Talk to your agent 4x faster than you type

With AI you're going to be typing a lot, and the more detailed you are, the better your results will be.

This is a reversal you need to adapt to. Before AI, the less detail you gave Google the better, as you didn't want to be overly specific or you'd get nothing back. With an agent it's the opposite.

The problem is that the level of detail required to take an app from start to finish is basically book length. Typing all of that out manually is prohibitive.

So what you want is voice transcription, where you speak into your computer's microphone and just talk. Just yap for minutes on end and hand that to the AI. Most people type around 40 words per minute and speak closer to 150, so you're getting three or four times the detail for the same effort.

You might be wondering why you can't just use the dictation already built into your computer. You can, but those use older technologies that generally haven't been adapted to AI yet. Apple's built-in dictation also cuts out after roughly thirty seconds of silence and tends to drop words in longer sessions.

The bigger issue is that the older engines transcribe sounds rather than meaning, so they'll cheerfully write "react rooter" and "sue pa base" and leave you having to clean up a mess which slows you down. The newer tools run the audio through an actual language model, which is why they get technical words right, and why they'll drop your ums and false starts instead of typing them out.

There are a lot of dedicated tools that have come up recently. I use Wispr Flow and I highly recommend it. It's very easy to set up. Any time you want to start yapping you press the Fn button on your MacBook, you talk, and you release the button. If you want to talk longer and pace around, press Fn followed by space, and now you can release the Fn key and it will keep recording. Once you're done, everything you said will appear in whatever text box is currently focused. It's very accurate and it uses the latest AI models to transcribe.

Because it types into whatever's focused, it works in your terminal, in your browser, in your notes app, etc. There's nothing you need to integrate or configure.

It's called Wispr Flow because you can talk really low, and you can even whisper, so it's useful in a quiet environment like a library or when you're out and about. You don't have to talk loudly for it to work.

The free plan gives you a couple thousand words a week on desktop, which is enough to see whether this fits how you work but not really enough to run on. The paid plan is around $12 to $15 a month. There are free alternatives if you'd rather not pay, and your AI chat assistant can point you at the current ones, but I'd honestly try the free tier for a week before deciding either way.

Set this up as early as you can, because if you resort to typing, it could honestly damage the quality of your work. You'll be reluctant to type so much, and your fingers will get tired in the middle of a prompt. You'll think forget it, this is good enough, I'll just send what I have. Now the AI is off working on something that is half-baked.

Protip: no need to speak super cleanly or in finished prose. Just ramble, repeat yourself, contradict yourself, and let the agent sort it all out, since it's very good at pulling the intent out of a messy paragraph. Better that you say even a couple of unfinished or poorly worded thoughts on something than having the agent guess at what you meant.

This is going to be useful for a few things in this course. The genesis rant in the first class is a voice-based exercise, where you will talk for ten minutes about why your product should exist and then hand the transcript to an AI. And when you get to writing specs for features, the difference between a two-sentence request and a five-minute explanation of what you actually want is going to make a world of difference in the results you get.

Most agentic engineers eventually arrive at this point on their own, where they think you know what, I wish I could just use my voice. And then they do. That's where most of us are right now, and I recommend you get here quickly as well.