Workflow Basics: Automating Multi-Step Tasks
Workflows let you chain multiple AI agent actions into automated pipelines. Instead of manually copying output from one conversation to another, workflows connect the steps for you.
Core concepts
Steps
Each workflow is a sequence of steps. A step defines:
- Which agent runs it
- What prompt to execute
- How to use output from previous steps
Template variables
Steps can reference previous outputs using template variables. For example, step 2 can include {{step1.output}} in its prompt to pass the result from step 1.
Execution
When you run a workflow, each step executes in order. You can monitor progress in real time and review the output of each step.
Use cases
- Research pipeline: Search the web, summarize findings, generate a report
- Content pipeline: Draft an article, review for tone, format for publishing
- Code pipeline: Analyze requirements, generate code, review for bugs
- Data pipeline: Extract data, transform format, generate visualizations