Before writing code, collaborate with the LLM to create a comprehensive project plan, which you can refer back to throughout the development process. After creating the first draft of your plan, remove features you don't like and mark certain features as 'won't do' if they're too complicated, while keeping a section of ideas for later to inform the LLM that they are out of scope for now.

Spend ample time using LLMs to define the scope and architecture of your project before transitioning to coding tools like Cursor.

Monitor the LLM's responses closely; if it keeps generating incorrect code, reassess the context you're providing to improve its output.

Write clear instructions for the LLM, using the appropriate naming conventions for each tool. Founders have successfully written extensive instructions to guide their AI coding efforts.

Work with the LLM to implement your plan section by section, explicitly stating which section to work on at a time, and check that it works before moving on.

Write high-level integration tests instead of low-level unit tests. Simulate user interactions to ensure features work end-to-end before moving on to the next feature.

AI tools will make it easier to generate low-level code, improving autocomplete capabilities and potentially producing high-quality code.

Classical training in coding helps engineers identify good versus bad code, which is essential for effective vibe coding.

Vibe coding is a practice that can significantly improve your coding skills if you're willing to experiment and adopt best practices.