Pipeline
A pipeline is a linear, ordered sequence of processing stages that data or a task flows through, from an initial input to a final output.
Pipeline is the tightest, most literal member of the workflow family. Where “workflow” can loosely cover human steps and looser processes, “pipeline” usually implies something closer to a fixed, code-defined chain: input goes in one end, passes through a series of well-defined stages in a fixed order, and a transformed output comes out the other end.
The term comes from data engineering and CI/CD, and it carries over to agents mostly unchanged: an agent-based pipeline might take a raw document, run an agent step to extract structured data from it, run a validation step to check the extraction, and output the validated result. Because each stage has a clear input and output, pipelines are relatively easy to test stage-by-stage and to reason about.
The limitation is exactly what makes it useful: a plain pipeline is linear. If a real process needs parallel branches that later converge, that's a dag-workflow, not a plain pipeline.
Example
A resume-screening pipeline: input a PDF resume → transform (extract text and structured fields) → validate (check required fields are present and dates parse) → output (a structured candidate record).
How it differs
Pipeline vs. workflow: pipeline usually means a stricter linear chain of processing stages, closer to a fixed program; workflow is the broader umbrella term and comfortably includes human approval steps or looser, less code-defined processes.
Common misconceptions
FAQ
What is a pipeline in an AI agent context?
Is a pipeline the same as a workflow?
Last checked: 2026-08-28