Hi there!

Welcome to the 25th edition of Work in Beta.

In this edition, we explain why so many people are switching from typing to speaking when they work with AI and the real reason it makes the work better. It isn’t speed.

So, let’s dive in!

You opened AI this morning and typed a quick prompt. Two lines. Clean, to the point. But in your head there was a whole paragraph - the background, the constraint you forgot to mention, the example you didn’t bother typing out. AI answered the two lines. It never saw the paragraph.

That gap between the two lines you typed and the paragraph you were actually thinking is the real story behind everyone suddenly talking to their AI instead of typing.

Most people explain the shift as a speed story. You speak at around 150 words a minute and type maybe 40, so talking is roughly three times faster. Stanford found exactly that in a controlled test: 153 words a minute by voice, 52 by keyboard. True. But speed is the shallow reason. The deeper one is what you leave out when you type.

Here’s what happens when you type a prompt. Typing feels expensive, so you trim. You drop the background. You skip the caveat. You leave out the “oh, and also” that would have changed the whole answer. You hand AI a thin version of a rich thought and then wonder why the reply feels generic.

We call this the Compression Tax. Every time you type, you pay it: you compress what you know down to what you can be bothered to type, and AI only ever sees the compressed version.

There’s a deeper reason typing is expensive. When you type, part of you is writing for an audience. You chase the clean sentence, the right word, the correct grammar - even when the only reader is an AI that does not care how polished you sound. So you slow down. You self-edit. And half the thinking never makes it onto the page, because it didn’t come out in a tidy sentence.

Speaking removes both problems at once. You stop performing and just keep going. The background, the sequence, the constraints, the messy “here’s what I actually mean” - it all comes out, exactly as it is in your head. And here is the part that surprises people: that mess is not a problem for AI. It is the high-quality input AI wants.

Typing compresses your thinking. Speaking expands it. For AI, richer input beats cleaner grammar every time.

This isn’t just our opinion. The people building these models keep pointing at the same thing: what AI gives you depends on what it sees. OpenAI’s own guidance tells people to feed the model the background and the relevant details, not just the request. Anthropic calls the discipline “context engineering.” Strip the jargon and it means one thing - AI is only as good as the material you hand it. Speaking hands it more material, with far less effort.

Picture a normal task: you need to push a deadline with a client.

Typed, your prompt looks like this:

AI hands back a competent, generic email. Apologetic. Vague on the reason. It guesses at the tone, because you didn’t give it one.

Now the same request, spoken - out loud, exactly as it falls out of your head:

Same task. Look at everything the second version carried - even with all the umms and ahhs: the reason, the history, the relationship, the tone, the exact date, the spin. AI now writes an email that fits this situation, not a generic one. The filler didn’t matter. AI ignores it and keeps the meaning.

Here’s the part that matters. You could have typed all of that. You just never would. It’s too much effort for a prompt, so you’d trim it to one line and accept the generic answer. Speaking, it came out without trying. That’s the Compression Tax, paid in full and it’s how we work now. We barely type anymore. The thinking behind most of what we make, this edition included, starts as us talking.

The email is one case. The pattern underneath it is always the same: speak the mess, then let AI structure it. You’re not talking instead of typing everywhere - you talk when you’re thinking, then hand the raw material to AI to clean up.

It works best when you’ve got a lot in your head and just need to get it out:

Basically, any time the job is to get everything out first and tidy it up later.

Pick one task this week you’d normally type a prompt for - an email, a brief, a plan. Don’t type it. Talk it through instead: out loud into any dictation tool, everything you know about it, in no particular order. Then hand that to AI and let it shape the result.

You’ll notice two things. How much more came out of you than you’d ever have typed. And how much better AI’s answer is when it has the whole picture instead of one trimmed line.

The shift isn’t from keyboard to microphone. It’s from compressing your thinking to expanding it. Speaking just lowers the cost of handing AI the full picture.

One honest catch: compression isn’t always the enemy. When you need precision - exact numbers, a tight instruction, the final wording of something you’ll send untouched - typing wins, because trimming is the point. But when the goal is to get everything out of your head and into AI, stop typing. Talk.

Most professionals know AI can do more for them. The gap isn’t awareness - it’s knowing where to start, what to change, and how to make it stick.

That’s what we work on through Work in Beta.

If any of this resonates, email us at [email protected] / [email protected] and we will figure out how to work together.