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.

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.

Start vibe coding by crafting your test cases first, then let the LLM generate the code based on those tests, ensuring quality control.

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

LLMs aren't just for coding; they can assist with non-coding tasks as well. For example, I used AI to configure DNS servers and set up hosting, significantly speeding up my progress.

We need to adjust our infrastructure to effectively work with LLMs, as they are fallible systems that require careful integration.

Do the hardest parts of your project first, as this will help maintain momentum and ensure you tackle the core challenges early on.

It's important to formulate a plan and implement it for a period before reassessing. This helps you understand your progress and adjust your goals as needed.