From FileUploadWorkflow to creative OS: How Layer scaled on Temporal

Layer.ai details their evolution from using Temporal for simple file uploads to building a sophisticated visual pipeline builder on top of it. They use Temporal child workflows for each node in their DAG-based pipeline, with a parent workflow orchestrating execution and using `workflow.wait_condition` against a shared durable data bus (`all_values` dict) to manage inter-node dependencies and data flow. They also highlight the adaptability of their `RunInferenceWorkflow` to handle multiple modalities (image, video, 3D, audio) without redesign, incorporating modality-aware timeouts and parallel post-creation steps. The integration of an MCP connector allowing LLM agents to directly invoke Layer workflows is also discussed as a key advancement for autonomous operations.