Most of us treat our AI tools like a new phone. We rush to fill every setting, load up context files, and stack instructions until the thing feels “personalized.” I used to think more setup meant more power. Then I saw a post that flipped that idea completely.
This LinkedIn creator shared something that stopped me mid-scroll: his entire Claude setup is almost completely empty, and he says it is the most productive version of Claude he has ever used. The kicker? He’s the same person who once championed building tons of .md files. It took him 194 days to unlearn all of it. I was genuinely surprised by how much he stripped away.
Let me break down what the author deleted, what he kept, and why the “less is more” approach actually works.
🗑️ What the author deleted
The original poster didn’t trim a setting here and there. He gutted almost the whole thing. Here’s his list, with the reasoning behind each cut:
- The folder system. He found that even when he told Claude “don’t check my folders,” it still peeked and pulled context he didn’t want in play. Deleting the folders removed the temptation entirely.
- The about-me file. Instead of a permanent .md file feeding context into every chat, he turned it into a skill. Same information, but now he invokes it manually, only when it’s actually relevant.
- Global instructions. This one is counterintuitive. The expert argues that too much standing context restricts Claude’s ability to think creatively. More background isn’t always better. Sometimes it just boxes the model in.
- Every setting he could switch off. He turned the whole menu off. The only exception is Connectors, and even then just the specific ones he leans on heavily (Gmail and Granola in his case).
- Repeated prompt templates. Rather than keeping a big prompt library sitting around, he converts recurring prompts into skills, so they’re on-demand instead of always loaded.
Why it matters: every file, folder, and global instruction is context Claude carries into every single conversation. The author’s insight is that constant context can crowd out fresh thinking. Pulling it in only when needed keeps each chat clean and sharp.
☑ What actually runs this “empty” Claude
So if almost everything is gone, what’s left? According to the person who posted it, just two things do the heavy lifting.
- Claude Skills. Think of these as “/” commands you trigger on demand. His about-me file and all his other .md files now live as skills, invoked only when they’re useful. Nothing runs in the background bloating every prompt.
- Projects. He keeps one per client, repeated task, or recurring workflow. Every chat inside a project remembers everything relevant, like past emails and campaigns, but none of it leaks into his other future chats. Clean separation, zero cross-contamination.
The one rule that ties it together
The whole system runs on a single, memorable rule from this savvy professional:
Write the same prompt twice? Make it a skill.
That’s the filter for everything. If you find yourself repeating an instruction, a bio, or a template, it shouldn’t live as permanent background context. It should become a skill you call when the moment calls for it. Simple, repeatable, and it keeps the clutter out.
How you can try this yourself
You don’t have to nuke your whole setup overnight. Here’s a practical way to test the author’s approach:
- Pick one global instruction you rely on and turn it off for a week. See if Claude’s answers get more flexible.
- Take your “about me” context and save it as a single skill you invoke manually instead of loading it everywhere.
- Spin up a dedicated Project for one client or recurring task and keep all its context inside.
- Next time you catch yourself pasting the same prompt a second time, convert it into a skill on the spot.
I think the reason this resonates is that it fights a habit most of us share: over-configuring in the name of productivity. This industry pro basically proved that a leaner setup can outperform a maxed-out one, because it lets the model breathe. His closing advice was blunt and I love it: do not overcomplicate it.
If you want the full reasoning and the exact settings the creator toggled off, go check out his original LinkedIn post. It’s a quick read that might just change how you set up your own AI workspace.