The holidays are just around the corner, and I will definitely get to know OpenRouter better – right now, I'm finishing a web application for creating photo zines, mainly for school students, and the application is based mostly on JS, HTML, and CSS. When working with AI, I divide tasks among different models (Gemini Flash, Pro, Flash Lite, and the free version of Claude). Drawing visual clues for the model is very helpful – often a prompt alone doesn't do much, and it's only when you map out the problem visually that the AI's reasoning gets a real boost. The planning mode is very helpful – it allows you to map out the required functions or the possibility of using appropriate libraries before coding, and only move on to the coding mode once the plan is approved. This saves a lot of tokens.