Skip to content

Chapter 5: Building with AI -- Resources

Curated resources for deeper exploration of topics covered in this chapter.

Frameworks from This Chapter

  • 8 Patterns for Effective AI Coding -- Context first, concrete examples, iterative refinement, architecture ownership, test-driven prompting, error escalation, checkpoint commits, and ruthless review.

Tools & Platforms

AI Coding Tools -- Inline Autocomplete (Level 1)

  • GitHub Copilot -- 88% of accepted suggestions survive to production; 30% acceptance rate; 55% faster task completion.
  • JetBrains AI -- AI assistant integrated into IntelliJ, PyCharm, and other JetBrains IDEs.

AI Coding Tools -- IDE Integration (Level 3)

  • Cursor -- AI-native code editor; supports Claude, GPT, and Gemini; Cursor 2.0 supports 8 simultaneous agents.
  • Windsurf (Codeium) -- AI-native code editor with conversational refinement capabilities.

AI Coding Tools -- Agentic (Level 4)

  • Claude Code -- Terminal-based agentic coding; plan mode, subagents for parallel execution, hooks for review checkpoints.
  • Devin (Cognition Labs) -- Autonomous AI software engineer; used by Goldman Sachs, Santander, Nubank; 20x efficiency on security fixes.

AI Coding Tools -- Orchestration (Level 5)

AI Coding Infrastructure

  • Model Context Protocol (MCP) -- 10,000+ active servers, 97 million monthly SDK downloads; standard for AI-tool integration.
  • AGENTS.md -- Convention for AI agent instructions; appears in 60,000+ open source projects.
  • Supabase MCP -- Database operations through MCP; create branches, test migrations, apply to production in a single conversation.
  • Remotion -- React video creation; Claude Code skill teaches animation APIs for programmatic video generation.

Chat Interfaces

  • Claude.ai -- Architecture discussions and learning new frameworks.
  • ChatGPT -- General AI conversations and code generation.

Further Reading

Research & Data

Community & Learning

  • ZoomInfo AI Development -- 400+ developers seeing 6,500 daily suggestions with 33% acceptance rate.
  • Skywork AI -- Turned a 6-month roadmap into 3 weeks using the Human-AI Development Loop.

The 5 Levels of AI-Assisted Development

Level Name Example Tools Key Metric
1 Autocomplete GitHub Copilot inline 88% survive to production
2 Generation Chat interfaces 8.69% more PRs (Accenture)
3 Iteration Cursor, Windsurf 10-15 min human checkpoints
4 Agents Claude Code, Devin 20x efficiency on targeted tasks
5 Orchestration GitLab Duo, Cursor 2.0 95% of enterprise pilots fail

Progression Guidance

  • Progress one level every 2-4 weeks for durable capability
  • Jumping two levels in a week causes regression within a month
  • Level 3 to Level 4 is the biggest jump -- requires shift from implementer to architect/reviewer
  • Invest 6-12 weeks accepting slower initial output to build review skills at Level 4