Claude 101

Business Context

Most people who get access to Claude use it once or twice, find the results mediocre, and go back to doing things manually. The problem is almost never Claude. It is the lack of a working mental model and a few practical techniques.

This lesson gives you both. By the end, you will know what Claude is, how to talk to it well, where to run it, and how to make it useful for the work you do every day.

Core Concepts

What Claude Is and How It Works

Claude is an AI assistant built by Anthropic. It reads text you give it and generates a response. That is the entire mechanic.

What makes it powerful is not magic. It is the result of training on an enormous amount of human-written text, then refining that training so it follows instructions, reasons through problems, and behaves safely. The model has no internet access by default, no live data, and no memory between conversations unless you give it one.

Three things shape every response Claude gives you:

Claude does not guess what you meant. It works with what you gave it. Better input produces better output, consistently.

What Claude can do:

• Read, summarize, and reason about long documents
• Draft, rewrite, and improve text of any kind
• Write, review, and explain code
• Answer questions, compare options, and think through decisions
• Work through multi-step problems step by step
• Maintain a conversation across many turns

What Claude cannot do:

• Browse the internet or retrieve live data (by default)
• Remember previous conversations unless you use Projects
• Replace your judgment on high-stakes decisions
• Guarantee accuracy on specific facts without cited sources

Your First Conversation with Claude

Open claude.ai and start a new conversation. Do not overthink it.

The most productive first message is a real task from your actual work, not a test. Paste in a document you need summarized, a decision you are working through, or a draft you want improved.

A good first message has three parts:

1. What you want Claude to do
2. What Claude needs to know to do it
3. What the output should look like

Example:

I'm a product manager preparing for a sprint review tomorrow.
Here is the ticket backlog for the last sprint: [paste]

Summarize what was completed, what slipped, and the main reasons why.
Format: three bullet sections. Keep it under 200 words.
Audience: engineering leadership who was not in the sprint.

That is all it takes. You will get a usable response in seconds.

How to Write Effective Prompts from Day One

Every effective prompt has three elements. Check for all three before sending.

Clarity: State the task precisely.

Weak: "Help me with my presentation." Strong: "Write an executive summary slide for a board presentation on our Q2 engineering KPIs. Five bullet points, one line each, framed around business impact."

Context: Give Claude what it needs to do the job.

Weak: "Review this code." Strong: "Review this Python function for edge cases and silent failures. It runs in a payment processing service handling 5,000 transactions/hour. Flag anything that could fail under load."

Constraints: Define what the output should look like.

Weak: "Summarize this doc." Strong: "Summarize this requirements doc in bullet points grouped by theme. Max five bullets per theme. Audience is a developer who has not seen it."

Most bad results come from missing context, not from Claude being wrong. If an output misses the mark, look first at what you did not tell it.

Getting Better Results

Once you have a working prompt, these five techniques will take your results from good to consistently useful.

Iterate, do not restart. If the first response is close but not right, follow up in the same conversation. Claude remembers everything in the current thread.

That's mostly right, but the tone is too formal. 
Rewrite the second section to sound like how an engineering lead 
would explain this to their team in a Slack message.

Give examples. If you have a format or style you want matched, paste an example before the task.

Here is an example of a PR description in our team's style: [paste]
Now write a description for this change: [paste]

Assign a role. Framing Claude as a specific expert changes how it approaches the problem.

You are a senior security engineer reviewing this architecture 
for common vulnerabilities. Focus on authentication and data exposure.

Ask for reasoning. For complex decisions, ask Claude to show its thinking before giving an answer.

Think through the tradeoffs between these two database 
migration approaches before recommending one.

Use the conversation. Claude is not a one-shot tool. The more context you build in a thread, the more useful the responses become. A ten-turn conversation that produces a polished result is better than ten separate mediocre prompts.

Key Points

• Clarity, context, and constraints are the three levers for prompt quality
• Iteration in the same conversation is faster and more effective than starting over
• Examples and role framing are the two highest-leverage techniques for improving output quality
• Most bad results come from missing context, not from Claude being wrong

Claude in the Browser and Desktop App

Claude runs in two main surfaces for day-to-day use. Each suits a different context.

Browser (claude.ai): The default entry point. Supports file uploads, Projects, artifacts, research mode, and most capabilities. Everything in this lesson applies here.

Desktop app: Runs on Mac and Windows. Gives you persistent access outside the browser with three modes: Chat, Cowork, and Code. Faster to open, always accessible, and no tab to lose.

Engineers working directly in a codebase use Claude Code, a CLI tool with direct filesystem access. That is covered in Claude Code 101.

Claude Desktop App: Chat, Cowork, Code

The Claude desktop app gives you persistent access to Claude outside the browser. It runs on Mac and Windows and opens three distinct modes.

Chat is the standard conversation interface, equivalent to the browser. The advantage over the browser is that it runs as a native app: faster to open, always accessible, no tab to lose.

Cowork gives you a structured environment for longer-form work. Use it when a task is too large for a single exchange: drafting a multi-section document, running a structured workflow, or working through a project in stages.

Code is Claude Code inside the desktop app. Engineers get the same terminal-based coding agent without needing to install a separate CLI.

Organizing Your Work and Knowledge

Topic: [What you want researched]
Scope: [Boundaries: time range, geography, industry, use case]
Compare: [What options, approaches, or alternatives to evaluate]
Output: [Format you need: table, report, bullet summary, recommendation]
Decision: [What you will use this to decide]

Research the current options for running LLMs on-premise for a 
financial services company that cannot send data to external APIs.

Scope: production-ready solutions available as of 2025.
Compare: vLLM, Ollama, and any enterprise-grade alternatives.
Evaluate on: deployment complexity, model support, performance at scale, 
and vendor support availability.

Output: comparison table followed by a recommendation with reasoning.
Decision: we are choosing our internal inference stack for a 12-month build.

Role: [Who should Claude act as? e.g., "a senior backend engineer", 
"a product manager reviewing requirements"]

Context: [What does Claude need to know? Situation, constraints, 
relevant background. Paste documents here.]

Task: [What exactly do you need? Be specific.]

Output format: [Bullets, numbered list, prose, table, code block, 
slide draft, etc.]

Audience: [Who will read or use this output?]

Constraints: [Length limits, tone, what to avoid]

You are a senior software engineer reviewing this code for a 
production system that handles [volume/context]. Flag issues by 
severity. Explain the risk, not just the line.

You are a product manager helping me structure thinking on [topic].
My audience is [engineers / leadership / customers]. 
Prioritize clarity and actionability over completeness.

You are an experienced engineering manager helping me prepare for 
[situation]. My goal is [outcome]. This person values [communication style].
Give me questions to ask, not scripts to read.

You are a technical advisor with deep experience in [domain].
I need to make a decision on [topic] with these constraints: [list].
Walk through the tradeoffs before recommending.

Access & Setup
- [ ] I have a Claude account or API credentials configured
- [ ] I know which Claude interface my team uses
- [ ] I know if my organization has a Claude enterprise agreement

Data Handling
- [ ] I know what data I am permitted to paste into Claude
- [ ] I have not pasted customer PII, private credentials, 
      or unreleased IP without confirming this is allowed
- [ ] I understand whether enterprise data controls are active

First Use
- [ ] I have run at least one prompt on real work
- [ ] I have reviewed the output critically, not accepted it unchecked
- [ ] I have saved one effective prompt for reuse

If your organization has custom skills configured, they appear in the / menu alongside built-in commands.

Actionable Takeaways

1. Run the practical exercise now. Take one real open task and run it through Claude before this lesson is closed. One real use is worth ten read-throughs.

2. Set up a Project for your main work area. Add your role context, team background, and one or two key documents. Every conversation in that Project will start with full context.

3. Save the prompts that work. Start a prompt library: a Notion page, a Markdown file, or a shared doc. One saved prompt is worth ten one-off interactions.

4. Choose the right surface for each task. Browser for day-to-day work and documents. Desktop app for persistent access and longer workflows. Research mode for anything that would normally take hours of manual investigation. Engineers working in a codebase: see Claude Code 101.

5. Check what integrations your organization has configured. If tools like Jira, GitHub, or Notion are connected, use them. Pulling live data into a conversation is faster and more accurate than copy-pasting.

Practical Examples

Claude in Action: Use Cases by Role

Engineer

Daily use:

• Generate boilerplate, write unit tests, and draft PR descriptions
• Review code for edge cases, security issues, or performance problems
• Explain unfamiliar code or debug errors with full context

Example prompt:

Review this Go function for race conditions and unhandled errors.
It is called concurrently in a high-throughput event processing system.
Flag any issues with severity: critical, high, or informational.

[paste function]

Next step: Claude Code 101 covers how to run end-to-end tasks directly in your codebase: implementing features, writing tests, and multi-file refactors without copy-pasting.

Product Manager

Daily use:

• Draft and refine product requirements, user stories, and acceptance criteria
• Summarize research, interviews, or feedback into structured insights
• Prepare for stakeholder meetings with pre-drafted agendas and talking points

Example prompt:

Here are 12 user interview transcripts from our discovery research: [paste]

Extract the top 5 pain points mentioned most frequently.
For each pain point: one sentence description, frequency count, 
and one representative quote.
Format as a table.

With Projects: Maintain a product brief, your team's terminology, and your roadmap context so every conversation starts with full context.

Engineering Manager

Daily use:

• Prepare for 1:1s, performance reviews, and difficult conversations
• Draft team updates, retrospective summaries, and sprint reports
• Think through organizational problems with a structured thinking partner

Example prompt:

I'm an engineering manager preparing for a performance conversation
with a senior engineer. The engineer delivers strong technical work 
but avoids cross-team collaboration and often goes silent in 
design reviews involving other teams.

Help me:
1. Draft 2-3 questions to open the conversation
2. Frame the behavior pattern without being accusatory
3. Identify what to listen for in their response

Tone: direct and constructive. This person values candour.

Technical Director / CTO

Daily use:

• Evaluate architectural options with a structured tradeoff analysis
• Draft engineering strategy documents and technology decisions
• Research emerging tools, patterns, and industry approaches

Example prompt (Research Mode):

Research the current state of vector database options for a 
production RAG system at scale. We are processing ~10M documents 
with semantic search latency requirements under 100ms.

Compare the top three options on: performance, operational cost,
managed service availability, and maturity. Recommend one with reasoning.

Implementation Workflow

Practical Exercise: Your First Real Prompt

This exercise takes 15 minutes. Do it now, not later.

1. Pick a task you have open right now. A document to write, a decision to think through, code to review, a meeting to prepare for. It must be real work.

2. Open Claude at claude.ai or in the desktop app.

3. Write your prompt using the three-element structure:

   • Clarity: What exactly do you need?
   • Context: What does Claude need to know? Paste the relevant material.
   • Constraints: What should the output look like?

4. Run the prompt and read the response critically. Is it useful? Does it reflect your actual situation?

5. Iterate once. Send one follow-up that corrects or refines the output. Note what you had to add: that is what belongs in your original prompt next time.

6. If the output is good, save the prompt. Use the Prompt Starter Template above.

7. Set up a Project for the work area this task belongs to. Add the context you just used as the project instructions.