Your keyboard is likely the single biggest bottleneck restricting your creativity right now. I realized this after reading a fascinating perspective from a LinkedIn user who is completely rethinking how we interface with artificial intelligence. This innovator describes wandering through Singapore, talking animatedly into his phone, looking like a “crazy man in black” to bystanders. But he isn’t on a call; he is performing complex prompt engineering through voice notes to supercharge his workflow.
🎙️ The Mechanics of Audio Prompting
The core concept this expert proposes is a shift from “text-perfect” prompting to “stream-of-consciousness” engineering. The author explains that voice is the most efficient medium for capturing ideas in their rawest, most truthful form. When we sit down to type, we instinctively edit our thoughts, filtering out details to be concise. However, this savvy professional argues that this filtering process actually hurts the AI’s ability to understand deep context. By using a Large Language Model like Gemini, the creator records long, unstructured audio files containing ideas, immediate self-corrections, and specific nuances. The AI then acts as a synthesis engine. It takes that chaotic audio—the “noise” in your head—and structures it into high-quality prompts, solutions, or content. This shifts the cognitive load: you provide the raw ore of intent, and the AI handles the refinement process.
💡 Capturing the Nuance in the Noise
The most profound takeaway from this post is the value of unfiltered data. The original poster highlights that raw audio captures the nuances that often get lost when we try to be formal in writing. When you speak to an AI, you might say something like, “I need a strategy for this product… actually, wait, it’s more of a service, so focus on the experience rather than the specs.” In a written prompt, you would likely delete the first part. However, that correction tells the AI exactly where the boundary lies between product and service in your mind. This expert leverages that stream of consciousness to give the LLM a richer dataset of intent. By allowing the “noise” to exist, the author ensures the final output aligns far more closely with his specific vision than a sterile text command ever could.
🚀 Accelerating Production Velocity
Speed is the ultimate advantage in the workflow this industry pro describes. He notes that he uses this method to generate everything from social media posts to complex training topics instantly. By removing the friction of typing, the gap between having a spontaneous idea and executing it effectively vanishes. If you are walking to get coffee and have a sudden breakthrough regarding a project, you do not need to hope you remember it until you get back to your laptop. You simply tell the AI immediately. This creator is essentially using the AI as an active listener and a secretary simultaneously. The result is that he can double down on productivity in the coming years without spending more hours at a desk.
📌 The Evolution of Multimodal Workflows
We are rapidly moving past the need for complex, text-only tech stacks. The LinkedIn user specifically mentions using Gemini, seemingly leveraging its multimodal capabilities to process information. This indicates a future where we stop treating LLMs like command lines and start treating them like colleagues. The author’s approach proves that you don’t need to learn complex syntax to be a prompt engineer; you just need to be able to articulate your thoughts clearly—or even unclearly—and let the model parse the logic. This is a massive shift in accessibility. It suggests that in the near future, the best prompt engineers won’t be the fastest typists, but the most articulate speakers who can verbally guide a model through a complex problem space.
⚠️ Potential Social and Technical Friction
While the benefits are high, this method does come with a few humorous and practical challenges. As the author amusingly points out, walking around talking to a machine makes you look a bit unhinged to the outside world. There is a genuine social barrier to dictating complex instructions in public spaces. Beyond the social aspect, relying on voice requires a model that handles transcription errors gracefully. If the AI mishears a specific technical acronym, it could derail the output. You have to be comfortable with a workflow that involves reviewing the AI’s interpretation of your audio before giving the final sign-off.
🛠️ The “Ramble Refiner” Prompt
To help you apply this expert’s strategy, I have drafted a system prompt you can use. Paste this into your LLM, then hit the microphone button and start talking.
Copy and paste this prompt:
I am going to provide you with a raw, unstructured voice transcription of my thoughts. This input will contain rambles, self-corrections, stammering, and stream-of-consciousness ideas. Your goal is to act as my Editor and Prompt Engineer.
- Analyze the transcription to understand my core intent, audience, and goal.
- Ignore filler words or corrected mistakes (e.g., if I say ‘Plan A… no, actually Plan B’, focus only on Plan B).
- Synthesize this information into a structured, high-quality prompt that I can run to achieve the desired result.
- Finally, execute that prompt and provide the output I was looking for.
This insight from the post’s author is a brilliant reminder that sometimes the best way to move forward is to stop typing and start talking!
Check out the full post for more context.