Voice transcription is a superpower for AI-assisted coding.
Not obvious until you try it. I picked this up from AI engineers here in Chiang Mai - they were raving about it. Now I get why.
The insight
Here’s the thing about LLMs: more context = better output.
You need to dump everything related to the problem. Edge cases. Constraints. Context. The weird thing that happened yesterday that might be related.
When typing, you self-edit. You skip details that seem obvious. You write shorter prompts because typing is slow.
When speaking, you ramble. And that’s the point.
Grammar doesn’t matter - the AI figures it out. What matters is getting everything out of your head and into the prompt. The AI builds a complete picture from your stream of consciousness.
What I b…
Voice transcription is a superpower for AI-assisted coding.
Not obvious until you try it. I picked this up from AI engineers here in Chiang Mai - they were raving about it. Now I get why.
The insight
Here’s the thing about LLMs: more context = better output.
You need to dump everything related to the problem. Edge cases. Constraints. Context. The weird thing that happened yesterday that might be related.
When typing, you self-edit. You skip details that seem obvious. You write shorter prompts because typing is slow.
When speaking, you ramble. And that’s the point.
Grammar doesn’t matter - the AI figures it out. What matters is getting everything out of your head and into the prompt. The AI builds a complete picture from your stream of consciousness.
What I built
I rebuilt a voice transcription tool last week (Ito was shutting down, I needed an alternative). Two weekend evenings, 600 lines of Rust.
Now I use it constantly with Claude Code. Hold Fn, speak, paste. That’s it.
The workflow
- See a problem
- Hold key, describe everything about it out loud
- Let Claude Code work with complete context
Try it
Your prompts will get 3x longer and your outputs will get 3x better.
FnKey (what I built): github.com/evoleinik/fnkey