factory-basics

Coordinating parallel and sequential tasks across multiple AI agents requires a programmable orchestration runtime to manage execution. You can spawn subagents and compose results across complex workflows using fan-out, pipelines, and synthesis patterns.

Multi-agent orchestration is a structured programming method for AI agents to execute distributed tasks and return aggregated results. It uses pattern-based execution for scalable automation, enabling end-to-end agent-driven workflows for data gathering, analysis, and reporting.

To spawn subagents and compose results, utilize a runtime API featuring spawn and join commands. This executes fan-out workflows where multiple subagents perform distributed tasks and return an aggregated synthesis of the combined results.

Yes, you can build end-to-end agent-driven workflows for data gathering and reporting through pattern-based orchestration. This involves spawning researcher subagents to collect data and synthesizing the aggregated results into a final comprehensive report.

Required runtime APIs for context propagation and session management include spawn, join, parallel, sequence, workspace, observe, and shutdown. These commands provide the programmable runtime needed to handle robust error handling and state coordination across agents.

Avoid pattern-based orchestration when tasks require simple linear execution without spawning subagents or coordinating parallel tasks. It is specifically designed for complex fan-out and pipeline workflows requiring robust error handling and context propagation.