Runway entrusts the construction of its pipelines to its conversational agent

Runway Agent now builds complex generation pipelines from simple text prompts. What this control means for mass creation.

Runway is connecting two building blocks it previously offered separately. Its Workflows—node-based pipelines where multiple generations are chained together to maintain consistent quality at scale—can now be controlled in natural language from Runway Agent, via a dedicated command. You describe the desired pipeline, and the agent builds the graph, connections, and settings, and can then execute it.

The system operates under three use cases. On command, the agent builds a sequence from an instruction—for example, starting with a product photo and a description to produce a short video ad—with the structure being submitted for approval before creation. Upon execution, it maps the provided inputs, requests what is missing, estimates the credit cost, and launches. It can also retrieve an existing workflow, indicate the inputs it requires, or modify a setting before running it again.

The benefit lies in joining two approaches that the developer previously kept separate. The classic node-based interface offers fine-grained control but requires manual patching, while the conversational agent prioritizes simplicity. Using natural language to assemble the graph aims to retain the former without the setup overhead. Furthermore, a pipeline built by the agent can still be opened in the classic editor for manual adjustment.

Building a workflow does not consume credits; only its execution does, with an optional confirmation step. The feature requires a Standard plan or higher to launch a pipeline from the agent.