Build Your AI Automation

Define Your Specialist

1. Choose one task you do regularly (weekly, daily, or multiple times per week).
2. Examples: writing emails to stakeholders, creating lesson materials, analyzing client data, drafting proposals.
3. Write 2-3 sentences describing what the specialist AI needs to do.
4. Example: 'My AI helps me write clear, professional weekly status reports for my manager. It knows my company style, our projects, and our metrics. It formats as: highlights, metrics, blockers, next week.'

Create Your Claude Project

1. Go to claude.ai.
2. Click 'Create Project'.
3. Name it: '[Your Task] Assistant' (e.g., 'Weekly Report Assistant').
4. You're ready to add instructions.

Write System Instructions (Using 5-Element Framework)

1. In your Project, open Project Instructions.
2. Write all 5 elements: Role & Purpose, Tone, Knowledge Context, Output Format, Guardrails.
3. Be specific about YOUR situation, not generic.
4. See example below.
5. Test with one real task. Refine if needed.

Upload Documents (Optional But Helpful)

1. In your Project, upload 1-2 documents that help the AI understand your context.
2. Examples: past reports (so it knows your style), company style guide, team handbook, metrics definitions.
3. Tell the AI: 'Use these documents as reference when writing reports.'

Test with 3 Real Tasks

1. Run your project on 3 different weeks of data or scenarios.
2. Look at output: Does it match what you need? Is the tone right? Is the format useful?
3. Edit your Project Instructions based on what you learn.
4. After 3 iterations, you're done. It's solid.

Use It for Real

1. Next week, use your Project for real work.
2. Time it. How much faster is this than without AI?
3. Save your outputs so you can refine the system over time.
4. You now have a tool you'll use forever.

Map Your Task Into 4-6 Steps

1. Choose a task that takes you 1-2 hours currently.
2. Break it into 4-6 clear sequential steps. Each step's output feeds into the next.
3. Example: 'Writing a blog post' → 1) Research topic, 2) Create outline, 3) Write draft, 4) Edit for flow, 5) Optimize for SEO, 6) Format for publication.
4. Write each step and what the AI output should be.

Pro tip: If your task has more than 6 steps, group related ones together.

Write a Prompt for Each Step

1. Using the RACE framework, write a prompt for Step 1. Make it focused and clear.
2. Test it in Claude or ChatGPT. Save the output.
3. Write a prompt for Step 2. Paste Step 1 output at the beginning.
4. Test it. Save output.
5. Repeat for all steps. You'll have a working chain by the end.

Document Your Workflow

1. Create a document: '[Task Name] Workflow'.
2. List all 4-6 steps in order.
3. For each step, paste your prompt and note how long it took to get good output.
4. This becomes your template. You'll reuse it.

Run the Full Workflow End-to-End

1. Pick a new instance of your task (different topic, different data).
2. Go through all 4-6 steps using your documented prompts.
3. Time it. How much faster than doing it yourself?
4. Note anything that needs tweaking. Update your documentation.

Target: You should cut your time in half. If not, refine the workflow.

Teach Someone Else (Optional)

1. Optional but powerful: Show a colleague your workflow.
2. Have them run through it with your documentation.
3. Watch them use it. They'll find gaps you missed. Refine.
4. Now you have a repeatable system others can use.

Install & Authenticate Claude Code

1. If not already done: npm install -g @anthropic-ai/claude-code
2. Get API key from console.anthropic.com/keys
3. Run: claude-code auth and paste your key
4. Verify: claude-code --version

Create SKILL.md

1. In your project folder, create a file: SKILL.md
2. Write what the skill does, how to use it, steps, output format.
3. Be clear so Claude understands your workflow.
4. See example from Lesson 10.

Create CLAUDE.md (If Applicable)

1. If this is code-related: create CLAUDE.md in your project root.
2. Define project structure, tech stack, code standards, how to run tests/build.
3. This tells Claude how to work in YOUR project.
4. See example from Lesson 10.

Give Claude Code a Real Autonomous Task

1. In your project folder, run: claude-code "[Your task description]"
2. Give it something that would normally take 30+ minutes.
3. Examples: analyze data and create report, organize files and create summary, process CSV and create visualization.
4. Watch Claude Code work. It reads SKILL.md and CLAUDE.md automatically.
5. Verify the output. If it's good, you've automated a real workflow.

Document & Repeat

1. Document what Claude Code did (steps it took, decisions it made).
2. Save your SKILL.md and CLAUDE.md in version control.
3. Next time you need the same task done, run the same command.
4. Refine your instructions based on results.