Ask HN: What's your vibe coding setup, what does it cost and how do you like it?

My experience with "vibe coding" (defined below) hasn't been as good as what others report on HN.

If you’ve done vibe coding, I’d love to hear about your setup — and how it’s working for you. Feel free to respond using the format below:

  Use case:
  
  [What you're building or doing with vibe coding]
  
  Tool used:
  
  [Name(s) of AI assistant(s) or platform(s)]
  
  Cost in \$:
  
  [Amount]
  
  Cost frequency:
  
  [per day / week / month]
  
  Cost includes:
  
  [subscription / API costs / both]
  
  Satisfaction:
  
  [Of the tasks you give to the AI, what percentage are completed to your satisfaction?]
  
  Usefulness:
  
  [Overall, how useful do you find vibe coding?]

Vibe coding is a style of software development characterized by a highly conversational, low-effort interaction with AI coding assistants (such Cursor or Aider), where the developer delegates most of the implementation details to the model. Instead of writing code manually, the developer gives natural language prompts — often vague or imprecise — and accepts the AI’s output with minimal review or editing.

The workflow typically involves:

- Asking for small, iterative changes (“reduce padding”, “fix this error”)

- Relying on the AI to understand intent and generate or modify code

- Copy-pasting error messages without explanation and letting the AI fix them

- Rarely reading diffs or digging into code unless something breaks

- Prioritizing speed and exploration over precision or deep understanding

Vibe coding is often used for prototypes, side projects, or creative experimentation — situations where quick results matter more than long-term maintainability or code clarity. It represents a shift from manual programming to prompt-based software shaping, where the developer becomes more of a director than a coder.

0. https://simonwillison.net/2025/Feb/6/andrej-karpathy/

  Use case:
  Building a personal RAG system as a weekend project.
  
  Tool used:
  OpenHands and Trea, both running in Docker.
  
  Cost in $: 
  $60
  
  Cost frequency:
  One-time (over 3 weekends)
  
  Cost includes:
  Mostly API usage and infrastructure
  
  Satisfaction:
  Low (10%?) — the final system worked much worse than `ripgrep`.
  The AI often produced unusable or overcomplicated results.
  I ended up spending more time debugging than... vibing.
  
  Usefulness:
  AI is helpful for certain programming tasks, and GitHub Copilot can be useful (when it’s not overeager). But in this case, the “vibe coding” approach felt inefficient and frustrating. Might’ve been better with higher-end tools, but hard to say.