ai-agent-design

AI Agent Design Skill

Patterns for designing AI agents—autonomous systems that use LLMs to reason, plan, and execute multi-step tasks.

Agent vs Chatbot vs Workflow

Aspect	Chatbot	Workflow	Agent
Autonomy	Low	None	High
Planning	None	Predefined	Dynamic
Tool Use	Limited	Fixed	Flexible
Memory	Session	None	Persistent
Error Recovery	Retry	Fail	Reason & adapt

Core Patterns

ReAct (Reasoning + Acting)

1. Thought: Reason about the task
2. Action: Choose and execute a tool
3. Observation: Process tool output
4. Repeat until complete

Example:

Thought: Need Seattle weather to answer umbrella question
Action: weather_api(location="Seattle")
Observation: {"temp": 52, "condition": "rain", "precipitation": 80%}
Thought: Raining with 80% precipitation. Recommend umbrella.

Plan-and-Execute

For complex multi-step tasks:

1. Planner: Create high-level plan
2. Executor: Execute each step
3. Replanner: Adjust based on results

Use when order matters and partial failures need recovery.

Reflexion

Self-improvement through reflection:

1. Attempt task
2. Evaluate outcome
3. Generate reflection on failures
4. Store reflection in memory
5. Retry with reflection context

Multi-Agent Patterns

Supervisor

Central coordinator delegates to specialists:

       Supervisor
      /    |    \
Research Writer Reviewer

Hierarchical Teams

Nested supervisors for complex organizations:

      Top Supervisor
       /         \
Research Lead  Writing Lead
   /    \         /    \
Web   Paper   Draft   Edit

Debate/Adversarial

Multiple agents argue to reduce hallucination:

Agent A (Pro) <--argue--> Agent B (Con)
              \    |    /
               Judge

Tool Design

{
  "name": "search_database",
  "description": "Search products. Use for availability/pricing queries.",
  "parameters": {
    "query": { "type": "string", "description": "Search terms" },
    "max_results": { "type": "integer", "default": 10 }
  }
}