I've been using it the same way. One approach that's worked well for me is to start a project and first ask it to analyse and make a plan with phases for what needs to be done, save that plan into the project, then get it to do each phase in sequence. Once it completes a phase, have it review the code to confirm if the phase is complete. Each phase of work and review is a new chat.

This way helps ensure it works on manageable amounts of code at a time and doesn't overload its context, but also keeps the bigger picture and goal in sight.

I find that sometimes this works great and sometimes it happily tells you everything works and your code fails successfully and if you aren’t reading all the code you would never know. It’s kind of strange actually. I don’t have a good feeling when it will get everything correct and when it will fail and that’s what is disconcerting. I would be happy to be given advice on what to do to untangle when it’s good and when it’s not. I love chatting with Claude code about code. It’s annoying that it doesn’t always get it right and also doesn’t really interact with failure like a human would. At Least in my experience anyways.