Introducing AI Agents: Autonomous Orchestration with Declarative Workflows

AI Agents in Kestra 1.0 introduce a new paradigm for workflow orchestration, shifting from static task lists to dynamic decision engines powered by Large Language Models (LLMs), memory, and tools. These agents autonomously determine actions and their order, allowing for adaptable and intelligent workflows that can reason, adapt, and make decisions on the fly. Unlike traditional workflows that require pre-defined sequences, AI Agents use system messages, prompts, memory, and a suite of tools to interact with external systems and maintain context across executions. They support multiple LLM providers and memory backends, offering flexibility and adaptability in creating declarative workflows. While AI Agents enhance dynamic orchestration, they are non-deterministic, making them unsuitable for mission-critical tasks requiring strict control and repeatability. Kestra offers a balanced approach by combining traditional task-based orchestration with AI Agents, maintaining workflows that are declarative, observable, and managed as code.