Sending Raw Romaji Directly to AI Is Just Too Easy

I'm starting to make it a habit to input text into AI using Romaji.

As of May 2026, voice input using the Groq+SuperWhisperxGroq combination is reportedly accurate and fast.

I often see posts on X saying Gemini 3.5 is the most accurate at recognizing messy handwriting. I compared the same text across Grok, Claude Opus 4.7, and Gemini 3.5, and Gemini won by a landslide.

Since voice input and handwriting recognition have become this accurate, I started wondering if typing errors even matter anymore. I tried a style where I just send raw Romaji (half-width alphanumeric characters)—not even Hiragana—directly to the AI, and it's so easy I can't go back.

I was already aware that my instructions to AI were becoming increasingly casual. Even with phrasing like "Remember that thing we did yesterday? Yeah, the continuation of that," the AI understands perfectly (as long as you pay the token fees).

However, this Romaji notation doesn't even look like proper Japanese anymore. Yet, modern AI quickly and accurately translates it into perfect Japanese sentences and follows the instructions.

I feel like I'm clearly losing something important. But it's human destiny to evolve while losing abilities—losing hunting skills to agriculture, agricultural skills to industrialization, and the ability to wait to smartphones. Even if left alone, we lose various abilities to old age. So, rather than fearing loss, as someone living in this era, I want to gratefully enjoy the technology.

No Need to Delete Typos

The merit of Romaji notation is that you don't have to worry about typos. Even if you make a mistake, there's no need to move your finger to the backspace key. Just keep typing. That level of input error isn't a problem at all. In fact, with this method, you don't even notice you're making mistakes. Your thinking isn't interrupted by the slight stress of seeing unexpected characters on the screen.

You don't even need punctuation. The AI inserts it at appropriate intervals, and if you want to explicitly separate sentences, you can just use a space like in English. You don't even need to type the long vowel mark "ー," which I used to hit with my pinky while making frequent mistakes. If you want to type "keyboard" (kiiboodo), just hit "kiiboodo."

Some people might use custom dictionary registrations extensively. Even that can be solved with a little ingenuity by letting the AI read your dictionary list. For example, if :k converts to your company email address, you can just tell the AI a rule like "if ' k ' is surrounded by spaces, convert it." There's no need to go out of your way to press keys far from the home position like ":".

No Need to Convert

The most comfortable part might be "not having to convert to Kanji." There's no need to stop your typing flow for incorrect conversions.

Stopping your thoughts for a split second to select the right candidate from a list is nothing but noise. The Kanji that fits the context is already in your head; pressing the space bar multiple times to select it feels like a truly wasteful task.

I never knew a world where you only need one type of character (the alphabet) without Kanji could be this comfortable. We've been using Twitter with a heavy handicap all this time.

Useful in Meetings Too

This input method is also very convenient for meetings where recording and transcribing isn't quite necessary or is difficult.

You can just keep your hands moving while listening to the other person. Since you don't need to worry about typos or conversion errors, you can focus on listening. If you can touch-type, you can listen while looking the person in the eye. They might wonder, "What is this person clacking away at...?" though.

AI of Tokens Over My Tokens

"Isn't it a waste to make the AI consume tokens to convert a string of Romaji into correct Japanese?"

That's certainly true.

However, the "brain tokens" saved by Romaji notation are far more valuable than the AI tokens saved by writing correct Japanese. In the coming era, what we should save is human tokens, not AI tokens. Last month, I realized this after almost losing my mind from a dopamine overdose caused by multitasking with multiple terminals while obsessed with AI. The power to think is finite.

Limits and a New Personality

There are certainly expressions and inspirations that only emerge through the style of putting thoughts onto fingers and completing a text while simultaneously deleting typos and conversions.

The merit of the usual writing style shines during the proofreading process to polish a text for others to read. For proofreading and editing, I have no choice but to use the usual method of deleting and adding.

However, when starting from zero, the method of throwing Romaji to AI for a clean draft is overwhelmingly faster.

One might say voice input is better if you want speed. But spoken and written language have clearly different properties as tools. Thinking patterns cultivated through reading can only be used with written words. If you want to think like Kant or Keikaku Itoh, you have to think while writing.

There are times when you have to dig deep into thoughts that are too embarrassing or scary to say out loud, and that's impossible with voice input.

In fact, introspection that requires such deep concentration and resolve is too fast even for typing. Handwriting is best suited for facing trauma.

Bilingual people say their personality changes depending on the language they speak. I also feel my personality differs between written and spoken language. In fact, it's completely different between handwriting, typing, and voice input.

Now, a new personality called "AI + Romaji" has been added.

I truly think I was born in a great era. Life is the most fun right now.

※https://pody.jp/player/ngpcDWPy6otqtyr7XYmx/gdfr60bU6m6Jtpq2jyQv#section-2