The multi-agent sequential pattern executes a series of specialized agents in a predefined, linear order, where the output from one agent is the direct input to the next.1
Diagram
Open the interactive version here to pan and zoom.
How it works
Orchestration is handled by a sequential workflow agent that operates on predefined logic — it does not consult an AI model to decide what runs next. That distinction is the whole economic case for the pattern, and it’s shared with the parallel and loop patterns. Contrast it with the coordinator pattern, where a model does the routing.
When to use it
Use it for highly structured, repeatable processes where the sequence of operations doesn’t change. The source’s example is a data processing pipeline: a data extraction agent pulls raw data, passes it to a data cleaning agent for formatting, which passes the clean data to a data loading agent that saves it to a database.
From the pattern-comparison guidance, the workload profile is:
- multi-step tasks following a predefined, rigid workflow,
- no need for model orchestration,
- a fixed sequence where each agent’s output is the next agent’s input.
Trade-offs
Compared with any pattern that uses a model to orchestrate, this one reduces both latency and operational cost — there are no reasoning calls spent on deciding what happens next.
The price is flexibility. A rigid, predefined pipeline is hard to adapt to dynamic conditions and can’t skip steps that turn out to be unnecessary, which causes inefficient processing and can raise cumulative latency when an unneeded step happens to be a slow one.
Footnotes
-
Samantha He, “Choose a design pattern for your agentic AI system”, Google Cloud Architecture Center, last reviewed 2026-05-28. Local copy:
references/google-cloud-agentic-design-patterns.md. ↩