Designing AI Agents

This repository is a collection of concepts, notes, research papers, articles, design patterns, tools, and examples useful for building AI agents.

🚜👷🚧🏗️ Note: This repo is under active development. Things will move around a lot.

About agents

AI Agents are software programs capable of:

• Interacting with their environment
• Collecting data
• Planning
• Making decisions based on data
• Performing predetermined goals
• Autonomously

Architectural patterns

• GenAI architectural patterns
• Navigating Complexity: Orchestrated Problem Solving with Multi-Agent LLMs
• 3 Patterns in Agent Design
• Design Patterns with Autogen

Architectural types

Simple Reflex Agents

• Simple reflex agents respond directly to stimuli from the environment without considering the history of the world.

Model-based Reflex Agents

• Model-based reflex agents use a model of the world to handle partially observable environments.

Goal-based Agents

• Goal-based agents act to achieve specific goals.

Utility-based Agents

• Utility-based agents aim to maximize their performance measure through a utility function.

Learning Agents

• Learning agents have the ability to improve their performance over time based on experience.

ReAct Agent (Reason + Action)

• Combines reasoning and action to improve decision-making processes in AI agents.

Task-Planner Agent

• Uses task planning to break down complex goals into manageable tasks for the agent to execute.

Multi-Agent Orchestration

• Coordinates multiple agents to work together towards common goals, enhancing their collective performance.

Examples Agents

• Awesome AI Agents

General Resources

• Attention is All You Need
• GPT-3: Language Models are Few-Shot Learners
• BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding
• State of GPT by Microsoft
• Busy Executive’s Guide on LLMs

Components of Agents

• Agent Personas and Goals: Developing and defining different agent personas and their objectives to tailor their behavior.
• Decision Making and Planning
  • Prompting Best Practices
• Automated Planning and Scheduling
  • Automated Planning and Scheduling
• Learning and Adaptation
  • Concepts and techniques for enabling AI agents to learn and adapt to new environments and challenges.
  • Cooperative Multi-Agent Reinforcement Learning
  • One-Shot Learning with Memory-Augmented Neural Networks
  • Zero-Shot Learning: A Comprehensive Evaluation of the Good, the Bad and the Ugly
• Multi-agent Systems and Coordination
• Interactions
  • Best practices and techniques for designing interactions between AI agents and humans.
  • Techniques for processing and managing inputs and outputs for AI agents.
  • Reflection (Reflexion)
• Hallucinations
  • Grounding
    • Grounding with Google AI
• Deployments, scalability and Performance
  • LangChain with Reasoning Engine
• Security and Privacy

Other Resources

• Startup School: Gen AI - Building AI Agents