Last week I got access to a Codex. One of the features I was most curious to try out was its voice chat mode. At first, I didn’t understand why some people were so hyped about it, aside from the cool demo shown at Astra’s release.
Voice as a presentation coach
With a month to prepare and a clear idea of what I wanted to show, I quickly built a demo using Kiro about decision model demo. The next step was to rehearse it. The issue is that I’m a productive procrastinator. The result was that I left it until the last two days. With the deadline approaching, I remembered codex’s feature. Opened my demo folder, clicked on voice chat and gave it this prompt:
I have to present the demo, set a timer, listen to me present and feel free to give me notes
And that was it. I suddenly had a calm and polite companion listening to me, offering an occasional “hmm” and transcribing what I said. After about 20 minutes, I got a really useful report on what ChatGPT suggested I could improve.
After another 20 minutes, I was able to get into a flow and focus on a single task. I went on to deliver a presentation to 50 people that received good feedback.
Why this worked for me
I had previously tried speech-to-text, but I think voice chat works so well in Codex because it packages all the following features together:
- It can interrupt me or let me interrupt it at the right time (especially with OpenAI’s 6.1 models).
- It works in real time, both on desktop and in the browser.
- The voice is pleasant (after configuring a tone I liked).
- It can understand my mumbling and turn it into a coherent request.
- It provides a concise but useful spoken response alongside a complete written version.
For me, these features take the experience from unusable to pleasant.
What now?
There’s clearly a lot going on here, and I expect to start typing less from now on. During coding sessions, I still type some of my prompts. But I now find myself automatically reaching for voice chat when:
- I need to check that I understand a concept or problem
- I want to verbalize my random thoughts and have the AI help turn them into something coherent (rubber duck)
- I want to rehearse a presentation
- I’m tired and keep losing focus
Photo: Unsplash.