← AI Agents & MCP theorytheory 0/14 · 0%

level 1

Curious · 0 XP

streak

perfect

badges

100 XP to Apprentice

Patterns · easy

2. ReAct: reason, act, observe

The classic prompting pattern that made tool-using agents reliable.

1 min read2 questionsup to 50 XPtopic 2/14

Read in another language

AI translates the explanation and quiz text while preserving code and answer scoring.

ReAct interleaves *thought*, *action*, and *observation*. Modern models do this natively through tool calling, but the pattern still explains why agents work: making the reasoning explicit before an action reduces impulsive, wrong tool calls.

- Thought: what do I know, what is missing? - Action: one tool call with concrete arguments. - Observation: the tool's real output, never invented.

text
Thought: I need the current ETH gas price before estimating the fee.
Action: http_get({"url": "https://api.example.com/gas"})
Observation: {"baseFeeGwei": 12.4}
Thought: Fee = 21000 * 12.4 gwei = 0.00026 ETH. I can answer now.
Answer: About 0.00026 ETH at the current base fee.

With native tool calling, keep a short `"Think about what is missing before each call"` line in the system prompt: it is the cheapest accuracy improvement available.

Gotcha: never let the model write the observation itself. If your loop appends model-generated text as a tool result, the agent will happily hallucinate API responses.

Exercise: run the same question twice — once with a ReAct hint in the system prompt, once without — and count tool calls and wrong answers over ten runs.

How to study this topic

1. Read the explanation once end to end, skipping the code.
2. Re-read the code snippets and say out loud what each line does.
3. Answer the quiz without scrolling back — that is the real check.
4. Read every explanation, even for the answers you got right.

Scoring

· Up to 30 XP from quiz accuracy.
· +15 XP for a perfect first attempt.
· +5 XP while a 3-day streak is running.
· 75% or higher marks the topic complete.

Check your understanding

1. What does the 'observation' step in ReAct contain?
The real output of the tool that was calledThe model's prediction of the outputThe user's next messageA summary of the plan
2. Why does explicit reasoning before a tool call help?
It surfaces missing information and reduces wrong argumentsIt reduces token costIt caches resultsIt bypasses rate limits

Badges

First light

Pass your first topic

Sharp shooter

3 perfect first-try quizzes

On a roll

Study 3 days in a row

Week warrior

Study 7 days in a row

Grinder

Earn 500 XP

Deep diver

Earn 1500 XP