Evolution of AI Agents
https://github.com/coleam00/evolution-of-ai-agents- Category
- Education
- Rank
- No. 2009Tools index
- Pricing
- Open Source
- Type
- TOOL
- Builder
- coleam00
- GitHub
- 37 stars
- Date
About
Three working examples showing how AI agent development evolved — from traditional RAG to SDKs and skill-based frameworks.
What it does
A comparative lab for understanding three agent architectures through runnable command-line applications. One answers from indexed documents, another coordinates research and writing agents, and the third discovers packaged instructions only when a task needs them.
Why it's ranked here
The comparison is unusually concrete because each architectural claim maps to executable code, storage choices, and visible tool behavior. It teaches more than a conceptual timeline. Still, these are demonstrations rather than a unified toolkit, and setup varies substantially between examples.
What's good
The examples expose the machinery that diagrams often hide. The retrieval agent covers ingestion, chunking, embeddings, vector search, streaming, and citations. The SDK agent shows delegation, monitoring hooks, built-in tools, structured note storage, and an in-process MCP server. The skills agent demonstrates staged instruction loading, typed dependencies, streaming, and multiple model providers.
Tradeoffs
Running the full comparison requires two languages, separate dependency tools, model credentials, and PostgreSQL services. The retrieval example depends specifically on OpenAI embeddings and pgvector. The SDK example still requires a database despite avoiding vector infrastructure, disables permission checks, and suppresses TypeScript checking. The repository presents separate demos, not one migration path or shared application architecture.
How to use it well
Use it for architecture workshops, prototypes, or deciding whether document retrieval, an agent SDK, or on-demand skills fit a problem. Run the examples side by side and compare infrastructure, context use, and tool control. Do not treat it as a production platform, deployment guide, or complete evaluation suite.
Technical notes+
README.md supplies separate uv and Bun workflows. rag-agent-demo/ingestion/ingest.py reads Markdown, text, and frontmatter, creates chunks, generates embeddings, and inserts documents plus vectors into PostgreSQL. rag-agent-demo/cli.py streams Pydantic AI responses and preserves message history. claude-agent-sdk-demo/agent.ts defines save_note and search_notes, registers them with createSdkMcpServer, configures researcher and writer subagents, logs tool use, sets permissionMode: "bypassPermissions", caps spend, and begins with // @ts-nocheck. pydantic-ai-skills-demo/src/agent.py injects discovered skill metadata through a dynamic system prompt and registers skill, HTTP, and note tools. pydantic-ai-skills-demo/src/cli.py exposes streaming text and tool events.
Observed
- Primary languages
- Python and TypeScript
- Install surface
- Python examples use uv; the TypeScript example uses Bun
- Interfaces
- Three command-line applications, plus an in-process MCP server in the TypeScript demo
- Retrieval storage
- PostgreSQL with pgvector stores documents, chunks, and embeddings
- Model provider support
- The retrieval example uses OpenAI; the skills example documents OpenRouter, OpenAI, and Ollama support
- Skill structure
- The skills example uses three disclosure levels: metadata, instructions, and resources
Read from README.md, rag-agent-demo/cli.py, rag-agent-demo/query.py, rag-agent-demo/rag_agent.py, claude-agent-sdk-demo/agent.ts, rag-agent-demo/utils/models.py, rag-agent-demo/utils/db_utils.py, rag-agent-demo/utils/providers.py, pydantic-ai-skills-demo/src/cli.py, rag-agent-demo/ingestion/ingest.py, rag-agent-demo/ingestion/chunker.py, rag-agent-demo/ingestion/__init__.py, pydantic-ai-skills-demo/src/agent.py, rag-agent-demo/ingestion/embedder.py.
What it can do
Demonstrate traditional RAG implementation
Document corpus and user queries → Retrieved and generated responses using basic RAG approach
Show SDK-based agent development
Development requirements and SDK configuration → AI agent built using software development kit framework
Illustrate skill-based agent framework
Skill definitions and task requirements → Modular AI agent with composable skills
Compare agent development approaches
Multiple AI agent implementations → Side-by-side comparison of different development methodologies
Execute example workflows
Sample tasks and test scenarios → Working demonstrations of each agent type in action
Tags
Tech Stack
Comments (0)
No comments yet
Editorially curated, with community endorsements as a secondary signal. Corrections welcome.