Open any AI chat tool (voice if you have it, or just type messily). Pick one real work goal and dump a 60-90 second ramble: your goal, constraints, what good looks like, and what you've already tried. Then ask: 'Echo my intent as 5 clean bullets + 1 open question.' You're done when you've written the one correction you still needed after the echo, in 12 words or less.
How-to · INT · INT.2
Voice ramble → cleaner mind meld
When you under-specify intent, a messy long ramble (voice or typed) can give the model more bits — then use its clean echo as the working brief.
Worked example
Here's a practical pattern from Karpathy: when you're too lazy to type out your full intent, just lean back and ramble — talk it out for a few minutes, stream of consciousness, even tell the model 'speech mode, typos ok.' Models are often surprisingly good at reconstructing that tangle into a cleaner echo. That's the actual skill: more bits in leads to a better mind meld and fewer corrections later. It's not magic, though — still don't dump secrets or client PII into the ramble.