Vibe coding, a term coined by Andrej Karpathy in February 2025, involves using AI to generate code through natural language prompts. Below are key tips for effective prompting, originally shared by Matt Palmer on Replit.
| Tip | Description | Example In-Practice |
|---|---|---|
| Checkpoint | Build in chunks, use rollbacks, avoid rewriting failed prompts | Find a point where the app "just works" then experiment. If you go down a bad path, rollback |
| Debug | Find root causes methodically, give LLMs relevant error context | Start high-level: "There is an error on the frontend", then drill-down as needed to provide context |
| Discover | Ask what YOU don't know; you don't know what you don't know | "What are some good options for a drag-and-drop interface in React?" |
| Experiment | Test different prompts, formats, and models; change what's not working | Experiment with passing the URL, wireframes, screenshots, or text to get better results |
| Instruct | Tell models what TO do, not what NOT to do | "Please add more padding to the button" instead of "Don't make the padding too small" |
| Select | Be selective with AI suggestions, avoid conflicting details | Ensure your instructions follow a cohesive train of thought. Focus on one feature at a time |
| Show | Attach screenshots and images, use full-file context snippets, and sample data | Attach screenshots and images, use our Element Selector for visual edits |
| Simplify | Keep prompts concise, pretend you're instructing a new colleague | Write instructions a human could follow. Focus on breaking down everything simply and logically |
| Think | Define the outputs precisely, consider edge cases upfront | "Please make the app responsive and mobile friendly. I'd like it to work well on my iPhone" |
Tips by Matt Palmer, Head of Developer Relations at Replit. Learn more in the DeepLearning.AI course "Vibe Coding 101 with Replit", published March 26, 2025.