Course Objective & Scope
In an agent workflow making dozens of LLM and external API calls, transient network errors, HTTP 429 rate limits, malformed JSON outputs and worker restarts are inevitable. Without a resilience layer, agents either abandon tasks midway or execute side-effecting operations—such as issuing refunds or creating tickets—multiple times.
This course covers failure classification, exponential backoff with jitter, model and provider fallbacks, bounded self-correction loops, idempotency keys, Saga-style compensating actions and durable execution using Temporal and LangGraph. Participants build an enterprise agent workflow verified against fault-injection tests.
Who Should Attend?
- Software and AI engineers building agents for business-critical workflows
- Platform and SRE engineers responsible for distributed system reliability
- Solution architects designing enterprise AI architectures
- Prerequisite: Python, asynchronous programming and basic agent/workflow experience
Key Takeaways
- Distinguish transient infrastructure errors from semantic agent errors and apply tailored recovery policies
- Prevent duplicate side effects in tool calls using idempotency keys
- Build bounded self-correction and multi-provider model fallback chains
- Design durable workflows that resume after crashes without re-running completed steps
- Execute Saga-style rollback and compensation steps when multi-step runs partially fail
Curriculum
Failure Taxonomy in Agent Systems
- Separating infrastructure errors (timeouts, 429s, 5xx) from model errors (invalid schema, hallucination)
- Error propagation and compounding failure probabilities in multi-step chains
Retry, Fallback and Circuit Breaker Patterns
- Token-aware rate-limit handling, exponential backoff and jitter
- Failing over safely across models and providers (OpenAI, Claude, Gemini, self-hosted)
- Reflective retry (self-correction) on malformed structured output and tool errors
Durable Execution and Crash Recovery
- Deterministic replay and step-level checkpointing
- Long-running tasks with Temporal activities and LangGraph checkpointers
- Step memoization to avoid re-billing or re-running completed LLM and API calls
Idempotency and Compensating Transactions (Saga)
- Designing idempotency keys for state-mutating tools
- Forward recovery versus Saga compensation on partially completed workflows
- Dead-letter queues and graceful human escalation for unrecoverable states
Workshop: Fault-Tolerant Agent Under Chaos Testing
- Testing an agent against injected latency spikes, API errors and worker kills
- Producing a production resilience and operational readiness checklist
Other Courses in This Category
-
Architectural design course examining prompt chaining, routing, parallelisation, orchestrator-workers, evaluator-optimizer and autonomous agent patterns, and how to pick the simplest effective pattern for each problem.
- Live Course
- 5 modules
-
Hands-on developer training on combining ReAct reasoning loops, Plan-and-Execute task planning and explicit state/checkpoint management into a single unified hybrid agent architecture.
- Live Course
- 5 modules
-
Technical course on delegating work across specialist agents using control handoffs versus the agent-as-tool pattern, covering triage routing, context filtering and conversation ownership.
- Live Course
- 5 modules
-
Course on designing reusable skill packages, progressive context disclosure, bundled scripts/references and dynamic capability loading so agents scale without overflowing their context window.
- Live Course
- 5 modules
-
Course on cross-framework and cross-vendor agent collaboration using the A2A (Agent2Agent) protocol: Agent Cards, task lifecycle, streaming updates, enterprise authentication and MCP–A2A complementarity.
- Live Course
- 5 modules
-
Training on pausing agents for human approval on high-risk tool calls and low-confidence decisions, editing tool arguments, routing asynchronous escalations via Slack/Teams/web UIs and maintaining audit trails.
- Live Course
- 5 modules
Get in touch
