@thinkwithv: Anthropic deleted 80% of Claude Code's instructions. On purpose. So I tested it on my own AI product. I graded 3 versions of my AI's instructions like my users would. The leanest won. The catch: leaner still has to be clear, and you still test the outputs. But a few sharp lines beat my old paragraphs. Makes me wonder how much of "being thorough" is just not trusting it to think. Where do you over-instruct your AI? #AI #PromptEngineering #Marginalia

V | AI for Thinkers

Open In TikTok:

Region: US

Saturday 15 August 2026 01:21:42 GMT