I’ve watched so many smart people treat the newest AI models like it’s still 2023. They open a fresh chat, paste a giant wall of text, hit enter, and hope for the best. Then they wonder why the output feels generic.
I came across a post from an AI professional that finally put words to this problem, and it stuck with me. The core idea is simple: Claude-5 might be the smartest model you’ve ever touched, but most of us are still prompting it like a basic chatbot. The creator laid out five concrete fixes, and each one changed how I think about working with these tools.
Let me walk you through what the original poster shared, step by step.
1. Stop re-explaining yourself. Use Projects.
Here’s the pain the author points to: every new chat starts at zero. No memory, no context. So you end up uploading the same PDF to five different chats and re-typing the same background every single time.
The fix: Delete your scattered folders and create a Project instead. Your files and your instructions live in one place, and every chat inside that Project can see them.
Why it works: You set the context once. After that, you’re building on top of it instead of starting over. I think this alone saves most people a shocking amount of repeat typing.
2. Turn your long prompts into Claude Skills.
The expert makes a sharp observation here. If you’re pasting the same prompt for the tenth time, you’re wasting context and, in his words, wasting your life.
The fix: The moment you paste something for the second time, turn it into a Skill. Your “about me” file? That should be a Skill. Save it once, then type / to call it forever.
Why it works: A Skill is reusable and instant. Instead of hunting through old chats for that perfect prompt you wrote last month, you trigger it with a keystroke. The savvy professional frames Skills as the layer almost nobody bothers to build, which is exactly why building it gives you an edge.
3. Let Claude interview YOU with AskUserQuestion.
This one flips the usual dynamic. Claude often knows exactly what information it’s missing. The problem is you’re the one holding that information, and you don’t always know to hand it over.
The fix: End your prompt with a line like this:
“Before you start, ask me everything you need to know to get this right.”
Why it works: Instead of guessing and filling gaps with generic assumptions, the model surfaces the missing pieces first. You answer a few pointed questions, and the final output lands much closer to what you actually wanted. I was genuinely surprised how much better results got just from adding that one sentence.
4. Hand over outcomes, not tasks.
The creator draws a clean line between a task and a goal. If you already know every step and you’re just dictating them, you gave it a task. And a task boxes the model in.
The fix: Hand Claude-5 the outcome you want, plus everything you know about the situation. Then let it figure out the rest. The mind behind this post notes that the newer model can plan, ask questions, and work for hours, so treating it like a step-follower wastes most of its ability.
Why it works: When you describe where you want to end up instead of every move to get there, the model can find paths you wouldn’t have thought of. You stop being the bottleneck.
5. Never run on default effort.
Effort, as this industry pro explains it, is basically how long the model thinks before it answers. Running the smartest available model with only ten seconds of thinking is a waste of what it can do.
The fix: Match the effort to the job:
- Use high effort for real, meaty work.
- Push to max effort for your hardest, most tangled problems.
- Drop to a lighter, faster model for quick drafts and throwaway stuff.
Why it works: You’re spending compute where it matters and saving speed where it doesn’t. Cranking effort on a simple email is overkill, but leaving it on default for a complex strategy doc leaves real quality on the table.
Why this all matters
The post’s author sums it up in a way I keep coming back to: the biggest gains aren’t from a fancier one-off prompt. They come from setting up your workspace so the smart model can actually be smart. Context, reusable Skills, letting it ask questions, handing over goals, and tuning effort. That’s the whole game.
What I appreciate about this breakdown is how practical it is. None of these fixes require you to become a prompt engineer. They’re small habit shifts that compound fast.
Try just one this week. My pick would be ending your next big prompt with “ask me everything you need to know first,” because you’ll feel the difference immediately.
Want the full context and the exact wording the creator used? Check out the original LinkedIn post for all the details.