Content
@
https://opensea.io/collection/dev-21
0 reply
0 recast
2 reactions
Stephan
@stephancill
Anyone here building large projects using Claude code? My current approach is to 1. write a context doc describing the problem and solution with product requirements 2. produce a technical specification via a back and forth with a thinking model based on the requirements doc 3. Get a thinking model to develop a daily plan for an engineer 4. Give Claude code the requirements, spec, and plan and tell it to implement one day and write do a write up in a standup doc when it’s done. 5. When it’s done I review the changes and ask it for revisions until I’m satisfied 6. After that I start another Claude session to implement the next day with all the docs and also include the relevant standup docs. Rinse and repeat This seems to work really well. Curious if others are doing it differently and getting better results Shoutout @jc4p for sharing this approach
9 replies
1 recast
51 reactions
agusti
@bleu.eth
I have a few tricks (cli tools) that let the AI keep context of the codebase (ast trees, files etc, import paths) all half baked into some repos as they grew and needed I should split these into separate mini-tools anyone can plug to a repo
2 replies
0 recast
11 reactions
agusti
@bleu.eth
after trying and failing a lot to build large-ish projects with llms, failure modes kinda emerge into patterns so trying to fight these as they arise. more than having any pre-conception or plan before going at it
1 reply
0 recast
2 reactions
agusti
@bleu.eth
imho the biggest unlock is when they can prepare/fetch their own context without relying on merely human input, as they can mid-course-correct or get extra context needed as they go
1 reply
0 recast
4 reactions
Dylan
@epicdylan
Sounds great!
0 reply
0 recast
1 reaction
