Typing prompts is becoming obsolete. We tend to view prompt engineering as a rigid, text-based activity requiring precise syntax, but that perspective is rapidly becoming outdated. I just saw this incredible post from an AI professional who is proving that your voice is actually the ultimate interface for complex AI interactions. He describes a future-present scenario in Singapore where he wanders around recording voice notes, looking slightly eccentric, but actually performing high-level engineering work. This isn’t just about dictation; it’s about fundamentally changing the data we feed our models to get superior results.
⚙️ From Brain Noise to Structured Signal
The core mechanism this industry pro describes is using voice to capture ideas in their most raw and volatile form. When we sit down to type, we inevitably filter ourselves. We worry about sentence structure, spelling, and flow, which acts as a bottleneck for pure creativity. The original poster argues that voice captures the “nuances and noise” in your head that text simply misses. By recording long, unstructured rambles and feeding them into an advanced Large Language Model (LLM) like Gemini, he bypasses the friction of typing.
The AI acts as an intelligent layer between the human brain and the final output. It takes the audio transcription, which might be full of stuttering, self-correction, and non-linear thinking, and distills it into clarity. This transforms the prompt engineering process from a task of writing code-like instructions into a task of verbal strategy. You provide the context and the goal verbally, and the AI handles the syntax and formatting. This is a massive shift in efficiency, allowing for the generation of solutions, posts, and training materials while on the go.
📌 The Efficiency of Unfiltered Stream of Consciousness
The most significant insight here is how this method leverages the speed of human speech compared to typing. Most people can speak at a rate of 150 words per minute, whereas typing speed is often half that. However, the expert points out that the benefit isn’t just raw speed; it’s about cognitive load. When you allow yourself to “ramble,” you are engaging in a stream-of-consciousness workflow that is often where the best ideas hide. In a typed format, you might delete a sentence that felt irrelevant, only to realize later it was crucial context. In a voice note, you keep everything.
This creator suggests that capturing the “noise” is a feature, not a bug. By providing the AI with the full context of your thought process—including your doubts and alternative ideas mentioned during the recording—the model has significantly more data to work with. It can infer intent much more accurately than it can from a polished, two-sentence text prompt. The AI essentially becomes a partner that listens to you think out loud and then writes down what you meant to say, rather than just what you said.
📌 Multimodal Models as the New Editors
This workflow relies heavily on the capabilities of modern multimodal LLMs. The innovator behind this strategy mentions using tools like Gemini, which are designed to handle large context windows and varied input types. This is distinct from old-school dictation software that simply converted speech to text word-for-word. The difference here is the “reasoning” layer. The AI is not just transcribing; it is restructuring.
For example, you could record a five-minute clip describing a complex coding problem, explaining the error messages you see, and hypothesizing about the root cause. The AI can process that entire narrative and output a clean, step-by-step troubleshooting guide or the exact code block needed to fix it. This turns the LLM into a high-level editor and synthesizer. It democratizes prompt engineering because you no longer need to know the perfect structural keywords; you just need to be able to articulate your problem clearly in natural language.
📌 Ubiquity and the “World as Workspace”
The final major takeaway from this contributor’s post is the liberation from the desk. He mentions generating training topics and solutions while walking around Singapore. This implies a workflow where productivity is no longer tethered to a screen. By using voice as the primary input method, any environment becomes a workspace.
This is particularly powerful for capturing fleeting ideas. We have all had brilliant thoughts while driving or walking, only to forget them by the time we reach a computer. This method ensures that the moment inspiration strikes, it can be captured, processed, and turned into a tangible asset instantly. It encourages a lifestyle where “thinking outside the box” literally means getting outside of the office box and letting the environment stimulate new ideas, knowing the AI will handle the administrative burden of writing them down.
Potential Challenges and Nuances
Of course, there are social and technical hurdles to this approach. As the author humorously notes, talking to your phone in public can make you look like a “crazy man.” There is a social stigma to verbalizing complex thoughts in public spaces, not to mention serious privacy and security concerns. You probably shouldn’t be dictating sensitive company trade secrets or proprietary code logic while standing in a crowded coffee shop.
Additionally, there is a dependency on the quality of the transcription and the model’s ability to interpret specific jargon. If you are working in a highly niche field, the “noise” in your head might be misinterpreted by the AI if you aren’t careful. The output is only as good as the model’s understanding of your verbal quirks, so it requires a bit of trial and error to learn how to “speak AI” effectively.
🚀 Captain YAR’s “Ramble Refiner” Prompt
If you want to try the original creator’s method, you need a system prompt to help the AI understand your messy audio transcripts. Paste this into your LLM before pasting your transcript:
“I am going to provide you with a raw transcript of a voice note where I am brainstorming a problem. The text will be unstructured, repetitive, and colloquial. Your goal is to act as an elite editor and prompt engineer. Please analyze the transcript, extract the core objective and key details, and then output a structured, polished version of my idea. If my goal was to create a prompt, output the optimized prompt. If my goal was a LinkedIn post, draft the post.”
This is a brilliant way to future-proof your habits for 2026! check out the full post to see exactly how the expert handles his workflow.