Compare the Six Patterns

Side by side on control flow, cost/latency, best fit and a representative example. Filter by kind, sort any column.

PatternKindControl flowCost / latencyBest forExample
Prompt chainingWorkflowFixed sequence, gates between stepsLow–mediumCleanly decomposable fixed subtasksWrite copy → translate it
RoutingWorkflowClassify, then branch to a handlerLow (+1 classify)Distinct categories handled separatelyTriage support tickets
ParallelizationWorkflowFan out, then aggregate (sectioning / voting)Higher (×N calls)Independent subtasks; multiple attemptsMany prompts vote on code safety
Orchestrator–workersWorkflowCentral LLM splits subtasks at run timeHighSubtasks can't be predicted up frontEdit code across many files
Evaluator–optimizerWorkflowGenerate → evaluate → refine, loopedMedium–high (loop)Clear criteria; refinement adds valueLiterary translation with feedback
Autonomous agentAgentLLM directs itself in a tool loopHighest, variableOpen-ended; steps can't be hardcodedCoding agent solving SWE-bench