Using LLMs as judges for evaluating outputs is promising, but it requires careful prompt engineering to ensure accurate grading of responses.

When the AI development environment gets stuck, try pasting your code directly into the LLM's website to get a different perspective on the problem.

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

Reading model outputs closely is essential; it helps identify misunderstandings and refine prompts for better performance.

The programming paradigm is evolving; now, prompts for LLMs are written in English, making software development more accessible.

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

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.

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

There are three types of evaluators: deterministic code-based evaluations, LLM-as-judge techniques for more complex assessments, and human annotation for real-time feedback.