6.1 Extended Thinking

Course: Claude Code - Power User Section: Advanced Input & Interaction Video Length: 2-5 minutes Presenter: Daniel Treasure

Opening Hook

Extended Thinking is like giving Claude a whiteboard to work through hard problems before answering. It leads to better reasoning, stronger error-checking, and clearer tradeoffs on complex tasks.

Key Talking Points

1. What is Extended Thinking?

A reasoning mode that lets Claude think deeply about complex, multi-step problems
Claude's internal reasoning is shown in a collapsible block in the UI
Trades processing time for reasoning quality
Ideal for architecture decisions, debugging, refactoring, and complex reviews

What to say: "Extended Thinking is like giving Claude time to actually reason through a problem. Instead of rushing to an answer, it shows you its thinking process—what it's considering, what trade-offs exist, and why it landed on a solution."

What to show on screen: Open Claude Code interface, navigate to a prompt box, show where Extended Thinking toggle appears in settings

2. When to Enable Extended Thinking

Architecture design and system planning
Debugging tricky, hard-to-reproduce failures
Large-scale refactoring decisions
Edge-case testing strategies
Complex pull request reviews
Ambiguous requirements or unclear problem statements

What to say: "You don't need Extended Thinking for every task. But when you're facing something genuinely complex—a refactoring that touches ten files, or a bug that only happens under weird conditions—that's when you turn it on."

What to show on screen: Highlight the /config menu, show the Thinking mode toggle

3. How to Enable Extended Thinking

Via /config → toggle Thinking mode on
Via environment variable: MAX_THINKING_TOKENS=10000 (sets thinking budget)
Via settings.json: alwaysThinkingEnabled: true
Keywords in prompts: "ultrathink", "think harder", "think deeply" trigger it automatically

What to say: "You can toggle it on per-session, or configure it permanently in your settings. You can even set a thinking budget—how much processing power Claude should spend reasoning. Ten thousand tokens is a good default for complex problems."

What to show on screen: Demo toggling the setting on in /config, show the env var syntax in documentation

4. What Extended Thinking Gives You

Thorough, multi-step reasoning visible in a collapsible block
Better handling of ambiguous or underspecified problems
Stronger error-checking and edge-case consideration
Clear articulation of trade-offs and assumptions
More reliable output for high-stakes decisions

What to say: "When you read Claude's thinking, you're not just getting an answer—you're getting the reasoning behind it. You see what it considered and rejected, what assumptions it made, and why. That transparency is gold for complex work."

What to show on screen: Show a real example of Extended Thinking in action—open the thinking block, scroll through the reasoning, then show the final answer

Demo Plan

Setup (0:15) — Open Claude Code interface, show /config menu, locate Thinking mode toggle
Enable Extended Thinking (0:30) — Toggle it on, explain the MAX_THINKING_TOKENS setting
Submit a Complex Prompt (1:00) — Give Claude a real architecture question (e.g., "Design a caching layer for a multi-service app with these constraints...")
Show Reasoning Block (2:00) — Click to expand the thinking block, scroll through the reasoning, point out key decision points
Review Final Answer (2:30) — Show how the answer reflects the thinking, highlight trade-offs and assumptions
Disable and Compare (3:30) — Turn off Extended Thinking, ask the same question, show the faster but less detailed response
Wrap-up (4:00) — Summarize when to use it and best practices

Code Examples & Commands

Enable via environment variable:

export MAX_THINKING_TOKENS=10000
claude code

Disable Extended Thinking:

export MAX_THINKING_TOKENS=0

In settings.json (permanent enable):

{
  "alwaysThinkingEnabled": true,
  "maxThinkingTokens": 10000
}

Trigger Extended Thinking via prompt keywords:

"I need you to think deeply about this architecture problem..."
"Ultrathink: How would you refactor this codebase?"
"Think harder about the edge cases in this test."

Gotchas & Tips

Extended Thinking uses more tokens — The thinking process itself consumes tokens. For simple questions, it's overkill and wasteful. Reserve it for genuinely complex problems.
Thinking is shown but not always complete — The collapsible block shows the core reasoning, but may not be exhaustive. It's a summary of Claude's internal process.
Max thinking budget matters — Setting too low (e.g., 1000 tokens) won't give Claude room to reason. 10000 is a good starting point; go higher for really hard problems.
Keyword triggers are automatic — If you say "think deeply" in your prompt, Extended Thinking activates even if you haven't enabled it in settings. This is usually helpful but can be overridden if needed.
Thinking mode affects response time — Expect longer wait times. Build in buffer if you're demoing live or using this in time-sensitive contexts.
Works best with clear context — Give Extended Thinking all the context it needs: the problem, constraints, current state, and what success looks like. Vague prompts mean longer thinking for weaker results.

Lead-out

"Extended Thinking is your secret weapon for the hard stuff. Up next, we'll talk about handling longer inputs—how to write multi-line prompts that stay clean and readable, which is crucial when you're giving Claude complex context."

Reference URLs

Prep Reading

Claude Code config options and environment variables
What "extended thinking" means in the context of LLM reasoning
Common use cases: architecture, debugging, refactoring
Token budgeting and cost implications

Notes for Daniel: Demo with a real problem statement (e.g., database sharding strategy, authentication refactoring) so viewers see genuine complexity. Expand and collapse the thinking block a couple of times so the feature feels tactile. Emphasize that this is a tool choice—not every prompt needs it. If thinking takes > 30 seconds, keep talking about why Extended Thinking is valuable rather than waiting silently.

Quick Reference