How Claude Thinks

Most participants arrive with one of two mental models. The first: Claude is Google with better sentences. The second: Claude is an alien intelligence that understands everything. Both models produce bad prompts. The search-engine model produces keyword queries. The magic model produces vague wishes.

This module installs a third model: Claude is a reasoning engine that processes language, follows instructions, and generates responses based on patterns. It does not retrieve information from a database. It does not remember previous conversations. It works with what you give it in the moment. Once people understand this, every prompting skill that follows has a foundation to sit on.

Every task you hand Claude falls into one of five categories. Generation: creating something from scratch (drafts, code, plans). Analysis: examining something that exists (data, documents, arguments). Transformation: changing the shape or format of content (summaries, translations, restructuring). Extraction: pulling specific information from a larger body (numbers from contracts, names from transcripts, dates from reports). Reasoning: working through logic, calculations, or multi-step problems where the answer requires thought, not just retrieval.

These categories matter because each one has different prompting requirements. A generation task needs voice, audience, and constraints. An extraction task needs precision about what to pull and what format to return it in. Reasoning tasks benefit from chain-of-thought prompting, which participants will learn in Module 06. For now, the categories are a sorting tool. They help people see that “use Claude” is not one skill. It is five.

This demo is the single most effective moment in the entire Tier I curriculum. It takes one topic (employee turnover) and shows three levels of prompting quality. Version A is a search query: “employee turnover.” Claude responds with a generic encyclopedia entry. Version B is a vague request: “Help me understand our turnover problem.” Claude gives general advice that could apply to any company. Version C is a structured prompt with role, company details, specific data, and format instructions. Claude produces work that could walk into a boardroom.

The teaching power comes from the contrast. Do not explain the lesson before the demo. Let the outputs speak. The room will see it.

Several tasks on the worksheet are deliberately ambiguous. “Summarize a 40-page report” could be Transformation (reshaping long content into short content) or Extraction (pulling key points). “Compare two vendor proposals” could be Analysis (evaluating each one) or Reasoning (deciding which one is better). The ambiguity is the point.

When pairs debate which category a task belongs in, they are developing the habit of thinking about what they are actually asking Claude to do. That thinking is more valuable than getting the “right” answer on the worksheet.

The answer key in the demo file provides the primary category for each task and notes where secondary categories apply. Use it to guide the debrief, not to declare winners.

The break-even calculation uses the scenario from the demo file: Greenfield Manufacturing is considering a new product line of custom aluminum brackets for aerospace. The numbers: $472,000 in annual fixed costs (equipment, facility, operators, certification, insurance, marketing), variable costs of $26.69 per unit (raw material $14.20, consumables $3.85, inspection $2.10, shipping $4.50, 3% sales commission on $68 = $2.04), and a selling price of $68.00 per unit.

The contribution margin is $68.00 minus $26.69 equals $41.31 per unit. Break-even is $472,000 divided by $41.31, which equals approximately 11,426 units.

Why this demo works: the math is straightforward enough that people in the room can follow it, but complex enough that doing it by hand would take ten minutes and invite arithmetic errors. Claude handles the computation correctly and shows every step. That transparency is the teaching moment. Claude is not pulling a number from a database. It is working through the logic, which is exactly what the Reasoning category means.

When the room sees the break-even answer, ask the follow-up from the demo file: “What if I told you the sales commission was 5% instead of 3%? Would you re-run the whole analysis yourself, or would you just tell Claude what changed?” This demonstrates that conversation (Module 03’s topic) is the natural complement to a good first prompt.

Common questions: “Can Claude handle more complex financial models?” Yes, but accuracy decreases with complexity. For models with many interdependent variables, Claude should be treated as a fast first draft, not a calculator of record. This program does not teach participants to trust Claude’s math blindly. It teaches them to use Claude’s math as a starting point and verify the numbers that matter.