Chain-of-Thought Reasoning

What is Chain-of-Thought (CoT)?

Chain-of-thought prompting encourages the model to show its reasoning step by step before arriving at a final answer. This dramatically improves performance on complex reasoning tasks.

Basic CoT Prompt

Q: A store has 47 apples. They sell 23 and receive a shipment of 31. 
   How many apples do they have?

Let's think step by step:
1. Start with 47 apples
2. Sell 23: 47 - 23 = 24 apples
3. Receive 31: 24 + 31 = 55 apples

Answer: 55 apples

Techniques

• "Let's think step by step": Simple addition that triggers CoT behavior
• Structured reasoning: Ask the model to break down the problem into numbered steps
• Self-consistency: Generate multiple CoT paths and take the majority answer
• Tree of Thought: Explore multiple reasoning branches and evaluate each

When CoT Helps Most

• Mathematical word problems
• Multi-step logical reasoning
• Code debugging and analysis
• Complex decision-making with multiple factors
• Any task where intermediate steps matter

Extended Thinking

Modern models like Claude support "extended thinking" — a dedicated reasoning phase before generating the response. This is particularly powerful for complex problems where the model needs to explore and evaluate multiple approaches.

🌼 Daisy+ in Action: Chain-of-Thought in the ERP

When Daisy+ digital employees process complex requests — like "find all overdue invoices for clients in California and draft follow-up emails" — they use chain-of-thought reasoning to break the problem into steps: query the ERP via MCP tools, filter results by state and due date, compose personalized messages for each client, and route them through the email system. Each step is logged in the ERP's chatter for full auditability.