https://www.youtube.com/watch?v=gYLNxUxVomY
Main ideas from the video "I spent 400+ hours in Cursor, here's what I learned":
📂 Create a folder for instructions or prompts.
Use markdown files and comments in the code to describe the project, technology stack, technology, and database structure, rules of behavior, and important instructions. Regularly updating these files helps AI better understand the project context. Cursor does not always parse framework/library documentation well, so it's better to copy them - check the documentation yourself!
Create a file roadmap.md to track the project's status, current goals, and future steps.
Use additional tools, such as Perplexity for searching information and Whisper Flow for voice input. Speaking is much faster than typing! Separately understand your architecture and brainstorm better concepts, this can be done with a "thinking" chat like o1. Do not let AI make big decisions.
⚠️ Danger: Not understanding the code created by AI accumulates "technical debt".
(Video from January 17 regarding r1 and o3) Choosing the right AI model is important - when generating code, Sonnet 3.5 has an advantage, as it provides high accuracy. Queries need to be built in detail: describe the task, add file tags, instructions, and context.
🗣 Chat: For questions, getting information, lists.
✍️ Composer: For making changes to the code, especially complex ones, use Agent mode.
There are tricks like "the less code, the better" or "think like a senior developer", "start with three paragraphs of thoughts", "do not delete comments", to get quality results. Do not overload AI with large instructions, but break tasks into parts.
- The fewer lines of code, the better
- Proceed like a Senior Developer
- DO NOT STOP WORKING until...
add:
- v0: For quick creation of first design versions. 🎨
- Claude chat: For consultations, brainstorming. 🤔
- Lavable/Bolt: For quick creation of MVP backend (with caution, control technical debt).
mentions https://repoprompt.com/
#cursor #promts