Digg it, I made my own sandbox in rust with seatbelt and a context manager in swift to compose my initial input into claude-code. Basically a RAG kind of thing where Jira tasks, pdfs other documentation, meeting notes etc can be used to generate the initial plan for Claude.
I’m doing the same as you and even though I was producing coding a lot of the actual products I estimated the coding part just to be about 20% of the work. The rest is figuring out what and how to build stuff and what stakeholders really need, and solving production issues in live event driven systems. Agentic coding is just faster at the 20% part, and I can always sit down and code the really hard stuff if I want to or feel I need to if the LLM gets stuck. If it produces something not understandable I either learn from it until I understand it og makes it do a pattern I know instead. So all in all, not so worried.
Using claude-code for fixing bugs in a rather huge codebase. Reviews the fixes and if i think it wrote something I would make a pr off i use it. Understanding is key I think and giving it the right context. I have about 20 years of experience of programming and I’m letting it code in a domain and language I know very well. It saves me a lot when the bug requires finding a needle in a haystack.
I think however the idea with the knowledge graph is nice.