I spent months believing the same thing everyone repeats: a smarter AI is a fully loaded AI. More files, more context, more custom instructions, more folders. Then I ran into a post that flipped that belief on its head, and I have not stopped thinking about it since.
The original poster is an AI professional who admits he used to be the exact opposite. His own words: “Yes, I’m the guy who asked to build .md files.” After 194 days of using Claude every single day, this creator stripped his settings down to almost nothing. His verdict? It’s the most productive Claude he has ever used.
I love a good myth-buster, and this one takes aim at the loudest advice in the AI world. Let’s walk through the misconceptions this expert is calling out, and the two things he actually kept.
Myth 1: More context always means better answers
This is the big one. Everyone says “give AI more context.” The author’s finding is more nuanced. Too much context, and Claude gives you the same answer every single time.
His take is sharp: heavy context works great for contracts, where you want consistency and zero surprises. But for ideas, it becomes a ceiling. You keep asking for something fresh and Claude keeps handing you a slightly reworded version of the last thing. The fix isn’t zero context. It’s context you control, invoked only when the task actually needs it.
Myth 2: A folder and output system keeps Claude organized
Sounds logical, right? Keep your outputs tidy so the AI can reference them. The original poster deleted the whole folder system, and here’s why it mattered to him.
Even when he explicitly prompted “don’t check my output folder,” Claude still peeked at it. Worse, it poisoned fresh answers with outdated work. Old drafts leaked into new requests. The organization he built was quietly sabotaging his results. I found this genuinely eye-opening, because it’s the kind of problem you’d never blame the folder for.
Myth 3: You need an always-on “about me” file
The common belief is that Claude should always know who you are, so you load a permanent about-me file. This industry pro made one clean swap instead.
He turned the about-me file into a skill. Same information, same context, but now it only fires when he calls it. No background noise, no weekly upkeep, no stale details bleeding into unrelated chats. The context is there the moment he needs it, and invisible the rest of the time.
Myth 4: Global instructions make Claude smarter everywhere
Global instructions feel like a shortcut to consistency. The creator switched them off, along with nearly every other setting in the menu. The whole menu, off.
The one exception he kept on was Connectors, specifically the Gmail and Granola connectors he actually uses day to day. Everything else that could be toggled off, was. His point: a setting you’re not deliberately using is a setting that’s shaping your answers without your permission.
Myth 5: Reusable prompt templates are the smart move
We’ve all got a notes file full of copy-paste prompts. The author argues that’s the slow way. If you keep pasting the same prompt, that’s a signal, not a habit to be proud of.
His rule is simple and I’ve adopted it already:
Write the same prompt twice? Make it a skill.
Instead of hunting for the right template, he turns the prompt into a callable skill. He mentions a couple of ways to do it, including a skill-creator command, or literally typing “make it a skill” inside any chat.
So what actually runs this “empty” Claude?
After all that deleting, the mind behind this system runs on just two features. That’s the whole thing.
- Skills: commands you call with a “/”. This is where his about-me file now lives, same info, invoked only when needed. Repeated prompts become skills too, so his most-used moves are one keystroke away.
- Projects: one per client. Every chat inside a project remembers everything, the emails, the past campaigns, the history. And crucially, nothing leaks out into the rest of his Claude chats. Clean separation, zero cross-contamination.
That’s it. Empty settings. Two features. A system so lean it looks broken until you understand the logic behind it.
The truth to act on
Here’s what I took away from this contributor’s 194-day experiment. The goal was never a maximally configured AI. It was a controllable one. Every deleted folder, file, and setting was a source of hidden interference he chose to remove.
If you want to try it, start small. Turn off one setting you never consciously use. Move your about-me info into a skill you call on demand. Give each client or project its own space so context stays where it belongs. Then notice whether your answers get fresher instead of flatter.
I was genuinely rethinking my own setup by the end of this. If you run on AI daily, the full breakdown from this savvy professional is worth a read. Check out the original LinkedIn post for the complete rundown, and see which of your own “best practices” might actually be holding you back.