I talk to my AI all day. It never talks back.

TL;DR: I dictate almost everything – ideas, fixes, complaints, whole briefs – and type almost nothing. The reason isn’t convenience alone. Speaking is how I think: I ramble, the AI organizes what I said, and I read it back knowing immediately whether it’s right or what still needs refining. The AI’s job is to read. It doesn’t need to talk back. The habit runs on a small pipeline that turns phone voice memos into plain text my AI can process and file. This is the piece of my setup I call the best thing I’ve ever made.

Let me introduce myself the way I once put it in a voice memo: “I’m not good at communicating. I’m shy. I don’t practice talking about things or formulating arguments. Yet I have a lot of information in my head. What I am good at is rambling.”

That’s the honest setup for this whole post. The person who talks to a computer all day is shy. If you also walk around with a full head and a bottleneck between it and the world, this one’s for you.

Why dictate to an AI instead of typing?

Because for some of us, speaking is the only input fast enough for how thinking actually happens. The discovery that changed how I work wasn’t a feature. It was noticing what happened when I stopped composing and started rambling: “I can dictate something about a topic I know about or have feelings about. And then it organizes and I read it back and I’m like, yeah – that’s it. That’s what I think, but much clearer.” Just as often the reaction is: that’s almost it, except for this one thing that needs refining. It’s never right the first time around, and that matters – the refining rounds are part of the thinking, not overhead on it.

The important part is the order. I don’t produce the clear version and hand it over. I produce the mess, the AI proposes the clear version, and I recognize what’s right and what still needs work. I can’t reliably generate the clean statement of my own position cold – but I know it on sight, I can tell you exactly why a wrong version is wrong, and the next round usually lands. For years I treated that as a flaw, but it’s not. It’s a mode of thinking, and dictation is the interface that frees it.

Talking to it, not with it

I am talking to the AI – that part is true. I’m getting my ideas out verbally, and saying them out loud is part of what refines them; the idea gets sharper in my mouth before any software touches it. What I’m not doing is talking with it. There’s no conversation. The AI’s contribution comes after I stop: it reads what I said and makes it useful – organized, filed, findable.

That one word is why I’ve never wanted the talking-back kind of AI. Voice assistants that chat with you out loud sell the conversation as the feature, and I don’t like it – the conversation isn’t the point. The listening is.

And typing can’t substitute, at least not for me. If I had to type everything out before a thought went away, I’d capture half of my ideas at best. Speaking is (usually) faster than thoughts evaporate. Typing isn’t.

Why capture ideas as voice memos? They don’t arrive at the desk

Here’s the pattern that makes the phone’s microphone the most important key on any keyboard I own: “During the day when things are busy and I’m sitting at my desk trying to think, that’s when I often have trouble seeing things clearly. It’s later on – working in the garden, going for a walk, just lying in bed in that time in the morning when I know I should get up but I’m just kind of lying there – that’s when I have clarity.”

The desk has all the tools and none of the clarity. The clear moments happen where the only capture device is the phone in my pocket, and the only input fast enough is my voice.

And those clear moments don’t keep. I used to think, oh, I’ll remember it, but I never remember it. Later I’d sit there trying hard to pull the idea back – really trying, the way you chase a dream after waking – and there’d be nothing left to pull. So I capture it the moment it’s there, mid-walk, because twenty minutes later there’s nothing left to capture.

One habit multiplies the value of every memo: I route it while I record it. Mid-ramble I’ll say something like, I’m telling you this story and I think it ties to this topic – grab it when we write about that. The tag costs me three spoken seconds. Months later, when that topic comes up, the memo surfaces exactly where it’s needed instead of rotting in a list I’d never reread.

How do you turn voice memos into text your AI can read?

With a small automated pipeline – and mine is the best thing I’ve ever made. A voice memo on its own is a recording nobody will ever listen to again. The value shows up when it stops being audio and becomes text an AI can read – automatically, every time, without me touching anything.

So here’s the machine, with credit where it belongs: I built mine from Thomas Frank’s Notion Voice Notes automation – his code, working as shipped; I changed how the output lands.

The shape is simple. I dictate a memo using the RecUp app on my phone. It lands in a Dropbox folder. A Pipedream automation picks it up, runs it through a transcription model, and drops the result as a plain text file into a voice memo folder in my vault, where my AI reads it and files it. Built once. Runs by itself. When I reviewed my whole setup recently – every connector, every automation, everything I’d tried and dropped – this is the piece I called “the best thing I’ve ever made.” (Thanks, Thomas!)

It gets real use. The other morning, thirteen memos rode that pipeline from my brain into my vault before I even made it out of bed. And everything in it runs on free tiers, on purpose – better tools exist, but I keep the stack lean; the full money picture is in What I actually pay for AI.

What happens after a memo lands – the filing, the retrieval, the daily rhythm of a vault the AI works in – is its own story, told in AI in Obsidian: how I actually use it.

Is dictating to an AI enough on its own?

No – capture needs a reader, and that’s the honest limit. Dictation alone did not change my work, and I can prove it, because I dictated for a long time before it paid off. The memos used to feed a Notion template I’d set up with the best intentions (Thomas Frank’s Ultimate Brain), and I struggled to learn it and keep it up to date. It turns out that capture without a reader is a pile of transcripts. And the funny thing is, that Notion board had Notion AI attached – at least for a little while, until the free trial ran out. It could search my memos. It could summarize them. And it couldn’t do anything with them, because it didn’t know me – didn’t know my projects, my plans, or where anything fit. (That era, and what it quietly taught me anyway, is part of my origin story.)

What finally made the memos useful was routing. Now that they land in a dedicated folder where an AI that knows my work can read them and put them where my future self will look, the content idea lands with the content ideas, the client thought lands with the client project, the half-formed worry lands where worries get triaged. Being able to query them is nice. The routing is the amazing part – and routing is only possible for an AI that knows the places.

So the claim I can stand behind is narrower than “dictate and your life improves.” Speaking gets the ideas out whole. An AI that knows you is what makes them useful. You need both halves.

How do you start dictating to your AI? The mic key you already have

You don’t need the pipeline on day one. I didn’t have it on day one. The microphone key on your phone’s keyboard – or whatever dictation app you already use – is the entire starter version: open a note, talk the thought through, paste it at your AI, and say “organize this and tell me what I said.” Read the result back, tell it what’s almost right, and let it refine. You’ll know within one sentence whether it caught what you meant – that recognition is the skill, and you already have it.

There’s a middle step, too, long before any automation: record ordinary voice memos and drop the files somewhere your AI can transcribe and read them. The habit is the trick – make voice memos you actually do something with, and the tooling can catch up later.

If you want a shaped first exercise, my first-file template is built to be filled by talking instead of typing – it’s the same habit with training wheels.

I can’t promise dictation is how your thinking works; plenty of people compose best through their fingers, and none of us gets to argue with our own wiring. But if your head is full and your typing can’t keep up, borrow my setup’s quietest feature: an AI that listens, reads, files – and never needs to talk back.