Learn what AI can do
Every Short I make teaches one thing. Each one has a page here with the prompt to copy and three steps.
AI document space: stop losing AI answers in chat
One living document per project that your AI keeps updating, so answers stop getting lost in the chat.
Split your AI chat: one chat, one job
A copyable prompt that turns any side job into its own short chat, so your main chat stays clean and answers stay sharp.
Voice to-do list: stop typing, just say it
A voice prompt that turns a messy morning ramble into a sorted to-do list you copy into your notes.
Never let your AI agent hit send
Run your AI agents in a line: one collects, one drafts, one checks. The checker can stop a draft, but it can never send it. A person approves first.
Stop fixing the same AI mistake twice
One hour a week, put each AI draft next to the version a person corrected. Any fix you see twice becomes a rule, plus a test so it can't come back.
Why AI videos look like slides (and the fix)
AI videos look like slides when you prompt them scene by scene. Ask for the storyboard first: what's said, what's shown and how long, for every scene. Approve it, then build.
Let an AI agent do your copy-paste
Computer use lets an AI agent use a website the way you do: it clicks, reads the screen and decides the next step. Hand it one boring task in plain words.
Max effort isn't the best effort (the 2-step test)
Max effort can cost about 13 times more than low and still not give the best result. Run your real task at medium, then once at high, and keep high only if it's clearly better.
Your AI deletes its best lessons every 30 days
Claude Code keeps your session transcripts for 30 days by default, then deletes them. Once a week, have it list the mistakes it repeated and turn each into one line in your CLAUDE.md.
Paying double for Opus? Try this test first
Opus 5.5 costs about double what Sonnet 5.5 does, yet Anthropic's own page puts Sonnet just two points behind it on a real-work test. Ask one question before you pick.