Lukas Hermann @lukashermann.dev · Dec 12 I've been very productive with Claude using this method: 1. Create a Project 2. Describe the project + tech stack 3. Add services and type definitions (project knowledge) 4. Start a new chat for every feature No. 4 is crucial, chats with lots of code deteriorate quickly!
23 likes 6 replies
Replies
Do you find that 3. gets out dated too quickly?
True that! Going to do some livestream coding with Cursor and that’s something I do a lot, too
Not only that, but you'll also use up more tokens faster by not breaking things up into new chats whenever possible. The longer the chat, the more tokens get used each time you ask a new question because Claude needs to look at the whole chat for context.
The most cumbersome about using Claude directly for me is keeping up with what the state of the files are. Once things change, it feels like it is not worth the effort to keep referencing them. What am I missing? How does that not drive you crazy with your flow?
Or just use an AI IDE like cursor!
That looks pretty cumbersome. Why not use Windsurf or Cursor? Same flow, but adding files is so much easier