Agent Architectures

What is an AI Agent?

An AI agent is an LLM-powered system that can autonomously plan, use tools, and take actions to accomplish goals. Unlike simple chatbots, agents can break down complex tasks, interact with external systems, and iterate on their approach.

Core Agent Loop

while not task_complete:
    1. Observe: Gather information (user input, tool results, environment state)
    2. Think: Reason about what to do next (LLM inference)
    3. Act: Execute a tool or provide a response
    4. Evaluate: Check if the goal is achieved

Common Architectures

ReAct (Reasoning + Acting)

Interleaves reasoning traces with actions. The model explicitly states its reasoning before each action.

Thought: I need to find the current stock price of AAPL.
Action: search_stock_price("AAPL")
Observation: AAPL is trading at $185.32
Thought: Now I can answer the user's question.
Answer: Apple (AAPL) is currently trading at $185.32.

Plan-and-Execute

First creates a full plan, then executes each step. Good for complex multi-step tasks.

Reflexion

Agent reflects on its failures and uses that reflection to improve subsequent attempts.

Key Components

• Memory: Short-term (conversation) and long-term (persistent storage)
• Tools: APIs, databases, code execution, web browsing
• Planning: Task decomposition and sequencing
• Error handling: Recovery from tool failures and incorrect reasoning

🌼 Daisy+ in Action: Digital Employee Architecture

Daisy+ digital employees follow a ReAct-style agent architecture: they Reason about the user's request, take Actions via MCP tool calls (search records, create tasks, send messages), and Observe the results before deciding on next steps. Each agent has its own Odoo user account with appropriate access rights — just like a human employee, but available 24/7.