Process over Magic: Beyond Vibe Coding - Takeaways

Some bullet points from the book to help me remember what I’ve learned.

I bought this book from Pragmatic Programmers. No affiliation, I just like their books.

My school computer room had a poster up on the wall that read, “To err is human, to really foul up requires a computer”. This is even truer with AI. It allows you amplify your progress, or your mistakes exponentially.

Prompting Strategies

• Guide the AI, don’t convince it
• Be precise, not vague (e.g., “Always run tests and verify they pass before committing code” instead of “be careful”)
• Avoid role-based prompting (e.g., “You’re an expert developer” tries to make the AI an actor which it’s not)
• Start with what you want the result to look like (Markdown, JSON, roles)
• Use examples or exclusions to narrow focus
• Focused prompts: one goal (add feature, fix bug, OR refactor)
• Stop the assistant from making things up by asking for clarification:
  • "If instructions are unclear, ask follow-up questions before continuing"
• Or getting stuck on impossible tasks:
  • "If a request is impossible, explain why and suggest alternatives"
• Iterate: plan, test, refine, compare, reuse prompts that work

Git & Version Control

• Start with clean git and passing tests
• One prompt, one commit
• Only commit working code
• Always review code before committing
• Stuck? Roll back. Don’t waste time fighting it. Try a different approach
• When you notice something else is broken: commit (if it’s working & passing tests), then immediately fix and commit separately
• Use git worktrees to work on multiple approaches in parallel

Planning & Documentation

• Large features: write detailed document (≤1000 words) as plan document
• Shape plan through iterations (can take several hours)
• Rewrite in compact format after shaping
• Walking Skeleton/MVP: start with simpler version, add to plan document, refactor to be more efficient
• Ask AI to update design docs with progress
• Write pseudocode directly into files, ask AI to implement
• Parallel designing: Ask AI to write/rewrite feature in different ways. Explore variations, different libraries. Review, tweak, rollback and try again. Think: AI guided spikes

Code Quality & Review

• Don’t trust confident responses - always verify and test
• AI doesn’t know how to write code, it just knows patterns
• AI may write bad or poorly performing code - iterate on it
• Ask AI to review for refactoring opportunities after committing
• Ensure adequate test coverage

Debugging & Problem-Solving

• Create step-by-step debugging plan (most likely → least likely causes)
• Large codebases: move in small steps; if instructions seem unclear, wrong, or unfamiliar, ask AI for more information before continuing
• Ask AI to explain how without writing code, then write it yourself (learn by doing, guided by AI)
• Ask AI to write a script for complex transformations (more efficient than having the AI do it directly)

Workflow & Sustainability

• Take regular breaks for sustainable, comfortable pace
• Stop after 5 minutes if no good results (exception: initial scaffolding)