Okay, I've been using the new Projects feature in Claude.ai for a couple of weeks now, and I have to say, it feels like a significant step up from just a persistent chat. It's not just a shared context; it's a proper workspace.
For my latest Python CLI tool, I created a Project and uploaded the whole codebase. The key difference? The AI seems to *understand* the project structure. When I ask "how does the config loader in `src/utils/config.py` interact with the main app?", it references the actual files correctly, not just a vague memory from a past message. I can also set specific instructions for the entire Project, like "This project uses type hints and pytest. Prefer descriptive variable names."
Here's what I'm finding most useful:
* **Centralized Instructions:** Setting a coding style or framework preference once and having it apply to all future conversations in that Project is a game-changer.
* **Better File Awareness:** It handles multiple files better than a chat where you're constantly re-uploading. The context feels more stable.
* **Onboarding Helper:** I shared the Project link with a colleague, and they could jump in and ask questions about *my* codebase directly, which was fantastic for collaboration.
However, it's not perfect yet. I've noticed it can still get "lost" if the codebase is very large, and I sometimes have to remind it to look at a specific recently-uploaded file. The big question is whether it justifies the price jump for the Team plan for solo developers.
Has anyone else moved from using a long-running "dev chat" to a dedicated Project? What's your workflow? Are you using it more for solo work or team stuff?
-- Weave
Prompt engineering is the new debugging