Matt Pocock – Claude Code for Real Engineers

Here’s What You Get:

Core Philosophy (Matt Pocock – Claude Code for Real Engineers)

1. AI is a planning + execution assistant (not a replacement)

• Claude Code works best when you:
  • Plan first
  • Validate thinking
  • Then let AI implement
• Treat AI output like junior engineer code → always review it carefully

Key idea: AI generates drafts, engineers ensure quality.

Key Workflow (Step-by-Step)

1. Start with Plan Mode (Critical Step)

• Before writing code, you ask Claude to:
  • Explore the codebase
  • Understand context
  • Generate a plan
• It returns:
  • Architecture steps
  • Unknowns
  • Clarifying questions

This prevents bad assumptions early

2. Force Clarifying Questions

• Claude should not jump into coding
• Instead, it asks things like:
  • Where data comes from
  • How strict validation should be
  • Edge cases

You refine requirements before implementation.

3. Break Work into Multi-Phase Plans

• Large features = multiple phases
• Each phase fits inside the AI’s context window

Example:

• Phase 1: Setup CLI structure
• Phase 2: Add parsing logic
• Phase 3: Handle edge cases

This avoids context overload

4. Manage Context with GitHub Issues

• Store:
  • Plan
  • Progress
  • Decisions
• Then:
  • Clear AI context
  • Reload from the issue later

This lets you work on large systems across sessions

5. Use Rules (Memory File)

Matt configures Claude with rules like:

• Be extremely concise
• Always list unresolved questions
• Follow specific workflows

This makes outputs:

• Cleaner
• More structured
• More predictable

6. Execute Fast with Auto-Accept Mode

• After planning:
  • Let Claude implement quickly
• Then:
  • Review changes in VS Code

Combines AI speed + human judgment

Tools & Stack

Typical setup:

• Terminal + Claude Code
• GitHub CLI (for issues & workflow)
• VS Code (for reviewing diffs)

What You Can Build

Using this approach, Claude Code can handle:

• Large feature development
• CLI tools
• Multi-file refactors
• Automation systems

And even:

• Full tools from plain English prompts (in some workflows)

Limitations & Risks

Matt (and real teams) highlight key issues:

1. Context Window Limits

• AI forgets things in long sessions
  → Solve with planning + external storage

2. Over-Reliance Risk

• Engineers may accept bad code
  → Always review like a PR

3. Wrong Directions

• AI can send you down bad paths
  → Expect some wasted time weekly

Key Takeaways

• Planning > Prompting
• Break work into phases
• Store context externally
• Force clarity before coding
• Treat AI as a draft generator

The real skill is not “using AI”
It’s designing a system where AI can’t fail badly

Who It’s For

Best suited for:

• Software engineers
• Technical founders
• Advanced AI users

Less ideal for:

• Beginners with no coding knowledge (without guidance)