AI / MLMarch 1, 20264 min read

The Evolution of AI Application Stack: 2020 to 2024

How RAG, LangChain, LangGraph, and GraphRAG built on each other to shape the modern LLM application stack.

The timeline

Modern AI applications did not appear overnight. They evolved through four major milestones, each solving a problem the previous layer could not handle alone.

ai_stack_timeline.txtTimeline

2020 ── RAG
         ↓
2022 ── LangChain
         ↓
2024 ── LangGraph
         ↓
2024 ── GraphRAG

What each layer does

ai_stack_definitions.txtDefinitions

RAG
 └─ Retrieves relevant documents for LLMs

LangChain
 └─ Framework to build LLM applications

LangGraph
 └─ Framework to orchestrate complex agents built with LangChain

GraphRAG
 └─ Advanced RAG using knowledge graphs

RAG solved knowledge grounding. LangChain made LLM apps composable. LangGraph added agent orchestration with cycles and state. GraphRAG extended retrieval with graph structure for global reasoning across large corpora.

In production systems like AI booking platforms, all four often coexist: RAG for document retrieval, LangChain for chains and tools, LangGraph for multi-step planning agents, and GraphRAG when cross-document relationships matter.