We stopped teaching one expressive skill and replaced it with a deliberately throttled one

I want to talk about something that has been bothering me for a while, and the more I look at it the more urgent it feels.

We used to teach children cursive for a reason. Not because it was charming or traditional, but because handwriting speed and legibility were direct prerequisites for professional participation. The faster and more clearly you could write, the more effectively you could think, record, and communicate. It was a practical skill with a practical justification.

Then keyboards became the medium of professional life and we made a sensible decision. We replaced cursive instruction with typing classes. We taught children to use the tool the world had adopted. That was reasonable. What was not reasonable, and what nobody seriously questioned, was what we were teaching children to type on.

The QWERTY keyboard was designed in the 1870s to solve a mechanical problem that has not existed for over a century. We have been teaching that solution to every generation of children since.

Christopher Sholes engineered key separation deliberately to prevent typewriter jams. The layout was not optimized for human speed or comfort. It was optimized for the mechanical limitations of a machine that no longer exists. We carried it through electric typewriters, into personal computers, into smartphones, and now into the era of AI, and at no point did we stop to ask whether it still served us.

The average professional types at 40 to 60 words per minute on a QWERTY keyboard. Court reporters using stenography routinely exceed 200. That gap has always existed. What has changed is what is waiting on the other side of the keyboard.

We need to learn from China

For most of the twentieth century, Chinese was considered incompatible with modern typing technology. The writing system comprises tens of thousands of characters. Early mechanical Chinese typewriters required operators to memorize four-digit codes for six thousand characters and were agonizingly slow. Western observers used this as evidence that Chinese script was destined for obsolescence in the information age.

China refused to accept that. Chinese engineers and typists spent decades innovating around the constraint, inventing input methods, predictive systems, and structure-based entry techniques that broke characters into their fundamental components and mapped them to standard keys. Predictive text, now universal on every smartphone on earth, emerged in significant part from that effort.

In 2013, a young man named Huang Zhengyu won the National Chinese Characters Typing Competition using a structure-based input method on a standard QWERTY keyboard. His average speed across the competition was 221.9 Chinese characters per minute. His opening burst extrapolated to 372 characters per minute. Researchers writing in 2024, including those at the MIT Technology Review and Stanford professor Thomas Mullaney in his book The Chinese Computer, still cite this performance as the landmark benchmark precisely because the numbers remain extraordinary a decade later. That is not a sign the record has not moved. It is a sign the culture that produced it has only deepened. China now has over 900 million internet users who type Chinese daily, a thriving competitive typing scene, and by 2019 was already experimenting with brain-computer interface typing competitions. The 2013 numbers are a floor, not a ceiling.

Had Huang's speed been measured against English typing records it would have shattered every benchmark ever set. The fastest verified English typist ever recorded achieved 174 words per minute. The fastest unofficial record is 256. Huang was operating at an equivalent rate that exceeded both, in a language that a century ago the world wrote off as impossible to type.

A writing system the world considered a liability became a competitive advantage, because the people who used it refused to surrender it and built something extraordinary around the constraint instead.

English speakers faced no such constraint. We inherited a slow keyboard, taught children to use it, and called the problem solved.

The machine on the other end has changed

For most of the keyboard's history, typing speed was a professional convenience. Faster typists produced more output in less time. The gap between 45 and 200 words per minute mattered at the margins.

That is no longer true. The machine on the other end of the keyboard is now the most consequential communication partner a professional will ever have. AI systems do not compensate for slow or imprecise input. They execute exactly what they receive. A developer, analyst, writer, or strategist who can express a precise, well-formed thought at 200 words per minute is not just faster than one operating at 45. They are operating in a fundamentally different register of productivity and precision.

There is also a signal worth paying attention to, even if nobody fully understands it yet. OpenAI's reasoning model o1 has been observed switching into Chinese during its internal reasoning process, even when prompted in English. Researchers debate why. Some point to training data. Others argue the model may be gravitating toward more semantically dense representations for certain reasoning tasks, that Chinese characters, each carrying a complete concept, may be computationally efficient in ways that token-by-token English is not. No one has confirmed this definitively. But the fact that the most advanced reasoning AI in the world may be finding efficiency in the same direction that produced those 372-character-per-minute typing speeds is not nothing. It is a signal about the relationship between linguistic density, precision, and machine intelligence that English-speaking countries should be paying attention to.

Why voice is not the answer

The obvious response is that voice input solves the throughput problem. Humans speak faster than they type. Why teach stenography when children can simply speak to their AI?

Because speaking to an AI is not speaking to the AI. It is speaking to a transcription layer that converts your speech into text, which is then passed to the model. That transcription layer has its own error rate, its own biases, and its own failure modes. It struggles with technical vocabulary, proper nouns, domain-specific terminology, and precisely the kind of language that produces high-quality AI output. It is sensitive to background noise, dependent on connectivity, and introduces a step between thought and machine that the typist does not have.

Consider who invented stenography and why. Court reporters do not type, and they do not use voice recognition. They use stenography precisely because it is the only input method ever devised that outpaces voice itself, capturing live human speech at full speed with full fidelity and zero transcription layer. Stenography was not invented as an alternative to typing. It was invented as an alternative to voice. The argument for teaching it to children is not that it is faster than a keyboard. It is that it is more faithful than a microphone.

This is the game-of-telephone problem applied to a child's future career. Every layer between intent and execution is a fidelity loss. The professional who speaks a complex technical requirement to a transcription layer and then to an AI model has introduced two opportunities for their meaning to be garbled. The professional who types that requirement directly introduces none. As I wrote in my previous piece on AI development as a communication problem, fidelity loss compounds. One bad link and nothing arrives correctly on the other end. Voice adds a link. Stenography removes one.

This is a pipeline problem

In my previous article I raised the question of where the senior developers of 2040 are coming from. Organizations are eliminating junior roles on the assumption that AI makes them redundant, without understanding that senior expertise is not hired, it is grown, over years, through the kind of consequential mistakes and hard-won pattern recognition that cannot be shortcut.

This article is the earlier chapter of that same story. The children who learn to communicate with machines precisely and at speed are the professionals who will have the grandmaster advantage I described there. The ones who grow up typing at 45 words per minute on a layout engineered for mechanical throttling, into voice layers that garble their technical vocabulary, are the ones we will struggle to develop into the practitioners we need.

This is not just an education question. It is a national competitiveness question. The countries that take it seriously will produce professionals who operate at a different level than the ones that do not. China did not set out to dominate typing competitions. They set out to preserve their language in the information age and built an extraordinary capability in the process. English-speaking countries have no such forcing function. We have to choose this intentionally, or we will not choose it at all.

What stenography actually is, and what it costs

Stenography is a chorded input method. Where typing presses one key at a time, stenography presses multiple keys simultaneously, with each chord producing a syllable, word, or phrase. A trained stenographer does not type the letters in "intention." They press a chord that produces the whole word in a single motion. Court reporters have used this system to capture live speech in real time for over a century. Stenographers have been able to achieve astounding speeds, exceeding 400 wpm.

Until recently, learning stenography required expensive proprietary hardware and professional training programs. That barrier no longer exists.

The Open Steno Project has made the entire ecosystem free and open source. Plover, the project's core software, runs on Windows, Mac, and Linux at no cost and works with a standard keyboard, meaning a child can begin learning today on the computer they already own. We also now have an open-source system for these as well, Javelin, which gives you an embedded steno engine, meaning the device can plug in just like any keyboard. The community has produced free learning games, practice tools, and a Discord server where beginners can get help from experienced practitioners. Just keep in mind these are not professional organizations but open-source communities that collaborate and help one another.

For families who want dedicated hardware, hobbyist steno machines designed and sold by community members start at well under a hundred dollars. Professional machines used by court reporters cost thousands, but a child does not need a professional machine to develop the skill. They need the software, a keyboard, and consistent practice. You can even make your own without too much effort.

The skill is learnable. The resources exist. The barrier is awareness, which is the only thing this article is trying to address.

So, what do we do?

I am not suggesting stenography replace typing instruction in every classroom tomorrow. I am suggesting that the people who make decisions about what children learn, parents, educators, curriculum designers, and policymakers, take seriously the question of what expressive-speed skills we are equipping the next generation with, and whether the answer we currently have is sufficient for the world they are entering.

We made a reasonable decision when we replaced cursive with typing instruction. The keyboard was the medium of professional life and children needed to use it. The medium of professional life has changed again. The keyboard is still there, but what is on the other side of it is completely different, and the throughput and precision with which a person can express themselves to that system will define their professional ceiling in ways we have not yet fully reckoned with.

We are sending children into that world at 45 words per minute on a layout designed to not break a typewriter. That is not a curriculum problem. That is a national competitiveness problem, and the time to treat it as one is before the gap becomes obvious.