Ultimate Vibe Coding Setup: o3 pro, Cursor, Codex, Claude Code

Key Concepts

• Vibe Coding: A highly productive coding setup leveraging AI tools.
• Asynchronous AI Agents: AI tools that can work on multiple tasks simultaneously in the background (e.g., Codex, Cursor background agent).
• Context Engineering: Providing sufficient and relevant context to AI models for optimal performance.
• Prompt Engineering: Crafting effective prompts to guide AI tools.
• Vibe Learning: Continuously improving one's understanding and skills alongside AI tools.
• Agent Mode vs. Plan Mode: Agent mode (auto-accept) for quick execution, plan mode (ask mode) for understanding and learning.

Tool Overview

• O3 Pro: A tool for solving deep bugs and refactoring. Requires sufficient context.
• Cursor: The primary IDE for vibe coding. Offers features like auto-completion and background agents.
• Clot Code: An AI agent that runs locally, used for code review, automation, and more. Can be used in Cursor or any terminal.
• Codex: A cloud-based asynchronous AI agent for code generation and task completion. Offers features like multiple versions and internet access.
• Super Whisper: An AI voice tool for transcribing speech into prompts.
• Repo Prompt: A tool for constructing well-structured prompts for O3 Pro by selecting relevant files.

Workflow and Methodology

1. Task Loading: Load multiple tasks into Codex to work on them asynchronously.
2. Version Generation: Utilize Codex's 4x feature to generate four different versions of each task.
3. Version Selection: Review the generated versions and select the most suitable one.
4. Pull Request Creation: Create a pull request from the selected version in Codex.
5. Code Review: Use Clot Code to review the pull request, acting as a senior developer.
6. Merging: Merge the pull request to a side branch (e.g., a personal branch) for safety. Avoid merging directly to production.
7. Local Testing: Pull the changes to the local environment and test them.
8. Refinement: Use Clot Code in agent mode to make further refinements to the code.
9. Protocol Following: Use custom prompts (e.g., protocol.md) to guide Clot Code in performing tasks like pushing changes to GitHub.
10. Continuous Iteration: Repeat the process with other tasks, leveraging the asynchronous nature of Codex.

Detailed Tool Usage Examples

Codex

• Task Automation: Load multiple tasks into Codex and let it work on them in the background.
• Version Comparison: Use the 4x feature to generate four different versions of a task and compare them.
• Code Generation: Use Codex to generate code for new features or bug fixes.
• Codebase Understanding: Use Codex in ask mode to ask questions about the codebase and learn about its structure and functionality.
• Environment Variables: Attach all environment files and startup commands for complex setups.

Clot Code

• Code Review: Use Clot Code to review pull requests and identify potential issues.
• Automation: Use Clot Code to automate tasks like pulling changes from GitHub and pushing changes to GitHub.
• Contextual Awareness: Provide Clot Code with relevant context by attaching files and descriptions.
• Agent Mode vs. Plan Mode: Use agent mode for quick execution and plan mode for understanding and learning.
• Custom Prompts: Use custom prompts (e.g., cloud.md) to guide Clot Code in performing specific tasks.
• Command Approval: Approve safe commands in Clot Code to avoid repeated prompts.

O3 Pro

• Complex Problem Solving: Use O3 Pro to solve deep bugs and refactor complex code.
• Context Provision: Provide O3 Pro with sufficient context by using Repo Prompt to select relevant files.
• Screenshot Integration: Integrate screenshots into O3 Pro prompts for visual context.

Repo Prompt

• Prompt Construction: Use Repo Prompt to construct well-structured prompts for O3 Pro by selecting relevant files.
• Context Building: Use the context builder feature to generate a list of relevant files based on the prompt.
• XML Tagging: Structure prompts using XML tags to clearly delineate user instructions and file contents.

Super Whisper

• Prompt Generation: Use Super Whisper to transcribe speech into prompts.
• Vocabulary Customization: Customize the vocabulary to improve transcription accuracy.
• Prompt History: Access the prompt history to retrieve previously used prompts.

Key Arguments and Perspectives

• Context is King: Prompt engineering is more about context engineering. Providing sufficient context to AI models is crucial for optimal performance.
• Asynchronous AI Agents are Essential: Asynchronous AI agents like Codex can significantly improve productivity by working on multiple tasks simultaneously.
• Code Review is Critical: Code review is essential for ensuring code quality and preventing bugs.
• Vibe Learning is Paramount: Continuously improving one's understanding and skills alongside AI tools is crucial for long-term success.
• Model Diversity is Beneficial: Using different AI models (e.g., Codex, Clot Code, Gemini) for different tasks can lead to better results.

Data and Statistics

• Codex Success Rate: Codex has an 85% success rate in terms of pull request approval, making it the best asynchronous AI agent currently.
• Cursor Agent Success Rate: Cursor agent is the second best asynchronous AI agent.

Notable Quotes

• "Prompt engineering is more like context engineering." - Tobias Lütke, CEO of Shopify
• "When you are vibe coding you need to be vibe learning."

Technical Terms and Concepts

• IDE (Integrated Development Environment): A software application that provides comprehensive facilities to computer programmers for software development.
• API (Application Programming Interface): A set of routines, protocols, and tools for building software applications.
• MRR (Monthly Recurring Revenue): A measure of predictable revenue that a company expects to receive every month.
• MVP (Minimum Viable Product): A version of a product with just enough features to satisfy early customers and provide feedback for future product development.
• CLI (Command Line Interface): A text-based interface used to interact with a computer system.
• Tokens: Units of text used by AI models for processing.

Logical Connections

• The video begins by introducing the concept of vibe coding and the tools used in the setup.
• It then provides an overview of each tool and its purpose.
• The video then demonstrates the workflow and methodology by building an AI startup (vector.ai).
• It then discusses the importance of context engineering and prompt engineering.
• The video then presents data and statistics on the performance of different AI agents.
• Finally, the video concludes by emphasizing the importance of vibe learning and continuous improvement.

Synthesis/Conclusion

The ultimate vibe coding setup involves leveraging a combination of AI tools, including O3 Pro, Cursor, Clot Code, Codex, Super Whisper, and Repo Prompt. The key to success is providing sufficient context to the AI models, continuously learning and improving one's skills, and using different AI models for different tasks. The workflow involves loading tasks into Codex, generating multiple versions, selecting the best version, creating a pull request, reviewing the code with Clot Code, and merging the changes. Vibe learning is crucial for long-term success, as it allows one to develop the taste, skill, and judgment necessary to make informed decisions.