Talk out your context, capture it, and let AI shape it to your format.

In the early 2000s, I was the vice president of developer relations at Cloud.com, and the Cupertino office was so quiet you could hear a pencil drop — engineers ten feet apart, every design debate running silently over IRC.

Walk into an engineering office today and it sounds nothing like that. The Wall Street Journal calls it typing being replaced by whispering: developers murmuring to their vibe-coding agents all day. I vibe code too — and here's my correction to the eye-rolling. Talking to AI isn't just noise. Speech runs 3–4x faster than typing, so in two minutes of talking you hand a model far more of what's in your head than you'd ever type. It isn't about the words. It's about the context.

The workflow is three moves: talk it out and capture it, paste the raw transcript, then prompt with your format. Here's the prompt, or if you want, turn it into a reusable skill.md:

What changes: You stop rationing context because typing it out is tedious. A bare prompt returns the average of the internet. A prompt loaded with your voice-captured thinking returns your take, in your format — that's context engineering, and it's the step most people skip.

Where else this works: A sales rep dictating everything they know about an account before the follow-up email; a marketer talking through a campaign before the brief; a CEO narrating a board memo on the drive in. Anywhere the thinking lives in your head and the formatting is the chore.

We've talked to machines since Siri landed in 2011 — but that was dictation: timers, weather, one-line commands. A category of intelligent successors is here now, all built to turn rambling speech into clean, paste-ready text:

The capture is only half of it. Drop those transcripts into a notes app you trust — Obsidian if you want local-first Markdown, Notion if you want a database and collaborators on the same canvas — and they become a compounding context layer. One voice capture, every AI in your stack, reading from the same source. From there, the same material can feed ChatGPT, Claude, Gemini, a coding agent like Claude Code or Codex, or a custom agent that runs while you sleep — any chatbot or agent platform you already pay for.

Wispr Flow is what I run on the dog walk or a long road trip — twenty minutes of talking, the angle for an article and every half-formed argument around it, and it lands as text I can paste and shape.

The bigger thing is what opens up once you stop rationing context. You stop censoring half-formed ideas because typing them out feels like a tax. You stop settling for the generic answer because you couldn't be bothered to load the full picture. Your AI stops being a search engine for the average opinion and starts being a thinking partner for your opinion — sharper drafts, faster decisions, work that compounds because it carries your fingerprints instead of the internet's.

That's the real upgrade. Not faster output — truer output. The version of you that's been stuck in your head finally has a way out, and a collaborator on the other side that can keep up.

The tools are here. The bottleneck was never the model. It was how much of your thinking you bothered to move across.

Talk, don't type — and let the work start sounding like you again.