AI workflows
What is meta-prompting, and why does it improve AI output?
Short answer
Meta-prompting means never prompting AI directly for anything serious. Ask the AI to write the prompt first, then run that prompt. It's my core framework for getting high-quality output, and it works best paired with voice dictation.
Most people type a one-line request and get a one-line-quality answer. The AI is a better prompt writer than you are. Use it.
How I do it
- Brain dump the task out loud: what I want, who it's for, what good looks like.
- Ask the AI to turn that into the best possible prompt.
- Run that prompt, in a fresh chat if it's a big job.
The shape of step two, ready to copy:
Here's what I'm trying to do: [brain dump].
Before doing it, write the best possible prompt for this task.
Include the context, the exact output I want back, and what to avoid.
Ask me anything you need to know first.Talk, don't type
Voice dictation is the one tool I'll never stop using. I use Wispr Flow. Talking strips out the self-censoring you do when you type, so the AI gets more raw material and writes a better prompt. I don't type more than a sentence anymore, and I code entirely with my voice.
Two related rules
- Ignore AI time estimates. When Codex or Claude says a build takes 20 hours, that's human developer hours. Just run it.
- Pick one platform for your team and stay on it. Consistency beats switching to whichever model is ahead this month.
Answered by TJ Larkin. From AI training and partner calls in May 2026. Last updated .