For thirty years, typing speed was a real measure of productivity. Offices hired by it. Schools taught it. Adults benchmarked themselves against it. There was a real, lived sense that the person who typed faster got more done.
In 2026, that is no longer true. Not because typing got slower, but because the bottleneck moved. Optimizing typing speed today is optimizing the wrong layer of the stack — like tuning the carburetor on a car you have stopped driving.
This is the contrarian case. It is not "typing is bad" or "everyone should dictate everything." It is something narrower and harder to argue against: the gain you get from raising your WPM is now consistently smaller than the gain you get from changing how you compose. And that changes what you should be practicing, if you are practicing anything at all.
The Old Model: WPM as Productivity
The argument for typing speed used to be straightforward. The faster you type, the more text you produce per hour. If your job is producing text, faster is better. The numbers worked: a 70 WPM typist out-produced a 40 WPM typist by 75% on equivalent tasks. That gap was worth training for.
The model assumed that the keyboard was the slowest link in the chain. You knew what to write. The only question was how fast you could get it onto the page. Practice closed the gap between mind and screen.
This model held for most of the post-typewriter era. There was no realistic alternative to the keyboard. Voice recognition was unreliable, expensive, and required training each system on each speaker. Mobile keyboards were worse than physical ones. So you trained your hands and accepted the ceiling.
What Changed
Three things changed in the last few years, and together they invalidated the old model.
1. Voice transcription actually works now
From roughly 2022 onward, AI-driven speech recognition crossed a quality threshold. Accuracy above 95% became normal even for accented speech and moderate background noise. Latency dropped below a second. Punctuation handling stopped being embarrassing. The output now looks like text, not a transcript.
This is the single biggest change. For the entire history of computing, voice-to-text was the technology that was "almost there." It is now here. Not perfect, but better than fast typing for a large fraction of text-production tasks.
2. AI rewriting closed the messiness gap
One of the historical objections to dictation was that spoken language is messier than written language. You repeat yourself, trail off, use filler words. Raw transcripts looked unprofessional.
That objection is largely defused. Modern dictation tools include a cleanup step that handles the messiness automatically — removing "uhms," normalizing sentence structure, even rewriting verbose passages into tight prose. The output reads like written language, even when the input was conversational.
3. The composition layer became the bottleneck
This is the deeper change. In 2026, the limiting factor for most knowledge work is not how fast you can type — it is how fast you can compose useful thought. AI assistants now help generate, refine, and structure that thought. Suddenly the bottleneck on your output is not your fingers at all. It is the speed of the conversation between your brain and the tools you use to think.
That conversation is much better served by talking than by typing.
The Layer Problem
Here is the engineering way of looking at it. The text production stack has three layers:
- Composition: deciding what to say
- Encoding: turning the thought into characters
- Rendering: getting those characters into the application
For most of computing history, layer 2 (encoding via keyboard) was the slowest. Layer 1 ran ahead of you — you knew what to write, and your fingers had to catch up. So training the fingers paid off, because they were the bottleneck.
In 2026, layer 2 has multiple options. Keyboards still exist; you can also dictate via voice. Voice runs at 3-4x keyboard speed. So layer 2 stopped being the bottleneck for anyone who switched to voice.
What is the bottleneck now? Layer 1 — composition. The slowest part is figuring out what to say. And that has nothing to do with WPM.
So when you sit down to practice typing, you are training the second-slowest layer. The gains are real but small, because the layer was already not the bottleneck. The leverage moved up the stack.
What Actually Slows You Down at a Desk
If you watch a knowledge worker for a day, you can see where the actual time goes. It is rarely typing speed.
Switching costs
Most of a workday is spent moving between contexts. From a meeting to a Slack thread to an email to a doc to a code editor. Each switch costs minutes of orientation. Typing speed does not affect this cost at all.
Composing the first sentence
The hardest part of any document is starting it. Writers stare at blank cursors for ten minutes before they type a word. The 40 WPM vs 70 WPM distinction is irrelevant during those ten minutes. Either way, no characters are appearing.
Re-reading and revising
Most of "writing" is re-reading what you wrote, finding the part that does not work, and rewriting it. The mechanical act of typing is a small fraction of the time. The thinking-about-what-you-typed is the bulk.
Reading documentation and references
For coders, technical writers, and analysts, a huge chunk of the workday is reading — docs, code, prior reports, emails. Reading speed and comprehension are the bottlenecks, not output speed.
Waiting
Waiting for builds, waiting for replies, waiting for meetings. Vast amounts of professional time are spent waiting for someone or something. Typing faster does not change the waiting.
When you actually log where time goes, pure text production is usually less than 25% of a knowledge worker's day. Within that 25%, doubling your typing speed gives you maybe an extra 12% of effective productive time. The lever is real but small.
The Composition Lever
Compare that to leverage at the composition layer.
When you talk through an idea instead of typing it, several things happen at once. First, you produce text at speech speed — 3-4x faster than typing. Second, you compose in the verbal register that you actually think in, which is usually faster to generate than the careful written register. Third, you tend to externalize ideas you would otherwise not bother typing because the cost was too high — the small clarifications, the digressions that turn out to be the actual point.
The output is rougher than carefully typed prose. But it has more in it. And modern AI cleanup tools can take that rough output and tighten it without losing the substance.
So the comparison is not "fast typing vs voice dictation" on identical content. It is "what you produce while typing vs what you produce while talking." The latter is consistently richer because the act of speaking surfaces more of what you actually know.
Where Typing Still Matters
It would be silly to claim typing speed is irrelevant for everything. There are still domains where it matters.
Programming
Code production is interrupted, deliberate, and full of symbols that are awkward to dictate. Most experienced programmers will tell you typing speed has never been their bottleneck. But editing, navigation, and refactoring are keyboard-bound, and being fast and fluent at the keyboard is part of the job.
Editing and revision
The "shape this paragraph correctly" work is keyboard work. Voice is bad at precise edits. You still want fast, accurate typing for the editing phase.
Quick replies and shortcuts
Short messages, command-line work, and other one-off interactions are usually faster by keyboard than by voice, simply because you avoid the overhead of holding a hotkey and saying a sentence to produce three words.
Quiet and shared environments
Open offices, libraries, shared homes — voice dictation has social friction that typing does not. The keyboard remains the default for any environment where speaking aloud is awkward.
So the keyboard is not going away. Touch typing is still a worthwhile skill — for the same reasons literacy is. But it is no longer the limiting factor in how much text you produce in a day.
The Number That Matters
Forget WPM for a moment. The number that actually predicts productivity for someone whose job is producing text is something like "words of useful output per hour, end-to-end, including thinking, composing, drafting, and editing."
For most knowledge workers, this number is far lower than their raw typing speed would suggest. A 60 WPM typist who composes well might produce only 200-400 polished words per hour of actual focused work, because most of the time is spent thinking, not typing.
Raising typing speed to 90 WPM might bump that number to 220-450. A meaningful gain but a small one.
Switching to dictation, on the other hand, can change the workflow entirely. Talking out a first draft in five minutes, then editing it in fifteen, often beats typing it in twenty-five minutes — both in time and in quality, because the looser composition surfaces more material.
What This Means in Practice
If you came here looking for advice, here it is.
If you cannot touch type, learning is still worth it. Hunt-and-peck at 25 WPM is genuinely a constraint, and getting to 50 WPM is achievable and useful. Below the threshold, basic competence matters.
If you can already type at the adult average (around 40 WPM) or above, stop chasing higher WPM unless typing is literally your job. The leverage is not there. Spend that effort somewhere else.
If you want to actually go faster at producing text, try voice dictation for the bulk-production tasks: long emails, message replies, document first drafts, notes, journaling. Use the keyboard for editing, code, and short interactions. The combined workflow is faster than either pure approach.
If you have written voice off because you tried it years ago and it was bad — try it again. The technology is in a different place now.
The Tool
Voice Keyboard Pro is one of the easier on-ramps on the Mac. It is a menu bar app, 1.7MB, no big UI to learn. You set a hotkey, hold it, talk, release — text appears at your cursor in whatever app you are using. Email, Slack, Notion, your code editor, the address bar of a browser. It does not care; it sends text wherever your cursor is.
It uses the Voice Keyboard Pro Whisper API for fast cloud transcription, with an offline Apple Speech mode for when you do not want audio leaving the device. Audio is not stored on servers. There is a free tier; the Pro plan is $4.99/month or $34.99/year and adds Smart Rewrite (which handles the spoken-vs-written cleanup), Voice Profile (improves accuracy for your voice), Voice Isolation (handles background noise), and custom vocabulary for domain terms. An iPhone keyboard brings the same workflow to mobile, which is where the typing-vs-voice gap is even larger because phone keyboards are slower than desktop ones.
Typing speed used to be a serious productivity lever. It is now a small one. The new lever is composition speed, and the way you pull it is by talking instead of typing.
The Reframe
This whole article can be compressed to a single observation: the layer of the stack that used to matter most for text production no longer does. The mechanical conversion of thought-to-characters has multiple options now, and the keyboard is no longer the fastest. Optimizing typing speed in 2026 is optimizing a layer that has been demoted from "bottleneck" to "one of several options."
The right thing to optimize is the layer that is still the bottleneck — composition. And the way to optimize composition is, paradoxically, to take the encoding step out of your way. Stop thinking about how to spell things and where commas go and which key your pinky needs to reach for. Just talk. Let the machine handle the rest.
Twenty years from now, the question "what is your typing speed" will sound like "what is your shorthand speed" sounds today — a relic of a workflow that used to be central and is now a niche specialty. We are early in that transition. But the shape of where it lands is already visible.
If you want to test it for yourself, you do not need to commit to anything. Voice Keyboard Pro has a free tier — dictate one long message and compare it to typing the same message. The difference is large enough to settle the argument on a single example.