100 Agents in 100 Days

This project is what happens when over-planning is asked politely to leave the room. Instead of designing the perfect AI agent architecture, the goal is to build 100 small agents in 100 days. Most are deliberately simple and focused on one agentic technology.

Here they are: the useful agents, the odd agents, and the agents who made a strong case for better supervision.

Custom agents

Agents built from the ground up, without any third-party agent framework.

• 

  Customer complaint agent

  Initial version of a bespoke agent

• 

  Website status checker

  Introduction of a bespoke Otto agent framework

• 

  Birthday card helper

  Improving the bespoke Otto agent framework

• 

  Obsidian vault summarizer

  Adding guardrails to the bespoke Otto agent framework

OpenAI Agents SDK

Agents built using OpenAI’s Agents SDK.

• 

  History tutor agent

  First steps with the OpenAI Agents SDK

• 

  Webpage image repairer

  Using tools with the OpenAI Agents SDK

• 

  Bedtime story teller

  Using guardrails with the OpenAI Agents SDK

• 

  Rubber duck debugger

  Human approvals with the OpenAI Agents SDK

• 

  Office fridge labeler

  Using an MCP server with the OpenAI Agents SDK

Google’s Agent Development Kit

Agents built using Google’s Agent Development Kit.

• 

  New York guide

  First steps with Google’s Agent Development Kit

• 

  Greeting card writer

  Creating an agent team with Google’s ADK

• 

  Internal acronym explainer

  Using guardrails with Google’s ADK

• 

  Color style agent

  Using plugins with Google’s ADK

IBM’s BeeAI framework

Agents built using IBM’s open-source BeeAI framework.

• 

  Travel landmark agent

  First steps with the BeeAI framework

• 

  Document summarizer

  Using guardrails with the BeeAI framework

• 

  Movie finder agent

  Retrieval-augmented generation with BeeAI

• 

  Gherkin test checker

  Non-agent workflow with the BeeAI framework

• 

  Gherkin test writer

  Multi-agent workflow with the BeeAI framework

LangChain ecosystem

Agents built using LangChain, LangGraph, or related tools from the LangChain ecosystem.

• 

  Marcie chatbot agent

  First steps with LangChain

• 

  Leave request checker

  Non-agent workflow with LangChain

• 

  Customer note taker

  Using guardrails with LangChain

• 

  Accessible HTML coder

  First steps with LangGraph

• 

  Meeting room finder

  Using tools with a LangGraph agent

• 

  Software project summarizer

  First steps with DeepAgents

Anthropic agents

Agents built using Anthropic’s tools and frameworks.

• 

  Refactoring coder

  First steps with Anthropic’s Python SDK

• 

  Model recommender

  First steps with the Claude Agent SDK

• 

  React card coder

  Using screenshots with the Claude Agent SDK

• 

  White belt hacker

  Using tool governance with the Claude Agent SDK

• 

  Web resource checker

  Using skills with the Claude Agent SDK

Microsoft agents

Agents built using Microsoft’s tools and frameworks.

• 

  Risk classifier

  First steps with Microsoft’s Agent Framework

• 

  Scribble sorter

  First steps with workflows in the Agent Framework

• 

  Postage calculator

  Using skills with the Agent Framework

• 

  Slimy car salesman

  Using guardrails with the Agent Framework

CrewAI agents

Multi-agent systems built using CrewAI.

• 

  Topic instructor

  First steps with CrewAI

• 

  Price checker

  Using tools with CrewAI

• 

  Movie suggester

  Using guardrails with CrewAI

• 

  TxtSpkr

  Using skills with CrewAI

Coding agents

Agents whose purpose is to assist with coding tasks.

• 

  Broken link checker

  Evaluating Cline CLI

• 

  Security headers auditor

  Evaluating Codex CLI

• 

  Webpage content checker

  Evaluating Google’s Jules

Other framework agents

Agents based on other frameworks.

• 

  Ghost story raconteur

  First steps with the LlamaIndex framework

• 

  Vehicle trip analyst

  First steps with Hugging Face’s smolagents

• 

  Bank support officer

  First steps with the Pydantic AI framework

• 

  Contact assistant

  First steps with the Haystack AI framework

• 

  Note router

  First steps with OpenClaw

Sandboxed agents

Agents deployed to sandboxed locations.

• 

  Stock local machine sandbox

  Agent running in a stock local machine

• 

  Stock virtual machine sandbox

  Agent running in a stock virtual machine

• 

  Stock micro virtual machine sandbox

  Agent running in a stock micro virtual machine

• 

  Stock container sandbox

  Agent running in a stock container

• 

  Read-only container sandbox

  Agent running in a read-only container

• 

  Container sandbox with Squid sidecar

  Agent running with a Squid sidecar container

• 

  Container sandbox with fewer services

  Agent running in a lightweight container

• 

  Container sandbox with fewer commands

  Agent running in a tightened container

• 

  Container sandbox with a Syscomp profile

  Agent running in a restricted container

• 

  Container sandbox with fewer resources

  Agent running in a restrained container

• 

  Container sandbox with a minimized image

  Agent running in a minimized container

• 

  Container sandbox with runtime control

  Agent running with runtime control

• 

  Container sandbox with network socket control

  Agent running with network socket control

• 

  Container sandbox with no desktop automation

  Agent running with no desktop automation channel

• 

  Container sandbox with hardware control

  Agent running with hardware control

• 

  Container sandbox with no shell access

  Agent running with no shell access

Docker sandboxed agents

Agents deployed to a Docker sandboxed location.

• 

  Sandbox agent

  Agent running with minimal, declarative sandbox

• 

  MCP sidecar

  Agent using a simple MCP Server sidecar

• 

  Jina Reader sidecar

  Agent using a Jina Reader sidecar

• 

  Code execution sidecar

  Agent using a code execution sidecar

• 

  Ollama model sidecar

  Agent using an Ollama model sidecar

• 

  HAProxy sidecar

  Agent using an HAProxy sidecar

Docker sandboxed agent teams

Multiple agents deployed to a Docker sandboxed location.

• 

  Two-agent sandbox

  Using simple agent-to-agent (ACA) messages

• 

  Movie poster team

  Using sequential agent-to-agent (ACA) tasks

• 

  Bug report team

  Using parallel agent-to-agent (ACA) tasks

• 

  Red green team

  Using iterative agent-to-agent (ACA) tasks

Local model agents

Agents using local models.

• 

  Local model evaluation

  Evaluating local models for language and reasoning

• 

  Local model coding

  Evaluating local models for coding tasks

• 

  Local Stable Diffusion

  Evaluating the SD model for image generation

• 

  Local Stable Diffusion XL

  Evaluating the SDXL model for image generation

• 

  Local image understanding

  Evaluating local models for image understanding

• 

  Local OCR

  Evaluating local models for optical character recognition

• 

  Local text moderation

  Evaluating local models for text moderation

• 

  Local image moderation

  Evaluating local models for image moderation

Agents with memory

Agents using memory, following the CoALA paper.

• 

  Takeaway agent Alison

  An agent which uses only short-term working memory

• 

  Takeaway agent Beverly

  An agent which uses long-term episodic memory

• 

  Takeaway agent Catherine

  An agent which uses long-term semantic memory

• 

  Takeaway agent Diane

  An agent which uses long-term procedural memory

Microsoft Foundry agents

Agents deployed to Microsoft Foundry.

• 

  Chinese idiom explainer

  Foundry prompt agent using web search

• 

  Haunted hotel concierge

  Foundry prompt agent using text resources

• 

  Bakery analyst

  Foundry prompt agent using data resources

• 

  Caesar cipher

  Foundry prompt agent using MCP server tools

• 

  Mysterious Island

  Foundry code-first agent using built-in RAG tools

• 

  Customs team

  Multiple Foundry agents working as an A2A team

AWS Bedrock agents

Agents deployed to AWS Bedrock.

• 

  Nutshell summarizer

  First steps with AWS Bedrock and a Strands agent

• 

  Maths helper

  AWS Bedrock code interpreter tool with a Strands agent

Observed Agents

Adding tracing and observability to agent.

• 

  Adventurer Oliver

  Observability with OpenAI’s Agents SDK

• 

  Adventurer Adrian

  Observability with Anthropic’s Claude Agent SDK

• 

  Adventurer Makenzie

  Observability with Microsoft’s Agent Framework

• 

  Adventurer Ivan

  Observability with IBM’s BeeAI framework

• 

  Adventurer Langdon

  Observability with the LangGraph framework

• 

  Adventurer LLoyd

  Observability with the LLamaIndex framework

Designer Agents

Agents capable of designing a website.

• 

  Designer Addison

  Design using an OpenAI agent and model

• 

  Designer Blake

  Design using an Anthropic agent and model

• 

  Designer Chase

  Design using a Google agent and model