Mathieu Stennier

No pre-registration required.

Agentic orchestration is easy to demo, and hard to ship safely. In this masterclass, we’ll take a realistic proof-of-concept use case and walk step-by-step through the blueprint required to deploy it reliably in an enterprise production environment with Camunda.
You’ll learn practical strategies across the full lifecycle:

• Agent design that behaves predictably
  • Patterns for structuring agent decisions, guardrails, and human-in-the-loop (HIL) checkpoints
  • When to use tools vs. workflows vs. prompts—and how to keep behavior consistent over time
  • Bring enterprise knowledge base as a first-class citizen to build grounded and hallucination-free agents.
• Reliable integrations and enterprise-grade workers
  • Building robust workers and connectors for real systems (ITSM/CRM/data platforms)
  • Handling retries, timeouts, idempotency, and failure modes without breaking the business process
• Security and compliance from day one
  • Secrets management and key rotation
  • Access control and least-privilege tool execution
  • Data protection, environment isolation, and secure context handling (RAG, logs, prompts)
• Testing and evaluation you can trust
  • Offline testing: validate agent flows before production (simulations, golden sets, regression tests)
  • Online evaluation: measure quality and risk in real traffic (hallucinations, leakage, LLM-as-a-judge, HIL effectiveness)
• Observability and operational readiness
  • End-to-end process observability (what happened, why, and who approved what)
  • LLM usage monitoring (cost, latency, token spikes)
  • Business KPIs: automation rate, goal completion, auditability, and incident response readiness
• Deployment, versioning, and safe evolution
  • Controlled rollouts, versioned prompts/skills/workers, and backwards compatibility
  • How to iterate without losing governance—or breaking working automations

Takeaway

You’ll leave with a clear production checklist and reference architecture to move from POC to enterprise deployment—safely, measurably, and with full operational control.

Reliable AI automation requires more than strong models. It requires governed processes, secure runtimes, and deep visibility. This presentation shows how Camunda’s BPMN-based agentic orchestration and Amazon Bedrock AgentCore work together to deliver production-grade, multi-agent automation at scale. You will see how to connect Camunda AI Agents with Bedrock AgentCore to model guardrails, manage complex tasks, and monitor behavior with advanced observability. This session provides a practical blueprint for building governed, scalable AI agents on AWS.