What Happened to My Thinking When I Stopped Typing I checked recently. It turns out I now spend more time talking out loud to an AI than I do to my own wife. Or my colleagues. This is really interesting, and here’s why. Not because of what it says about me. Because of what it did to the quality of the ideas coming out the other end. My hypothesis, arrived at somewhere around the four hundredth conversation, is this: If you’re using AI for anything that needs creative thinking, you’re at a real disadvantage if you type. That covers more ground than it sounds like it does. It’s obviously true when you’re hunting for an idea. But it’s just as true when you’re executing one. Because execution is never a single fixed thing. You want variations. You want the second option, and the fourth, and the one you hadn’t thought to ask for. Talking gets you all of them. And it gets you them sharper. I use Wispr Flow. It sits in the background until you double-tap a key. Then you talk, and words appear. I’m using it now. The shift wasn’t “this is a faster keyboard.” It was “oh, I’ve been thinking with the handbrake on for twenty years.” Start with the mechanics. Every keystroke is a micro task-switch. You’re spending attention on the physical act of finding letters. And unless you’re an exceptional typist, you’ll make typos. Each one is a small handbrake pull on the thought you were halfway through having. Cognitive scientists have a name for this. They call it extraneous load. It’s the brainpower you spend turning a thought you’ve already had into finger movements. Note that bit. Already had. The thinking is done. Everything after it is admin. And you’re paying for that admin out of the same limited mental budget you needed for the thinking. It’s a tax on the privilege of using a nineteenth-century machine. One whose keys were deliberately scattered so the typebars wouldn’t jam. We are all, still, working around a mechanical problem that was solved in 1873. The speed gap is embarrassing too. Most people speak at 130 to 150 words a minute. We type at about 40. Stanford put numbers on it in 2016. Speech recognition was three times faster than the iOS keyboard. And it produced 20% fewer errors. That study is now a decade old. The models have improved by an order of magnitude since. But speed is the boring bit. The interesting bit is the editor in your head. When you type, you’re more deliberate. That sounds like a virtue. For a contract, it is. For finding an idea, it’s poison. Typing is expensive. So you edit before you commit. Which means you arrive at the prompt having already decided what you think. Which rather defeats the object of asking. Think about why water cooler serendipity works. It isn’t the coffee. It’s that someone says something apparently irrelevant. And the irrelevance is the entire point. The breakthrough hides in the tangent. But nobody types the tangent. Speak it, though, and it survives. It lands in the transcript. And a decent model will pick up the inkling you hadn’t noticed you’d had, then hand it back to you as a solution. I’ve lost count of the times the useful output came from the throwaway aside. Not from the carefully constructed question. There’s research that backs this up, and I find it quietly thrilling. Psychologists have spent years giving people fiendish puzzles, then watching whether talking helps them crack it. The answer turns out to depend entirely on whether anyone is listening. One study is gloriously titled “Verbalization Toward Others Facilitates Insight Problem Solving”. It found the benefit shows up only when you speak to another person. Mutter the same words to yourself and you get nothing. Say them to someone and things start to give. Talking at a wall is just talking. Talking to somebody is thinking. And a voice conversation with an AI, whatever else it is, is definitely somebody listening. Then there’s the unglamorous argument. The one nobody puts on a keynote slide. My wrists. I used to have genuinely bad ones. Now I don’t, because I talk. Around 450,000 UK workers have upper limb RSI. It costs British industry somewhere between £5bn and £20bn a year. And the fix turns out to be a microphone. Which is why I’d wager keyboards will be largely relegated sometime in the next ten years. Not gone. Relegated. The way handwriting was. Still there, still occasionally lovely, no longer the default. The direction of travel isn’t subtle. WhatsApp handles roughly 7 billion voice messages a day. And a 2024 Uswitch survey found nearly a quarter of Gen Z say they never answer phone calls. 61% prefer text in almost every scenario. Which is a wonderful paradox. They’ll happily send you a four-minute voice note. But they won’t take a thirty-second call. Because a voice note is speech with the obligation removed. All the bandwidth of talking. None of the tyranny of listening back in real time. And that, in the end, is why AI is the perfect recipient. It’s infinitely patient. It never sighs. It never checks its watch while you circle the point for ninety seconds. It is the ideal audience for a half-formed thought. I speak to Claude more than to my wife. Not because Claude is better company. Because Claude never needs me to get to the point. Which is, I suspect, exactly where the good ideas were hiding all along.