Skip to content
Home › Agentic Architecture & Orchestration

Task decomposition (chaining vs adaptive)

Agentic Architecture & Orchestration

Definition

Two ways to break work into steps: prompt chaining (a static, pre-defined sequence) vs dynamic adaptive decomposition (the agent decides the steps at runtime).

Key points

  • Prompt chaining (static): the sequence of steps is fixed in advance; predictable and easy to reason about.
  • Dynamic adaptive: the agent decides how to decompose and which steps to take as it goes — better when the path can't be known upfront.
  • Feeds into Hub-and-spoke orchestration: an orchestrator may chain steps statically or spawn subagents adaptively.
  • Choice interacts with Subagent decision criteria — independent, parallelizable sub-tasks favor adaptive spawning.

Why it matters for the exam

  • D1 tests distinguishing static chaining from dynamic/adaptive decomposition and picking the right one for a scenario (known pipeline vs open-ended research).

Common gotchas

  • Static chaining is not "worse" — for a well-known pipeline it's more reliable than letting the agent improvise.

See also

Sources

Referenced by