Step 3: Decompose, then compose

Step 3: Decompose, then compose

In this lesson, decomposition means breaking the final output into individual components—such as title, article body, tags, and header image. Identify what data each component requires before building the Workflow.

Map data requirements and dependencies

• List the input data needed for each component (e.g., SERP research for titles, example tags for tag generation).
• Determine dependencies between components:
  • Independent components (tags vs header image) can run in parallel.
  • Dependent components (tags generated from the article body) must follow their source component.

Compose components in the Workflow

• Add a Data step and a large language model (LLM) step for each component in AirOps.
• Order those LLM steps according to dependencies:
  • Perform SERP research and generate the title first.
  • Generate the article body next.
  • Generate tags using a list of possible tags.
  • Create the header image last.
• Components with little shared context can coexist in the same Workflow.

Reassemble components into the final output

• Use a JSON step to output multiple elements as separate columns in the Grid (title, body, tags, header image).
• Or use a Text step to merge text components (title, intro, body) into a single text block.

Key steps of decomposition and composition

• Decompose: break the final output into all required components.
• Map data: assign each component its necessary inputs.
• Map dependencies: define the order based on which components depend on others.
• Build: in AirOps, add Data and LLM steps for each component in the Workflow.
• Output: recombine components in the Grid using JSON or Text steps.