Patrick Loeber@patloeberOctober 8, 2025
As someone who does a lot of [not just vibe] coding using voice input, I'm 100% convinced that much (maybe most) programming will be done by voice.
I still type a lot, too. But once you get used to voice mode and change your workflow to be effective with voice, it feels scratchy going back to typing.
A few notes.
- My setup is often 1) a clean-up and processing voice loop, 2) with a fair amount of LLM prompting and iteration before I send the text into the coding environment. As you see in the video @patloeber reposted below, a little bit of cleanup is helpful. I would argue that *a lot* of cleanup and contextual rewriting is useful right now, too. LLMs are really flexible, but also giving them the right tokens really matters. Obviously this should just be subsumed into the coding environment tooling if voice input is going to have first-class support.
- The change in workflow feels like being in "planning planning" mode. One effective way to use LLM coding environments is to go back and forth with the LLM to create a planning document, and then have the LLM execute that doc step by step. In voice mode, I'm often doing a few rounds of refining each individual thing that I "say" to the coding environment while creating the planning doc.
- You really want everything to happen quickly with voice. Latency breaks your sense of flow.
- Voice conversational latency is largely a solved problem. (See, eg @pipecat_ai.) The bottle-neck here is often waiting for the coding LLM to figure out what it wants to do. This is the same with voice as with typing. But you feel this even more with voice, partly because talking is so fast and easy.
- Sometimes I go back to typing because I know the coding LLM is going to be slow enough that I don't gain much by talking. As coding LLMs get faster, voice input will become more and more useful.
- I love TUIs, but some of the experiments we're doing with other kinds of voice-driven interfaces feel like they would be transformative for coding environments. It's so early, I don't have a clear sense of this yet. But imagine saying "show me that call chain" and getting a graphical, annotated stack trace that combines the LLMs understanding of the code plus what it can see from the logs of the last few test runs.
Patrick Loeber@patloebervibe coding with voice input is now supported in ai studio🔥