Sequential
Linked nodes in a horizontal pipeline; each output feeds the next.
Verified against anthropic.com
A sequential agent runs as a fixed pipeline: each node takes the previous node's output, transforms it, and passes it on, in a set order, until the last node produces the result. There is no re-planning and no loop; the path is decided in advance. It is the simplest and most predictable shape, because every run follows the same steps, which makes it easy to trace, test and cost, but unable to adapt when a step's output is surprising. In anatomy terms it is a chain of scoped reasoning or effector stages with no central orchestrator.
A fixed chain of stages, each a scoped model call or effector, wired output-to-input; no orchestrator and no feedback loop; the perimeter gates any external write within the chain.
- ·Document pipelines: extract, classify, summarise, route.
- ·Voice and media pipelines: transcribe, analyse, respond (Deepgram, ElevenLabs, Abridge).
- ·Compliance flows where the steps are fixed and must be auditable end to end.
- ·Maximally predictable and traceable: the same steps every run.
- ·Easy to test, cost and reason about.
- ·No orchestration overhead.
- ·Failures localise to a single stage.
- ·Cannot adapt when a step's output is unexpected.
- ·A single stage failure breaks the whole chain.
- ·No parallelism: latency is the sum of every stage.
- ·Rework means re-running from the failed step.
Mapping classified by the Field Book from each agent's documented control flow.
Description, use cases and trade-offs: Field Book analysis, synthesised from established agent-architecture literature.
The corpus does not rewrite vendor documentation. Where copy is quoted verbatim it is marked as such and attributed to its source.