Chapter 7 Advanced Techniques: Power User Features
7.1 When to Use Advanced Techniques
You’ve mastered the basics. Now: - ✓ Tackle complex problems - ✓ Work more efficiently - ✓ Get better answers - ✓ Build sophisticated systems
These patterns complement Anthropic’s current Claude Code best practices. They are starting points to test, not guarantees of output quality.
7.2 Technique 1: Plan Before Implementation
Problem: Claude rushes to answers, misses nuance
Solution: Ask for an inspectable plan, assumptions, and verification strategy before implementation. Evaluate the plan itself rather than relying on an unverifiable claim that the model “thought harder.”
Problem: Build a data pipeline that:
- Reads from 3 sources
- Cleans inconsistent data
- Validates quality
- Outputs to database
Before asking for code, ask:
"Before you write code, walk me through:
1. Data source details (formats, schemas, sizes)
2. How you'd validate consistency
3. Error handling strategy
4. Performance considerations
5. Testing approach
Then show me the implementation."7.3 Technique 2: Context Stacking
Problem: Claude forgets earlier context
Solution: Explicitly reference previous decisions
Earlier we decided:
- Use pandas for data processing
- Write to PostgreSQL
- Handle nulls by forward-fill
- Validate row counts match source
Now implement step 4: [your step]
Keep these decisions in mind.7.4 Technique 3: Role-Based Prompting
Problem: Generic answers that don’t fit your context
Solution: Tell Claude what role to play
Review this from the perspective of a senior data scientist at a marketing agency.
I'm building a customer segmentation model.
The stakeholders are:
- VP Marketing (needs actionable segments)
- Data Engineering (needs reproducible pipeline)
- Analytics (needs SQL-friendly outputs)
Design the segmentation approach that satisfies all three.7.5 Technique 4: Constraint-Based Design
Problem: Solutions don’t fit real constraints
Solution: State constraints explicitly upfront
Design a recommendation system with these constraints:
- Python 3.10 (the version pinned in production)
- No TensorFlow/PyTorch (not allowed in company)
- Must run on company GPU (NVIDIA A100)
- Must score 1M items in < 2 seconds
- Storage limited to 100GB
What's the approach?7.6 Technique 5: Assumption Surfacing
Problem: Claude makes wrong assumptions about your needs
Solution: Have Claude list assumptions, then validate
"Before you build this, list all your assumptions about:
1. Data shape and size
2. Performance requirements
3. Error tolerance
4. Team skill level
5. Business constraints
I'll tell you which are wrong so you adjust."7.7 Technique 6: Comparative Analysis
Problem: One solution might not be best
Solution: Ask Claude for multiple approaches
"Show me 3 different ways to [task]:
Approach A: [description]
- Pros:
- Cons:
Approach B: [description]
- Pros:
- Cons:
Approach C: [description]
- Pros:
- Cons:
Which would you recommend for [specific context]?"7.8 Technique 7: Systematic Testing Strategy
Problem: Incomplete test coverage
Solution: Ask Claude to design test strategy first
"Design a comprehensive test strategy for [function]:
1. What are the possible inputs?
2. What could go wrong?
3. What edge cases exist?
4. How do you test robustness?
5. How do you test performance?
6. How do you test failure modes?
Then write tests for all categories."7.9 Technique 8: Documentation-First Design
Problem: Code without clear purpose
Solution: Have Claude write documentation first
"Write the documentation for this feature first:
Title: [what it does]
Purpose: [why it matters]
How to use: [examples]
Inputs: [types and formats]
Outputs: [expected results]
Edge cases: [special handling]
Performance: [expected runtime]
Then implement code that matches this documentation."7.10 Technique 9: Architectural Review
Problem: Missing the big picture
Solution: Have Claude diagram/explain architecture
"Draw (in ASCII or describe) the architecture for:
1. Data inputs (where from?)
2. Processing stages (what happens when?)
3. Storage (where does it live?)
4. Outputs (where does it go?)
5. Monitoring (how do we know it works?)
6. Failure handling (what if it breaks?)
7. Scaling (what if data grows 100x?)
Then show the implementation."7.11 Technique 10: Iterative Refinement
Problem: One pass isn’t perfect
Solution: Use feedback loops
Round 1: "Give me a basic implementation"
[you review]
Round 2: "Add error handling for [issues found]"
[you review]
Round 3: "Optimize for [performance issue]"
[you review]
Round 4: "Add docs and tests"
[you test]
Round 5: "Final review - anything missing?"
[ready for production]7.12 Advanced Pattern: Complete System Design
Combine techniques for major projects:
Step 1: UNDERSTAND
"I need to build [system].
Constraints: [list]
Team: [size and skills]
Timeline: [how long]"
Step 2: DESIGN
"Before coding, design the architecture.
Show: inputs → processing → outputs"
Step 3: ASSUMPTIONS
"What assumptions are you making?
Let me validate them."
Step 4: COMPONENTS
"Break it into 4-5 key components.
For each: purpose, inputs, outputs."
Step 5: TESTING
"Design the test strategy.
What could fail? How do we prevent it?"
Step 6: IMPLEMENTATION
"Now implement [Component 1] with full docs and tests"
Step 7: INTEGRATION
"Connect [Component 1] to [Component 2].
Show integration points."
Step 8: FINAL REVIEW
"Review the complete system.
Any performance or robustness issues?"
Step 9: DEPLOYMENT
"How do we deploy this safely?
What monitoring do we need?"7.13 When to Use Each Technique
| Situation | Use This |
|---|---|
| Simple task, clear requirements | Basic prompting |
| Complex problem | Multi-step thinking |
| Need consistent context | Context stacking |
| Domain-specific needs | Role-based prompting |
| Real-world constraints | Constraint-based design |
| Unsure of requirements | Assumption surfacing |
| Choosing approach | Comparative analysis |
| Quality concerns | Systematic testing |
| Unclear purpose | Documentation first |
| Large system | Architectural review |
7.14 Asking Claude About Advanced Techniques
I'm building [system].
Current challenge: [specific problem]
I want to use [technique name] to solve it.
Here's my context: [describe situation]
Can you:
1. Help me apply [technique] correctly
2. Show me the process
3. Help me think through the problem
4. Suggest what I might be missing