Claude Code is all you need
TL;DR Highlight
A vim and terminal-centric developer's honest experience using Claude Code since June — covering both the potential and limits of vibe coding.
Who Should Read
Developers who've tried Cursor, Copilot, or similar AI coding tools but found them a poor fit for their workflow. Fullstack developers wanting to rapidly prototype with vibe coding.
Core Mechanics
- The author tried Cursor, Cline, Zed, etc. but they didn't fit the vim + terminal workflow. Claude Code was terminal-native from the start, requiring zero adaptation.
- Cancelled GPT $20/month subscription and switched to Anthropic, then upgraded to $100/month MAX plan within days for Opus access. Claude Code is fundamentally a developer productivity tool.
- For new projects, writing a 500+ word SPEC.md with explicit tech stack choices is essential — without tech stack guidance, the model picks overly complex architectures that fail at basic functionality.
- When Claude tries to process files one-by-one manually (e.g., converting 100 JSON files), redirecting it to write a script is dramatically more efficient.
Evidence
- Interview problem: out of 9 technical interviews, only 2 candidates could code without AI. The rest couldn't write basic SQL queries when AI was removed, and didn't understand tradeoffs of AI-generated code.
- `--dangerously-skip-permissions` usage tip: the author recommends it for trusted projects to avoid constant permission prompts, significantly speeding up agentic workflows.
- One-shot project generation success depends heavily on SPEC.md quality and explicit tech stack specification.
How to Apply
- For one-shot new projects: write a 500+ word SPEC.md first, explicitly specifying tech stack (language, framework, DB). Without tech stack guidance, models choose overly complex setups that don't even achieve basic functionality.
- When Claude tries to manually process files one-by-one (e.g., 100 JSON files), redirect it: 'Write a script to do this in batch' — orders of magnitude faster.
- For trusted projects, `--dangerously-skip-permissions` eliminates constant permission prompts and dramatically speeds up agentic workflows.
Code Example
snippet
# One-shot app creation example
claude -p "Read the SPEC.md file and implement it"
# Shorthand for dangerously-skip-permissions (sandbox environments only)
IS_SANDBOX=1 claude --dangerously-skip-permissions
# Delegating repetitive tasks to a script
# Bad example: Claude manually processes 100 JSON files one by one
# Good example: "Write a script that converts 100 JSON files from old_shape to new_shape"Terminology
vibe codingBuilding software through conversation with AI without directly viewing or modifying code. The developer only provides requirements and lets AI handle language, framework, and implementation choices.
one-shotGetting a complete working result from a single prompt. If a working app comes out on the first try with no follow-ups, that's a one-shot success.
--dangerously-skip-permissionsA Claude Code flag that skips file access permission prompts. Speeds things up but grants broad access, so only use in trusted environments.