Skip to content

Latest commit

 

History

History
157 lines (103 loc) · 4.44 KB

tool-selection.md

File metadata and controls

157 lines (103 loc) · 4.44 KB
title description
Tool Selection
Choosing the right AI coding tool for structured development

Tool Selection

TLDR

Use Cline (VS Code extension) for most projects. Its Plan Mode → Act Mode workflow matches this methodology perfectly, and terminal approval keeps you in control.

Use Claude Code (terminal) for quick exploration or if you prefer command-line workflows.

Use Claude Web for brainstorming sessions and phase audits—not for writing code.

Extended thinking is required. Without it, AI makes shallow decisions that cost you later.

Cline: The Default Choice

Cline is a VS Code extension that works in two modes:

Plan Mode: AI reads your project, proposes an approach, asks clarifying questions.

Act Mode: AI executes the plan, creating/editing files and running commands.

This matches our workflow exactly:

  1. "Can we plan task 3?" → Plan Mode
  2. Review the plan, adjust if needed
  3. "Proceed" → Act Mode
  4. Approve terminal commands as they come
  5. Verify the result

Why terminal approval matters:

Cline wants to run: rm -rf node_modules && npm install
[Approve] [Reject] [Edit]

You see every command before it runs. This prevents disasters and keeps you aware of what's happening.

Setup:

  1. Install "Cline" from VS Code extensions
  2. Configure with your Anthropic API key
  3. Enable extended thinking
  4. Set autoApproveCommands: false

Claude Code: The Alternative

Terminal-based tool. You type natural language, it writes code.

$ claude "add authentication to this Express app"

Advantages:

  • Faster for quick tasks
  • No VS Code dependency
  • Explicit thinking levels (think hard, ultrathink)

Disadvantages for this methodology:

  • Less structured Plan → Act separation
  • You have to manually enforce approval workflows
  • Context loading is more manual

Use it when: You're doing exploratory work, quick fixes, or you just prefer terminal.

Skip it when: You're doing structured, multi-phase development.

Claude Web: For Specific Phases Only

Use for brainstorming (before development):

"I want to build X. Help me scope an MVP."

Extended thinking shines here. Let AI reason through complexity before you commit to building.

Use for phase audits (between development phases):

  1. Push code to GitHub
  2. Start fresh Claude conversation
  3. "You're a senior dev. Review this repo for issues."
  4. Fix what it finds

Don't use for: Writing code during active development. The web interface isn't designed for file management and iterative coding.

Extended Thinking: Non-Negotiable

Extended thinking is when AI reasons through a problem before responding. Without it:

Simple prompt → shallow answer:

"Design a matching algorithm"

AI: "Here's cosine similarity..." [generic solution]

With extended thinking:

AI thinks: "1000 users, needs to be fast, they mentioned budget constraints, vector DB might be overkill for MVP..."

AI: "For your MVP scale, simple tag overlap will be faster and cheaper than vector similarity. Here's why, and here's when you'd upgrade..."

Extended thinking catches problems before they become expensive.

Enable it:

  • Claude Web: Settings → Extended Thinking → On
  • Cline: Automatic with Claude models
  • Claude Code: Use think hard or ultrathink keywords

Cost Expectations

Task Type Tokens Cost (Sonnet)
Simple task 5-10k $0.05-0.10
Medium task 15-25k $0.15-0.25
Complex task 30-50k $0.30-0.50

Typical MVP (25-35 tasks): $200-400

The methodology adds ~20% cost overhead (more documentation, confidence scoring). It saves 10x that in avoided rework.

Quick Decision

Building something production-worthy?
├─ Yes → Cline
└─ No, just exploring → Claude Code

Need structured workflow with approval?
├─ Yes → Cline
└─ No, I want speed → Claude Code

Brainstorming or auditing?
└─ Claude Web

For 90% of readers following this guide: use Cline.

Proof It Works

This methodology built RISE—a desktop Electron app—in 4 weeks for ~$400 in tokens. Production-ready, documented, maintainable.

Not a toy demo. A real application with multi-process architecture, build pipelines, and comprehensive tests.

Next: The Brainstorming Session — The conversation that prevents expensive mistakes.