If you already knew all 26 hacks, unfollow me.
Almost nobody makes it past Claude hack 6:
(# 11 surprised even me):
1. Use the Claude Desktop app. The browser is the demo. The app is where you can use Cowork.
2. Use Cowork on the app, not the chatbot. If you're typing into a chat box, you're using it like it's 2025.
3. The longer your chat, the dumber it gets. It rots as the conversation grows. Anthropic's own prompt engineers said it.
4. So clear your chats often. A bloated chat doesn't just confuse Claude. It quietly costs you money.
5. A token is roughly one word. And every word you send & receive is money. Stop saying "how are you."
6. Here's where most people quit reading. Don't give Claude steps. Give it the goal & get out of the way. Steps drag it back down to the old Claude.
7. Hand it your hardest, longest task. The one you'd struggle with. That's where Claude actually wins.
8. Stop writing those 500-word prompts. A well-specified problem beats a long one. Say what you want, not how to do it.
9. One thing it never tells you: write positive instructions. "Do this," not "don't do that." Action verbs make it ship.
10. Use Research mode (the '+' at the bottom-left). Ask it something genuinely hard and watch.
11. Skills fire themselves. You don't prompt them. You just type /command. That's it.
12. Use "AskUserQuestion" in your prompt. Make Claude interview YOU before it answers. It prompts itself better than you ever will.
13. Set up one folder. 3 subfolders inside. That's the whole system. People overcomplicate this.
14. This is the one that got me: the about-me file is everything. Tell it who you are, what you love, what you hate. Same task, completely different answer.
✦ To build your "about me" file, copy the prompt:
✦ Go to https://t.co/psB7XxAv8w. Subscribe for free.
✦ Don't pay anything. Open my welcome email.
✦ Hit the automatic reply button inside.
✦ Open Notion link. Go to "Claude cowork"
✦ Copy the "Taste interview prompt".
15. Then trim that about-me file. A bloated file is as bad as no file.
16. Plug in Connectors. Claude reads your Gmail and Slack now. Most people never turn this on.
17. Name every output. Name the order you want them in. Vague in, vague out.
18. Thinking is hidden in "+", "thinking". For a hard task, turn it on.
19. For hardest tasks, reach for the Opus model.
20. It will sound 100% sure even when it's dead wrong. So make it audit its own answer before you trust a single line.
21. Don't trust agreement either. Claude is a sycophant. It wants to please you. That's not the same as being right.
22. The first draft is yours to fix, not to ship. Always.
23. Last one, and the only rule that matters: outsource the thinking, never the understanding.
24. Claude is built for coding & knowledge work. Not health. Know what it's for.
25. The first draft is yours to fix, not to ship. Always.
26. Last one, and the only rule that matters: outsource the thinking, never the understanding.
@ScottPresler@KenPaxtonTX Wish I could’ve made it from North Houston. I would’ve loved to meet you and Ken.
But tonight was parents cs sons soccer practice. Priorities.
But I’m so glad it was a packed house.
@ScottPresler I respect the shit out of what you are accomplishing.
Just wish I could vote for you here in TX.
If only PA had no voter ID laws…
But even then, I wouldn’t, because I literally follow the law.