mediumAI Engineering

What is the ReAct pattern and why did it become the default for agents?

Q: What is the ReAct pattern and why did it become the default for agents?

**TL;DR.** **ReAct = Reason + Act**: the model alternates explicit **thoughts** ("user wants X, I should check Y") with **actions** (tool calls), feeding each **observation** back before the next thought. Verbalized reasoning improves action selection and makes trajectories **debuggable** — which is why the structure stuck. thought plan next step action to

66 views

Understand the problem

Interleaving reasoning and acting: thought, action, observation on repeat.

react-patternagentsreasoning

Attempt it yourself

Sketch your approach before reading the solution — that's what interviews test.

Stuck? AI Nudge Available

Get a conceptual hint to guide your logic without spoiling the final implementation.

Study the solution

The solution is waiting

Give it an honest attempt first — then compare your thinking with the full walkthrough.

Read the code

A ReAct trajectory (what you read when debugging)

Thought: The user asks for Q2 refunds for customer 811. I need order data first.
Action:  search_orders({ customer_id: "811", from_date: "2026-04-01",
                         to_date: "2026-06-30", status: "returned" })
Observation: 3 returned orders: #90312 ($120), #90417 ($60), #90583 ($75)

Thought: Refund totals come from the payments tool, not order totals.
Action:  get_refunds({ order_ids: ["90312", "90417", "90583"] })
Observation: refunds: 90312=$120, 90417=$60, 90583=$0 (pending)

Thought: Two completed refunds ($180), one pending. I can answer precisely.
Final:   "Q2 2026: $180 refunded across 2 orders; 1 refund ($75) still pending."

Join the discussion

Discussion (0)

No responses yet. Be the first to share what you think.

How does function calling (tool use) work under the hood?

What should you log and trace in an LLM application?