I thought I knew my way around Claude. Then I stumbled onto a post that made me feel like I’d been driving a race car in first gear the whole time. The original poster laid out 26 hacks, and by hack number 8 they warned that almost nobody keeps reading. I read every single one, and I’m still thinking about a few of them.
What I love about this list is that the creator isn’t precious about being right. One point is literally them deleting their own past advice. That kind of honesty is rare, so I wanted to break down what this AI professional shared and explain why each hack actually matters for how you work day to day.
Here’s the full rundown, cleaned up and explained.
The 26 Claude hacks, unpacked
- Use the desktop app. The expert says the browser is just the demo. The app is where a feature called Cowork actually lives.
- Use Cowork, not the chatbot. If you’re still typing into a plain chat box, you’re using it like it’s an old habit. Cowork is the upgrade.
- Forget .md files and folders. This is the point where the author deletes their own past advice. Folders leak old outputs into your context, and let’s be honest, nobody maintains those files every week.
- Skills and Projects are the whole system. The creator boils it down beautifully: a skill is a capability you teach once, and a project is a place you go back to.
- Rebuild your about-me as a skill. Instead of a static file, the original poster turns personal context into a skill so it follows you into every chat automatically. Same info, but it actually shows up when you need it.
- Turn every setting off. Empty your global instructions. Leave Connectors on. That’s the whole config, according to this contributor.
- Projects remember between chats. And you can share one with your entire team in two clicks, which makes collaboration painless.
- Long chats get dumber. This is the hack where most people quit reading. The longer your conversation runs, the worse the output gets. The mind behind this post points out that Anthropic’s own engineers have said it out loud.
- Restart instead of correcting. Stop typing “no, I meant.” Click “Restart the conversation from here” on an earlier message and start clean from that point.
- Every word costs money. A token is roughly one word, and each word you send and receive adds up. The savvy professional gives a wild example: message 30 costs you 31 times what message 1 did.
- Point your sharpest model at your hardest problem. The author suggests Fable 5 for your single toughest challenge, two turns, then switch models. Extra turns are the most expensive habit you have.
- Opus 5 High for everything else. Note the word High, not Max. The expert is specific about this for a reason: you rarely need the heaviest setting.
- Give the goal, not the steps. Handing Claude a rigid list of steps drags it down. Give it the outcome you want and get out of the way.
- Never paste secrets. No logins, passwords, or keys. Ever. Simple, non-negotiable.
- Skills fire themselves. You don’t prompt a skill. You type a slash command and it runs.
- It sounds sure even when it’s wrong. Confidence is not accuracy. The creator flags this so you don’t get fooled by a convincing tone.
- Don’t trust the agreement either. This LinkedIn creator calls Claude a sycophant. Pleasing you and being right are two different jobs.
- Use Research mode. It’s the “+” in the chat box, and the original poster says almost nobody knows it’s there. Save it for something genuinely hard.
- Add “use AskUserQuestion” to your prompt. Claude then interviews you before it answers. As the innovator puts it, it prompts itself better than you ever will.
- Talk, don’t type. Cowork reads 100,000 words in about 15 seconds, then waits on you at 60 words per minute. You speak at 150, so use your voice.
- Ask for five very different versions. Then pick one. That final choice is taste, and taste is the part the tool can’t do for you.
- Make it audit its own answer. Have Claude check its own work before you trust a single line of it.
- Too much context backfires. Overload it and it loops, giving the same answers with zero creativity. Keep context in the project and the chat clean.
- The first draft is yours to fix. Not to ship. The person who shared this is firm on that one.
- Outsource the thinking, never the understanding. This is the rule the author says matters most, and it’s the perfect note to end on.
Why this hits so hard: most of these aren’t about clever prompts. They’re about treating Claude like a fast, overconfident collaborator that needs a clean workspace and a human with taste in the loop. Keep the context tight, stay skeptical, and own the final call.
Where to start today
If all 26 feel like a lot, I’d pick three to try first. Move to the desktop app so Cowork is in reach. Restart long chats instead of arguing with them. And always make Claude audit its own answer before you trust it. Those three alone changed how I think about a normal work session.
The original poster packed a ton of hard-won detail into this one, and there’s nuance in the full breakdown worth reading for yourself. Check out the complete LinkedIn post to see every hack in the creator’s own words.