Agentic Workflows: Learn AI Agent Patterns
An interactive visual guide to the 5 core AI agent workflow patterns defined by Anthropic. Free, no signup required.
1. Routing
Routing classifies an input and directs it to a specialized followup task. Works well for complex tasks where there are distinct categories that are better handled separately.
2. Parallelization
Divides work into independent subtasks (sectioning) or runs identical tasks multiple times (voting) for parallel execution. Effective when subtasks can be parallelized for speed.
3. Prompt Chaining
Decomposes a task into a sequence of steps, where each LLM call processes the output of the previous one. Ideal for tasks that can be cleanly decomposed into fixed subtasks.
4. Evaluator-Optimizer
One LLM call generates a response while another provides evaluation and feedback in a loop. Particularly effective when there are clear evaluation criteria.
5. Orchestrator-Workers
A central LLM dynamically breaks down tasks, delegates them to worker LLMs, and synthesizes their results. Well-suited for complex tasks where you cannot predict the subtasks needed.
Additional Patterns
6. Verification Pattern
A secondary agent checks the primary agent output against sources, flagging errors before delivery.
7. Combined Workflows
Multiple patterns chained together: route first, then parallelize, then pipeline the results into a final output.
8. Task Delegation
AI passes tasks to the next available worker based on capacity and specialization, like tickets in a queue.
Who This Is For
- Office workers automating email, reports, and meeting prep
- DevOps and AI Solutions Architects building CI/CD, incident response, and monitoring workflows
- Anyone automating personal tasks: trip planning, job hunting, learning
Source
Based on Building Effective Agents by Anthropic (December 2024).
Built by Sameer Goel