The “product-level memory” aspect is interesting to me. Code history can tell you what changed, but not necessarily what the product became.
I can see this becoming more valuable as coding agents make smaller, more frequent changes. At some point, someone still needs a higher-level explanation of what actually exists in the product.
Curious how you decide what qualifies as a feature versus an implementation detail.
The core idea is identifying features where you answer "yes" to the question: "Could you demo it?"
Regarding the value of such tools when working with codebases whith heavy coding agent use - absolutely. Especially when code generated by models such as Astra can be pretty unreadable (https://news.ycombinator.com/item?id=49654229)
The “product-level memory” aspect is interesting to me. Code history can tell you what changed, but not necessarily what the product became.
I can see this becoming more valuable as coding agents make smaller, more frequent changes. At some point, someone still needs a higher-level explanation of what actually exists in the product.
Curious how you decide what qualifies as a feature versus an implementation detail.
This is the template for the prompt that is used here (everything below "{copy_rule}"): https://github.com/netizer/feature-ledger/blob/main/assets/B...
The core idea is identifying features where you answer "yes" to the question: "Could you demo it?"
Regarding the value of such tools when working with codebases whith heavy coding agent use - absolutely. Especially when code generated by models such as Astra can be pretty unreadable (https://news.ycombinator.com/item?id=49654229)