Lately, I’ve been thinking about the difference and similarities between speaking, writing and thinking. Paul Graham wrote an article about how writing is thinking and that if you think you’re thinking without writing, you only think you’re thinking.

I haven’t known what to do with this idea for a while but think I’ve found something interesting recently.

In using LLM agents to build software, sometimes I use voice to text functionality to get thoughts out of my head as fast as possible. From a pure speed perspective, this is usually the fastest way to get the agent something to start working on. In practice, however, the transcription of what I’ve said isn’t always as exacting or precise as what I had in mind.

I think the appreciation of this freestyle nature is to appreciate what it is to think through writing.

When you speak to communicate, you’re generally trying to communicate some information. But speaking has certain constraints. While you’re speaking, you need to figure out what you’re going to say next while you’re saying the thing you’re already saying.

This is quite different from writing. Oddly enough, I never realized how different this was until I watched a bunch of people demonstrate how they use coding agents.

They would talk about their process, explain what they were going to do, and then would start typing to the agent, at which point, their verbal speed and cadence would change entirely.

They would dictate or type things like the following:

Create a React, Typescript application, using the Tailwind CSS framework …

Usually monotone. Usually slower than natural speaking speed. Usually completely different from how the presenter was previously addressing the audience.

In a presentation forum, this is highly unnatural to listen to.

It’s the conversion of the writing process into something consumable to a speaking presentation format, but it’s awkward.

Having noticed how strange this transition feels, when I do presentation coding with agents now, I tend to prefer a voice to text approach. This approach allows me to fluidly explain myself to the audience and the agent at the same time, whenever I want to, by recording myself when I am talking about what I am going to build and using that transcription as my prompt.

That’s a fun party trick, but back to writing and thinking.

Writing slows you down to a speed where you can care more about where you are going. The constraints of speaking require tradeoffs you don’t have to make when writing.

If you can’t think of quite the right word to describe what you’re thinking, you can fall back to a more simple description or a synonym. This works in conversation, but is a more lossy version of your actual thought and what you were trying to say.

Words and speech are already lossy, but when we take time to write something down, we have the best chance of using the best words we know how to communicate what we mean.

When we consider what this means in the context of writing to think, we might suggest speaking is insufficient to think because it requires mental energy to be committed to other things as well. If you’re presenting, you may be considering your cadence, your body language, your tone, your volume. You may be watching the audience and picking up on cues from them to get a sense of how they are receiving what you are saying. If you’re conversing, you’re listening to your conversation partner and picking up similar cues. In the end, what I think this means is that your mental bandwidth cannot be maximally focused on thinking when you are speaking.

There are too many other things going on. Maybe the closest thing to writing when speaking is to record yourself by yourself with no conversation partner and nothing to distract you. But even still, it’s not quite like writing. You’re looking around, absorbing other stimuli, and subconsciously incorporating it in what you say.

If you close your eyes and dictate, you might come close to thinking like writing, but you still can’t avoid the constraint of being forced to pick the next word to string it together with the last. Unless of course, you are going to pause and deliberately and carefully pick each. and. every. word. as. you. say. it. aloud. But is that really even speaking?

There is a sort of freestyle component to speaking. In our minds, we might think we generally know where we’re going (what point we’re going to make) but we don’t always know how we’re going to get there.

I’ve played with this idea a lot lately.

The next time you’re in a casual conversation, observe the topic and listen to the participants. Get a sense of what is being talked about and stay engaged.

See if you notice multiple threads of conversation. See if it seems like the people speaking are understanding each other. See if you can follow the conversation evolving from one topic to another.

Conversations are rarely planned. They emerge, evolve, and eventually fade away.

Even after an engaging and enjoyable conversation, there is a moment of palpable exhale where you realize it’s time to turn the page.

So what are we doing when we converse? Are we thinking?

If we need more of our mental bandwidth to focus fully, then maybe we’re thinking to a certain degree but at a diminished capacity.

If we’re debating, we’ll hear the argument of our opponent and then be expected to respond, advocating for our perspective over theirs. We’re thinking on our feet, but maybe not as deeply or clearly as we’re capable of doing.

We’re kind of thinking, but really we’re reacting. We’ve practiced debating and when we hear an argument we only have so much time to respond. There are lots of techniques you could use to buy time.

Take a deep breath. Pause. Vamp. Talk about something unrelated.

All to buy time. Time to think.

Because when speaking, time is the tightest constraint limiting the clear communication of your ideas.

When speaking, it’s more difficult to manage a conversation. Harder to retrace to previous topics. To prune earlier branches. With a partner it’s nearly impossible and would likely make the conversation feel quite unnatural.

There is a natural give and take of a conversation. If someone is talking about their day, it would be unbecoming to interrupt or abruptly change the topic to something else.

When speaking, you live in the flow of time. You need to speak words in an order that make sense. You can’t return to a sentence from 5 minutes ago, edit that, then reorganize it to better state your point. When speaking you kind of can, but it takes practice and you have to expend energy bringing your partner or audience with you.

I know I said that you can think when you speak but I don’t think I said that quite right. What I mean was, you can do an approximation of thinking while speaking but to truly think with clarity you must write.

If you say that, people can follow what you’re talking about, but you have to lay a trail of breadcrumbs. When writing, the thoughts are more pure and uninterrupted.

If you have the thought that leads to speaking the quote — maybe something like

My point is that writing is thinking and speaking isn’t quite thinking so I need to ensure I communicate clearly that speaking isn’t quite clear thinking

You can change your prose to reflect that. You don’t need to bring anyone along with you. You don’t have to justify your hard pivot. You can think and respond to your thought alone without the decorum or ceremony required when speaking.

To respond to this thought while speaking compared to writing requires different orders of magnitude of mental bandwidth. So much more, I could almost imagine myself letting the misunderstanding go while speaking because it would just be so much work to return and correct the record with linear conversation.

When writing, the nonlinear edits are possible and easy. The ideas exist concretely in words, to be sculpted into ideas.