An LLM is a brilliant engineer that has just woken up with total amnesia. It knows every programming language in the world, but it knows absolutely nothing about your specific project.
1The Danger of Zero-Context
If you ask an AI to 'Add a loading spinner to the button', it will write the code using a generic HTML <button> and CSS. But what if your project uses a custom <CoreButton> React component and ChakraUI? The AI's code is useless. You must explicitly inject the file containing the <CoreButton> component into the AI's context so it understands your specific architectural constraints.
2Explicit File Targeting
Modern AI IDEs provide syntax (usually the @ symbol) to attach files directly to your prompt. When writing a prompt, you are no longer just writing text; you are curating a specific data packet. If building a new API route, your prompt should explicitly attach @DatabaseSchema, @RouteInterface, and @AuthMiddleware. This gives the AI the exact blueprints it needs to succeed.
3Attention Dilution
While context is required, too much context is fatal. Neural networks use 'Attention Mechanisms' to weigh the importance of input tokens. If you feed the AI 30 unrelated files, its attention becomes diluted. It might accidentally pull a variable name from a completely unrelated CSS file and hallucinate it into your backend logic. Be surgical. Only provide the exact files required for the immediate atomic task.
