🎙️ Communication & dictation

In the previous track, you saw that AI "ate" the work tactical (writing the code) and that its value shifted to the side strategic (deciding what and why). But deciding isn't enough — you need to transmit the decision. And here's the key insight from this module: how you communicate your intent to the agent has become the skill that most separates those who take off from those who get stuck. Pocock calls this, plainly, communication overpowered — "ridiculously overpowered in the world of development."

Put the vision into words is different from giving an order. “Do the login” is a vague order. “I want an email and password login that reuses our authentication pattern, with friendly error messages and test coverage” is a verbalized vision: includes what, why, and the done criteria. The agent is only as good as how clearly you express yourself. The reason of this is simple: AI can’t read your mind—it only has the words you gave it. The clearer the vision in your words, the less AI has to guess—and guessing is exactly where it goes wrong.

The average person types ~40 words per minute and speaks ~130. That’s more than 3× faster. When your work stopped being typing code and became describe intent for the agent, this speed difference translates directly into a productivity difference. That’s why Pocock is emphatic: "Anyone who's not doing dictation is just so much slower." It’s not motivational exaggeration — it’s the math of bandwidth.

Dictation is exactly this: you speak and the computer transcribes your prompt in real time, in any text field. Instead of typing three paragraphs of context, you speak three paragraphs in a third of the time—and still share details you’d never have the patience to type. The reason is important: the bottleneck in working with AI is no longer how quickly AI can write (it’s instant); the bottleneck has become you being able to express what you want. Dictating breaks this bottleneck.

Pocock has a phrase that sums up the whole game: what matters is "how fast you can output tokens from your brain and input them back into your brain." In other words: the speed at which you can remove ideas from your head in the form of tokens (words) and then return to your mind the answers the AI produced. Working with an agent is this cycle in motion: brain → tokens → agent → tokens → brain, repeatedly.

The two sides of the cycle have different bottlenecks. On the output, the limit is how quickly you can speak—and that’s where dictation helps (3× more words per minute). On the input, the limit is how quickly you read and absorb the response—and that’s where speed reading, well-formatted diffs, and even listening to a summary come in. The reason to think about both sides: there’s no point in speaking quickly if you spend 10 minutes reading each response. The goal is to widen the pipe in both directions.

Here's the central thesis, and it's provocative: communicating well is "ridiculously overpowered" in the world of software development—in Pocock’s words. For decades, “communication” was treated as a soft skill, secondary to technical talent. With agents, it became a out-of-the-box hard skill: it’s the channel through which all your strategy gets carried out. If you’re clear, the agent gets it right; if you’re unclear, no model in the world can save you.

The good news: communication is a skill, not a gift—you can train and improve it. And it has three practical muscles: (1) sharpness — say what, why, and the completion criteria; (2) structure — order your thinking (context → objective → constraints) instead of dumping everything together; (3) brevity without losing meaning — cut what doesn’t change the decision; keep what does. The common mistake is confusing “being technical” with “being clear”: dense jargon impresses people, but confuses the agent just as much as it would confuse a junior.

The tool Pocock uses is Whisper Flow. The name comes from Whisper, OpenAI’s speech recognition model—extremely good at transcription, including accents and technical terms. “Flow” is the app that plugs this engine into your whole system: you trigger a shortcut, speak, and the text appears wherever your cursor is. Whether it’s the Claude Code terminal, an email field, or the editor—it works anywhere.

There are alternatives (the operating system’s own built-in dictation, or other apps based on Whisper), but the point isn’t the brand — it’s adopt the habit. O reason a dedicated tool instead of native dictation: the good ones have low latency, automatic punctuation, formatting, and work globally (in any app), which makes the difference between “use it once in a while” and “use it all day.” The common mistake is installing it, trying it once, finding it strange, and giving up—the learning curve takes a few days; after that, there’s no going back.

Wrapping up the module: overpowered communication doesn’t come out of nowhere—it’s a built habit. At first, talking to the computer feels strange (everyone goes through this), and you’ll want to go back to the keyboard. Stick with it for a week. The goal isn’t to speak eloquently; it’s to speak structured and clear automatically, the way you'd explain it to a capable junior. Below is a mental checklist—a "human skill" to follow until the habit comes naturally. Copy and paste it onto a sticky note next to your screen:

TREINO DE COMUNICAÇÃO OVERPOWERED — 1 semana

1. INSTALE um app de dictation (ex.: Whisper Flow) e defina um atalho global.
2. REGRA: por 1 semana, todo prompt longo é DITADO, não digitado. Sem exceção.
3. ESTRUTURA ao falar (sempre nesta ordem):
   - CONTEXTO: onde estamos / o que já existe.
   - OBJETIVO: o que quero alcançar (a VISÃO, não só a ordem).
   - PORQUÊ: por que isso importa.
   - LIMITES: restrições, padrão a seguir, o que NÃO fazer.
   - PRONTO: como sei que terminou (teste / critério).
4. REVISE a transcrição 1x antes de enviar (pega erro de reconhecimento).
5. LOOP: feche o ciclo lendo o diff/resposta rápido; não releia tudo devagar.

Meta: falar nitido e estruturado vira automatico. Ai voce esta "so much faster".