I keep running into this: hand a Figma file to Claude Code/Cursor, get a good first implementation, then the design changes and the agent has no memory of what changed or why. Curious if others have hit this and what you're doing about it, if anything.
Comments URL: [https://news.ycombinator.com/item?id=49595961](https://news.ycombinator.com/item?id=49595961)
Points: 2
source & further reading
news.ycombinator.com — original article
Ask HN: What are other "uncrackable benchmarks"?
Ask HN: How has AI changed your QA validation workflows?
Ask HN: How do you break AI agents out of the "apologize and repeat" loop