Deep dives into how real AI systems work and how to build systems that work in production. The aim is to explain complex Ai topics in a simplified and practical manner.
For almost two years, I used the same Claude prompt without questioning it.
It worked, so I never had a reason to.
Then I changed just seven words.
Same task. Same model. Same day.
Claude followed my instructions noticeably better.
I wasn’t discovering a clever prompt trick. I was discovering that one of my oldest prompting habits had quietly become outdated.
That sent me down a rabbit hole through Anthropic’s own documentation.
What I found surprising was that several of the prompting habits that made Claude better in 2023 now have official alternatives — or are no longer the default recommendation at all.
If you’ve been prompting Claude the same way for a year or two, there’s a good chance you’re carrying around…
Deep dives into how real AI systems work and how to build systems that work in production. The aim is to explain complex Ai topics in a simplified and practical manner.
There is this useful command in Claude Code "/claude-api prompt-audit" which analyzes your prompts, CLAUDE.md, skills etc. and recommends how to rewrite some of them for new model to digest better. Don't be missled by 'claude-api' part in the…
