Cursor users spend more time prepping context than writing code
A developer who uses Cursor heavily noticed that the actual coding has become the fast part — the slow part is everything around it. Before each AI request, they have to locate the right files, re-explain the project's structure and design decisions, clarify what changed recently, and double-check that the AI is looking at the correct code.
This overhead repeats every session. Their proposed fix is a lightweight tool that and watches the repository continuously.
Instead of rescanning everything on each request, it tracks only what actually changed and keeps an up-to-date map of , , and recent edits. When you ask Cursor to do something, this tool would have the relevant context already loaded — fewer tokens consumed, fewer wrong assumptions, no more "please read these 14 files first." The developer posted to ask whether others share the frustration and whether such a tool would actually get used.
Key points
- Explaining project structure, finding files, and describing recent changes can take longer than the coding itself
- The frustration comes from repeating this setup work every session, not just once
- The proposed solution is a local agent that watches the repo and updates only changed parts incrementally
- Having context pre-loaded would reduce wasted tokens and cut down on the AI making wrong assumptions
- Cursor's @codebase feature attempts something similar, but automatic incremental change tracking is not yet a solved problem