The keyboard is effectively a bottleneck for human creativity. We often struggle to get ideas out of our heads and onto the screen fast enough to keep up with our racing thoughts, resulting in lost details and diluted concepts. I recently discovered a fascinating post from a forward-thinking AI enthusiast who has completely abandoned the traditional typing interface in favor of something much more fluid.
This innovator describes a scene where he might look like a “crazy man” walking around Singapore muttering to his phone, but he is actually performing high-level prompt engineering. By shifting his primary input method from text to voice, the author is leveraging the raw speed of speech to interact with Large Language Models (LLMs) like Gemini. He argues that this method isn’t just about convenience; it is about fidelity.
When we type, we edit. When we speak, we flow.
This workflow captures the “nuances and noise” of the original thought, allowing the AI to process the raw data into structured gold. I believe this shift from syntax-focused typing to intent-focused speaking is where the next leap in productivity lies.
⚙️ The Mechanics of Voice-First Engineering
The core concept the original poster highlights is the efficiency of capturing ideas in their “original form.” Most of us have trained ourselves to translate our thoughts into written sentences before we even touch a key. This internal translation layer slows us down and often filters out the creative messy parts that lead to breakthroughs. The author’s approach bypasses this filter entirely.
Here is how the mechanism works in practice. You utilize the voice input feature of a multimodal model like Gemini or ChatGPT. Instead of trying to craft the perfect prompt sentence by sentence, you simply ramble. You describe the problem, the context, the constraints, and the desired outcome in a stream of consciousness. Because modern LLMs are incredibly adept at parsing intent from unstructured data, they can sift through the “ums,” “ahs,” and disjointed sentences to find the core instruction.
The expert notes that while the raw audio isn’t the final product, the AI instantly cleans, structures, and executes based on that audio.
It turns a five-minute ramble into a pristine blog post, a coding solution, or a strategic plan. This essentially treats the LLM as an intelligent editor that doesn’t just correct grammar but synthesizes logic from noise.
📌 Insight 1: Capturing the Nuance of “Messy” Thoughts
The most compelling argument the creator makes is regarding the preservation of nuance. When you sit down to write a prompt, you are forced to structure it linearly. However, human thought is rarely linear; it is associative and branching. By forcing our thoughts into a text box, we often lose the subtle connections that make an idea unique.
This voice-first method allows you to include those tangents. For example, if you are brainstorming a marketing strategy, you might verbally wander into a side note about a specific customer demographic or a tone of voice you want to emulate. In a text prompt, you might delete that sentence to keep things concise. In a voice note, you keep it. The AI can then be instructed to weigh that “side note” heavily in the final output. The author specifically mentions using this for generating training topics and solutions. By verbally “dumping” the entire context of a business problem, including your frustrations and uncertainties, you give the AI a richer dataset to work with than a sterile, well-written paragraph would provide. The “noise” in your head becomes valuable “signal” for the model.
📌 Insight 2: Speed as a Quality Multiplier
Quantity has a quality all its own, and speed allows for rapid iteration. The LinkedIn user emphasizes that he started this habit in 2025 and is doubling down for 2026 because of efficiency. We speak significantly faster than we type. The average person types at about 40 words per minute but speaks at around 150 words per minute. This means you can provide three times the amount of context in the same amount of time.
Consider a scenario where you need to debug a complex situation or write a detailed Standard Operating Procedure (SOP). Typing out every step and condition is exhausting. With this method, you can simply narrate the process as you visualize it. You can say, “Okay, step one is this, but be careful because sometimes X happens, and if X happens, we need to do Y.” The AI listens to this entire logic flow and can output a perfectly formatted, numbered list or a flow chart description. The author uses this to generate “posts and solutions” instantly. The barrier to entry for creating complex documents drops to near zero when the input mechanism is as effortless as a conversation.
📌 Insight 3: The “Meta-Prompt” Technique
A subtle but powerful implication of the author’s post is the idea of using voice to generate the actual prompts for future use. He calls it “prompt engineering via voice notes.” This suggests he isn’t just asking the AI to write a post; he is asking the AI to write a prompt that will write the post.
Imagine you are walking your dog and you have an idea for a specific image generation style or a complex data analysis query. You don’t have the syntax memorized. You can dictate: “I want a prompt that tells an AI to act as a senior data analyst. It should take a CSV file, look for trends in Q3, and output a table. Make sure the prompt includes instructions to ignore outliers.” The LLM takes your voice instruction and outputs a sophisticated, technically accurate text prompt that you can copy and paste later. This allows you to “code” or “engineer” systems while away from your desk. It turns the world into your workspace without requiring you to bury your face in a screen. You are essentially programming natural language with your voice.
Potential Challenges and Nuances
While this workflow is powerful, there are practical limitations to consider. The first is privacy. As the author jokingly admits, he looks like a “man in black” talking suspiciously to his phone. You cannot easily do this in a crowded coffee shop or an open-plan office without disturbing others or revealing sensitive information. It requires a private space or a lack of inhibition.
Secondly, the AI’s ability to understand is high, but not perfect. If your “ramble” is too disjointed or contradictory, the model might hallucinate connections that aren’t there. It is crucial to review the output. The raw audio is the source of truth, but the AI’s interpretation is still an estimation of your intent. You must develop a habit of verbally indicating structure, such as saying “New topic” or “Important note,” to help guide the model through your audio stream.
Final Thoughts
The original creator challenges us to “think outside the box” for 2026. The habit of tethering ourselves to a keyboard is a relic of the pre-AI era. By embracing voice, we remove the friction between thought and execution. I highly recommend reading the full post to see the author’s enthusiasm firsthand and perhaps try a voice session yourself today.
Check out the full post here!