Agent Workflow Patterns
Five workflow patterns + the autonomous agent — the building blocks of effective LLM systems, with a citation behind every claim. Start with the concept, then dive into each pattern.
Core Concept: Workflow vs Agent
Before the six patterns: what an 'agentic system' is, the building block they all share, when to use which — and why the 'how many patterns' count differs between sources.
1 · Prompt Chaining
Workflow · Decompose a task into a fixed sequence of steps; each LLM call processes the previous output, with optional checks between.
2 · Routing
Workflow · Classify the input first, then send it to a specialized follow-up — a different prompt, tool chain, or model.
3 · Parallelization
Workflow · Run multiple LLM calls at once and aggregate the results. Two variants: sectioning and voting.
4 · Orchestrator–Workers
Workflow · A central LLM dynamically breaks down the task, delegates to worker LLMs, and synthesizes their results.
5 · Evaluator–Optimizer
Workflow · One LLM generates a response while another evaluates and gives feedback, in a loop — iterative refinement.
6 · Autonomous Agents
Agent · No fixed path. The LLM plans and acts in a loop, using tools and environmental feedback to direct itself toward a goal.
Compare the Six Patterns
Side by side on control flow, cost/latency, best fit and a representative example. Filter by kind, sort any column.
Practice
Make it stick: look up the jargon, flip the cards, then test yourself. Everything lives in this session only.
Sources & References
Every claim on this site traces back to a primary or authoritative source. Filter by origin, open any link.