We still mostly type at the model. A whole lifetime of writing emails and Slack messages and texts has trained us to clean our thoughts up before sending them anywhere — structure it, edit it, make it presentable for whoever's on the other end. But the model isn't a person, and it turns out it handles mess better than it handles polish.
Read about an engineer who broke his arm and had to rebuild his whole workflow around voice because he physically couldn't type. Forced experiment. Turned out it wasn't just a workaround — he got more done talking than he ever had typing.
That stuck with me because it flips the assumption. We treat voice as the fallback — slower, sloppier, worse than typing, or that's the instinct anyway. But for working with AI it might genuinely be the better mode. You dump the full mess out loud — half-formed thoughts, tangents, the part where you change your mind mid-sentence — and let the model do the structuring. The messier the input, weirdly, the better the output. You're not supposed to clean it up first. That's the whole point.
So spew it out next time you sit down with one of these. Have it play the mess back to you distilled before you move forward. You save the typing time, you probably get clearer on what you actually meant, and what comes back is better because the model had more to chew on.
And I think this is just the start. Voice is one door. Meta's glasses, OpenAI's audio devices, those always-listening table pucks, humanoid robots — all of it is the same bet, embedding the model deeper into daily life in ways that have nothing to do with a keyboard. Every one of those expands how we actually talk to these things, and how much we get done alongside them.