Section 2 · Reasoning Patterns

Different problems need different thinking strategies. This compares reasoning patterns in AI agents — ReAct, Plan-and-Execute, Reflection, Tree-of-Thoughts and Self-Consistency — and when to use each.

How it works

1. 1ReAct (think → act → observe): good for tool use and open-ended problems.
2. 2Plan-and-Execute (plan first, then execute): good for complex, multi-step workflows.
3. 3Reflection (critique → improve): improves quality and factual accuracy.
4. 4Tree-of-Thoughts: explores multiple paths for hard reasoning.
5. 5Self-Consistency: multiple attempts, take the majority answer.

Key points

• ReAct is simple and works with tools; Plan-and-Execute reduces mid-way changes.
• Reflection catches self-mistakes; Tree-of-Thoughts finds better solutions but costs more.
• Self-Consistency boosts accuracy on reasoning-heavy tasks.
• Golden rules: start simple (ReAct), add planning/reflection for complexity, always measure.

Why it matters

Choosing the right reasoning pattern balances accuracy, cost and speed. It's a core decision when designing capable, efficient agents.

Frequently asked questions

What is ReAct?

A pattern interleaving reasoning (thoughts) with acting (tool calls and observations), step by step.

When use Tree-of-Thoughts?

For hard problems where exploring multiple reasoning paths finds a better solution — at higher cost.