AI · Framework
Building Effective Agents: Workflow and Agent Patterns
Architectural framework categorizing LLM workflows and autonomous agent patterns
An architectural taxonomy based on Anthropic's research that classifies LLM applications into structured workflows and autonomous agents. It breaks down design patterns ranging from basic prompt chaining to dynamic tool-use feedback loops, helping developers choose the simplest reliable architecture for a task.
The framework
Three categories:
1. Workflows — LLM is embedded in predefined code paths
- Prompt chaining: In → LLM call → LLM → Out
- Parallelization: In fans out to two LLM calls, each → Out
2. Workflows — LLM directs control flow through predefined code paths
- Orchestrator-Worker: Orchestrator routes to multiple Workers, then a Synthesizer produces output
- Evaluator-optimizer: Generator looped with an Evaluator
- Routing: Router sends input to one of several LLM calls
3. Agent — LLM directs its own actions based on environmental feedback: In → LLM call → action on a Tool → Out, with a feedback loop back into the LLM.
All patterns are built from scratch in LangGraph (video: https://youtu.be/aHCDrAbH_go, tutorial: https://langchain-ai.github.io/langgraph/tutorials/workflows/); original post: https://www.anthropic.com/research/building-effective-agents
What it's for
- Choose between deterministic workflows and autonomous agent architectures for an LLM application
- Implement iterative quality control loops using the evaluator-optimizer pattern
- Design parallelized or routed multi-step prompt chains in frameworks like LangGraph
- Structure tool-use loops where LLM actions depend on runtime environment feedback
Key points
- Three-tier classification system separating static workflows, dynamic routing, and autonomous agents
- Sequential prompt chaining pattern for multi-step deterministic tasks
- Parallelization pattern for concurrent LLM processing and aggregation
- Orchestrator-worker pattern for decomposing complex goals into subtasks
- Evaluator-optimizer pattern for iterative generation and refinement
- Routing pattern for directing inputs to specialized prompts or tools
- Autonomous agent pattern with environment feedback loops and tool execution
Pricing
Free
Details
- Type
- Framework
- Category
- AI
- Topic
- Design Systems
- Open source
- No
- Platforms
- other
- For
- AI engineers and software developers building LLM-powered applications
- Tags
- agents, prompt engineering, llm architecture, langgraph, workflows, frameworks
Why it was saved
- @LangChain: A mental model for deciding whether a task needs a simple workflow, a controlled pipeline, or a true agent.
Related
Curated collection of plain-text design systems extracted from popular tech websites.
Quick typography, color palette, and iconography reference for clean product interfaces.
Open-source, token-driven design system and Figma UI kit by GR8 Tech.
AI-readable DESIGN.md style references extracted from popular websites.
Curated UI and UX checklists for websites, apps, and user flows
A concise spec sheet of typography, colors, and radii for minimal UI
Updated