For most of the history of software, making something meant writing code. Then AI moved us one step up: we stopped writing code and started writing prompts.

The next step is already here, and it's the most natural one of all. We don't even need to type. We can just speak.

In an AI and agentic world, voice isn't a convenience feature. It's the interface that removes the last barrier between an idea and the thing that gets built.

This site is the proof

The website you're reading was built without me writing a single line of code. That's true literally: AI wrote all of it.

But it goes further. I didn't type the prompts either. The essays, the page structure, the design direction, the fixes, the domain and hosting setup: almost all of it was directed by speaking to Claude, in rough, conversational sentences, the way I'd brief a colleague.

I've led product development for 25 years. This is the first project where the gap between what I said and what got built was a single conversation.

Why speaking changes the work

It's faster than typing. Most of us speak several times faster than we type. The idea gets out before it fades.

You think out loud. Typing makes you edit as you go. Speaking lets you ramble, change your mind mid-sentence, and add context you'd never bother to type. That extra context is exactly what makes AI output better.

Rough is enough. My spoken prompts were full of half-sentences and transcription errors. It didn't matter. The AI understood the intent, and intent is what counts.

It removes the last skill barrier. First AI removed the need to code. Voice removes the need to sit at a keyboard and craft a prompt. Anyone who can explain what they want can build.

What this means for those of us who build products

In Nobody wants an agent, I argued that users just want to talk to a box and have the work done. I chose the word talk deliberately. Voice is how that becomes real.

Design for spoken intent, not typed commands. People speak in context, with hesitations and corrections. Products need to understand meaning, not parse syntax.

Make the conversation the workspace. When people build by speaking, the product has to show progress, ask the right follow-up question, and let them redirect mid-flow.

Treat voice as primary, not a bolt-on. A microphone icon next to a text box isn't a voice product. Voice-first means the whole experience works when nobody types.

The next generation of software won't be typed into existence. It will be talked into existence. This site was.