Vibe Coder · Unit 2: How Claude and ChatGPT Work · Lesson 5 of 7

Thinking and tools

A quick reply, thinking first, or an agent that plans, acts and checks. What each one is and when to use it.

Read the short notes and pictures.Do everything inside a blue frame.

Three ways a model can work

So far you have seen the plain reply: the model reads your message and writes an answer in one pass. Claude and ChatGPT can do two more things, and most of this course depends on them. They can think before answering, and they can use tools.

Thinking first

With extended thinking on, the model works through the problem before it writes the answer: it lays out the steps, tries an approach, and checks it. You can open that thinking and read it. The answer takes longer and is better on hard problems, such as planning a product, finding a bug or comparing options with trade-offs.

Quick reply

Seconds

Answers at once. Right for simple questions, rewrites and small edits.

Thinking first

Longer

Works the problem through, then answers. Right for plans, hard bugs and decisions.

Tools let it act

On its own, a model can only write text. A tool is something the app lets the model use to do more than write. The model decides when a tool would help, asks the app to run it, and reads the result.

A 3D scene of coloured blocks and cards. Tools let it act: search, run code, read files, use apps.
Four kinds of tool: search, running code, reading files, and the apps you connect.

Search

Looks things up on the web and reads the pages. This is how it answers past its cutoff.

Run code

Writes a small program and runs it, to do sums, analyse a spreadsheet or draw a chart, instead of guessing the numbers.

Read and write files

Opens the documents you attach, and creates new ones: a document, a page, an app.

Use your apps

Works in the apps you connect, such as your drive, your calendar or GitHub.

your codethe modelyour function1 question + tool definitions2 tool call: get_weather("Pune")3 run it {"temp_c": 29}4 the result5 "It is 29°C in Pune."
The model asks for a tool, the app runs it, and the result goes back into the context.

An agent puts them together

An agent is a model that keeps going on its own: it makes a plan, uses a tool, looks at the result, and decides what to do next, until the job is done. Claude Code, which you will use to build your product, is an agent. You give it a goal such as "add a sign-up page", and it reads your code, writes new files, runs the app, and fixes what breaks.

A 3D scene of coloured blocks and cards. An agent: plan, use a tool, look at the result, go again, done.
The agent loop: plan, act, look at the result, go again.
thinkthe model chooses an actionactyour code runs the toolobservethe result joins the historyhistoryexit: a final answer, or the step limit
An agent repeats the same cycle until the goal is met or it needs you.

Which to use

Quick reply

Simple questions

"Rewrite this sentence." "What does this error mean?"

Thinking

Hard problems

"Compare these three pricing models for my product and recommend one."

Agent

Many steps

"Build the sign-up page, connect it to the database, and test it."

ActivityBuild · step 5 of 6A fact-checked product brief

Plan the first version with thinking on

Paste your handover note and ask, with extended thinking on, for a plan of the smallest first version: what it does, what it leaves out, and the three riskiest assumptions. Read the thinking, then add the plan to the brief under "First version".

Done when

This plan is what you will prototype in Unit 7.

Keep these

  • A plain reply is written in one pass.
  • Extended thinking works through the problem first: slower, and better on hard problems.
  • Tools let the model search, run code, read and write files, and use your apps.
  • An agent plans, acts and checks in a loop until the job is done. Claude Code is one.