Building Mission-Critical AI Workflows

​Visual workflow builders are great for prototyping AI workflows. Add retries, conditional branching, and parallel LLM tasks, and they become hard to reason about and nearly impossible to debug.

​Harsh Mer, AI Data Engineer, ran into this building a multi-step AI pipeline: fetch from APIs, pass through an LLM, trigger follow-up actions. Simple flows worked fine. When he added real production requirements, the visual model broke down. He moved the pipeline to Kestra and the difference was immediate: declarative YAML, proper error handling, full observability without additional tooling.

​Join Harsh and Will Russell (Developer Advocate, Kestra) for a session on what he learned building AI workflows that hold up under production conditions. You’ll learn how to:

• ​Identify what separates a working AI prototype from a workflow you can feel confident deploying in production

• ​Structure multi-step AI pipelines with retries, error handling, and parallel execution using declarative YAML

• ​Debug and observe AI workflows in production using execution logs and task-level visibility