I need help with getting the most out of CC
So, this is a post of admission. I admit that I need help.
I am a professional engineer, been working in technology for ± 15 years. Had CTO role in medium sized company (300 FTE), had several other lead roles but always kept my hands dirty.
I try to get with the times and I try to get the most out of Claude Code. I have the 20x subscription and off I went. I have been using it on and off for a year now, and I have to say that I am underwhelmed with my ability to make it work.
- Note that I am trying in earnest. I do feel I have a fair assessment of the capabilities of today's models. I notice that my workflow typically ends up in this: I create a single large PLAN.md with subphases. I spent quite a bit of time on this. I also create an architecture document that holds some invariants (like tech stack, test strategy, etc).
- I have a prompt for planning a subphase, which consist out of an analysis, clarification and write phase. I also have a standard prompt for implementing a subphase using those detailed implementation phases.
- I use an adverserial review agent (actually as part of my implementation prompt), on Opus Max on each subphase. The implementation phase also details that each sub-subphase should be committed and every commit should have a green test suite run.
I just let it cook then. But here is the thing; I get lost a lot. In particular:
- I start of reading all the output (plans, docs, questions) in detail, but I get fatigue real fast and then just glance at them or scroll through them. I typically take the recommendation for any clafification or decision Claude asks me, and I really notice that my thinking input is getting more limited as the day drags on.
- I don't read the code anymore. I used to care about code structure for all the right reasons, but I just can't bring myself to read all the code it produces, it is just mentally draining. When I read other peoples code (say, 4 years ago), I always tried to establish what mental model the writer had in mind. That is something I can no longer do as there is no mental model, there wasn't a human involved at all. The result is that I just "yeah sure whatever" it.
- In the end, I have a project with an enormous code base that I don't know, with a structure that I don't know nor understand. The code might be good, it might be full of holes. The only thing I know is that when I use it, it **seems** to work but I have no idea about security, logic, etc.
Even with all these problems, you'd think I would just make smaller pieces of work so I retain a sense of control. But no, it is so tempting to just let it do larger and larger things that I find myself disappointed in myself not being able to withstand it. Nevertheless, I love the feeling that I am productive, until it is already too late and I have lost all control over the repo.
No, if you stuck around with me until now, thanks a million because here is my question: what am I doing wrong? I have tried to lookup ways of working with CC but it's all just shrills out there that suddenly have the power to create something without any regard for security. I am really looking for a way that maximizes the value of CC for me, which I guess would result in:
- A code base I feel I know how it works, where I know what logic is where and how it is proven.
- A way of working that is not about me reading pages and pages of text (it really is exhausting) and trying to convey my intent better.
- A way of working that is either engaging (so I stay focused) or hand off for hours on end (so I can do other focused work) with a strong preference to the latter.
I am sorry for the wall of text, but any help is appreciated. I really feel I am doing this whole thing wrong. I sincerely start to doubt if my brain is wired for this, and I need help learning new ways.