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.