Agents work best with more context, voice enables that

“Fix the checkout bug”. That’s the kind of prompt agents usually get. Agents can pull logs, check deployments, read support tickets – that part is solved. What they can’t pull is what you actually want. The tradeoffs you care about, the way you want it done.

When you talk, you don’t filter as much. You ramble. You over-explain. You mention things you weren’t sure were relevant. Voice gets people to stop editing themselves before hitting enter.

I tried going voice-only for a full day (Aqua Voice, press and hold right Cmd to talk). Claude, email, all of it. One minute of talking produces way more context than I’d ever type out.

The keyboard is the bottleneck. There’s so much waiting on the other side, but you have to type to get to it. Speaking just removes that wall.

Works for me at home. But at office with people around, there’s still a real hesitation to talk it all out. That friction hasn’t gone away yet.