The repeatable moves that make you look like a wizard. Most are free and work in the tools you already have.
Ask the same hard question to ChatGPT, Claude, and Gemini separately. Paste all three answers into a fourth chat: "You're the chair of a council — where do these agree, where do they really disagree, what's your synthesized verdict?"
Before it answers: "List what you're assuming and ask me anything unclear before you start." Kills confident wrong guesses.
Models don't know today's news. Feed them real results. Get a free key at exa.ai, run one search, paste the JSON into your AI, and ask for a cited brief — you just built a mini research agent:
curl -s https://api.exa.ai/search \ -H "x-api-key: YOUR_KEY" \ -H "content-type: application/json" \ -d '{"query":"your question","numResults":5}'
One AI drafts; another attacks it: "You're a harsh reviewer — find the 3 biggest flaws in this." Then feed the critique back to fix. Two passes beats one every time.
Paste a screenshot of any UI into Claude/ChatGPT or v0: "Build this as a working web page." Design in, code out.
Turn any script into a realistic voice clip for a demo, video, or app. Paste text → pick a voice → download the audio. Great for narration without recording yourself.
"Keep improving this until it passes [a clear test]. Stop after 5 tries." Let the AI iterate on its own work — with a step cap so it can't run away.
Pipe one AI's output into another as a mini-assembly line: draft → translate/critique → polish. Each step is a fresh prompt with one job. This is the seed of every agent — you already know how to build one.