Anthropic just shared something I think is worth paying attention to if you’re starting to build tools with Claude. And while I know we've covered this inside our AI Fundamentals lives, and other lessons... it's worth another mention now that Anthropic is also talking more about it.
Their main point? 𝗠𝗼𝗿𝗲 𝗶𝘀𝗻’𝘁 𝗮𝗹𝘄𝗮𝘆𝘀 𝗯𝗲𝘁𝘁𝗲𝗿.
A longer prompt, more rules, and the most powerful model can sometimes make your AI tool more expensive without making it better.
Anthropic recommends things like:
• Keep your instructions clear and simple
• Remove old or repeated rules
• Don’t ask Claude to “think deeply” when the task is simple
• Reuse information instead of making Claude process the same thing over and over
• Don’t generate long answers when a short one will do
This really matters if you ever turn one of your Claude experiments into a tool people pay to use.
When you’re just testing something, you’re usually asking: 𝗖𝗮𝗻 𝗜 𝗴𝗲𝘁 𝘁𝗵𝗶𝘀 𝘁𝗼 𝘄𝗼𝗿𝗸?
But once other people start using it, you also need to ask: 𝗛𝗼𝘄 𝗺𝘂𝗰𝗵 𝗱𝗼𝗲𝘀 𝗶𝘁 𝗰𝗼𝘀𝘁 𝗲𝘃𝗲𝗿𝘆 𝘁𝗶𝗺𝗲 𝘀𝗼𝗺𝗲𝗼𝗻𝗲 𝘂𝘀𝗲𝘀 𝗶𝘁?
One thing I especially liked from this article is the idea that our prompts can get cluttered over time...(Premium & VIP... this brought me back to my car trip analogy and infographic!)
We keep adding rules every time something goes wrong, and eventually we end up with a huge prompt full of instructions Claude may not even need anymore.
This is was we call 𝗽𝗿𝗼𝗺𝗽𝘁 𝗱𝗲𝗯𝘁. 😅 So here’s a good question to ask the next time you build something:
𝗗𝗼𝗲𝘀 𝗖𝗹𝗮𝘂𝗱𝗲 𝗮𝗰𝘁𝘂𝗮𝗹𝗹𝘆 𝗻𝗲𝗲𝗱 𝗮𝗹𝗹 𝗼𝗳 𝘁𝗵𝗶𝘀?
🔗 See Anthropic’s full guide: 𝗛𝗘𝗥𝗘