Compare Patterns

Sort, filter, and explore every entry across key attributes. Click a row to expand details.

Cost vs. latency map

Every entry plotted by cost and latency. Hover a bubble for its description; click to open its page.

Sweet spotLow cost · low latencyBatch friendlyLow cost · slow OKPremiumSpend money to be fastJustify itSlow and expensiveLowMediumHighVariableLatency →LowMediumHighVariableCost ↑Agentic RAGSagaEvaluator-OptimizerLong-HorizonMulti-AgentPlan & ExecuteReflectionEvent-DrivenOrchestrator-WorkerRAGMemorySub-agentsGuardrailsParallel CallsPrompt ChainingReActRoutingSkillsTool UseHuman in the Loop
Cohorts:workflowsagent patternsprimitivesmodifiers·bubble size = complexity
20 patterns
Pattern Category Complexity Latency Cost Best For
AgentAdvancedHighMediumCompound or multi-source questions where grounding and citations matter
AgentAdvancedMediumMediumAsync reactive systems on a queue or stream
AgentAdvancedHighHighTasks that span hours to weeks with checkpoint-and-resume across deploys
AgentAdvancedHighHighEnterprise systems with parallel specialization
AgentIntermediateHighHighComplex multi-step tasks needing upfront planning
AgentIntermediateMediumMediumKnowledge-intensive Q&A and generation
AgentIntermediateVariableMediumOpen-ended tasks requiring tool use
AgentIntermediateHighHighHigh-quality outputs needing self-critique
AgentBeginnerLowLowMulti-intent systems with specialized handlers
AgentAdvancedHighMediumLong-running workflows requiring compensation on failure
AgentAdvancedMediumMediumLayered policy checks plus a dual-LLM split against prompt injection
AgentIntermediateVariableLowHigh-stakes actions requiring human approval or correction
AgentIntermediateMediumMediumSessions requiring context persistence
AgentIntermediateLowLowRepeatable in-context procedures the agent should perform consistently across runs
AgentIntermediateMediumMediumDelegating bounded sub-tasks to role-scoped agent instances
AgentBeginnerLowLowStructured API calls and function execution
WorkflowIntermediateHighHighQuality-sensitive outputs needing iteration
WorkflowIntermediateMediumMediumComplex tasks with dynamic decomposition
WorkflowBeginnerLowMediumIndependent sub-tasks, aggregated results
WorkflowBeginnerMediumLowSequential, predictable transformations

Not sure which pattern to use?

Answer a few questions to get a pattern recommendation.

1
Question

Is the trigger an external event (cancellation, status change, scheduled job) rather than a user request?