Steven Bartlett stopped typing. He made voice his primary input across every device and every channel he communicates on, using Wispr Flow, and the time it takes him to send a message fell from thirty seconds to five. Ninety percent of what he speaks now needs no editing at all. Inside his company, eighty-nine percent of his team followed him onto it. The constraint was never his thinking. It was his hands.
The problem was accuracy, not speed
Bartlett had tried dictation before, and AI Central's account of his switch is blunt about why it never stuck. Even at a ten percent error rate, other tools meant re-editing every message. So he typed, and stayed slow.
That figure deserves a second look, because ten percent sounds tolerable. It is not. One wrong word in ten means you cannot trust a single sentence without reading it back. And the moment you have to proofread, you have not removed the work of writing, you have moved it, from composing to correcting, and correcting is the slower of the two.
This is the threshold most people miss when they judge a voice tool. They test how fast the words appear on screen. The number that actually decides whether the habit survives its first week is the edit rate.
What the bottleneck actually cost him
Bartlett hosts The Diary of a CEO, described as the world's number two podcast, and founded a four hundred and twenty-five million dollar creator holding company. Being the slowest link in a chain that size is expensive in a very particular way, and he named the cost himself.
I become a bottleneck in my own company.
That phrasing from Bartlett is more precise than saying he was busy. A bottleneck is not a person with too much to do. It is a point where other people's work stops. Work that needed him did not move.
The second cost was worse, because it never appeared on anyone's task list. Bartlett estimates he lost seventy to eighty percent of his best ideas. His best thinking happened while he was moving, walking, exercising, in transit, and he had no way to capture it. Ideas do not wait politely for you to reach a keyboard.
Voice as the default, everywhere, not somewhere
By AI Central's account the switch was not that Bartlett used a dictation app occasionally. He deployed Wispr Flow across every device and every communication channel, and voice became his primary input.
The word carrying the weight there is every. A dictation tool confined to one app forces a decision every time you open a different one, and a decision that repeats forty times a day is a tax that eventually kills the habit. When the same input works in the mail client, the messaging app and the document, there is nothing left to decide. You speak by default and type by exception.
Universal deployment is also what makes the capture problem solvable. If voice only works at the desk, the ideas you have on a walk are still lost. If it works on the phone in the street, they are not. Bartlett's two problems, the bottleneck and the lost thinking, turn out to have one fix, and only because the fix was everywhere.
The numbers, and how to read them
Bartlett's own figures on the result are specific. Message time fell from thirty seconds to five, and ninety percent of what he says now requires zero edits.
The gap between my thought and my delivery collapses.
Thirty seconds down to five, in Bartlett's telling, is a six-fold reduction on a single message, and the compounding across a working day needs no further claim to be obvious. The wider ceiling comes from a simpler pair of numbers AI Central puts at the end of the case: a typical typing speed of around sixty words per minute against a speaking rate of about two hundred and twenty. That is close to four times the throughput before anyone opens an app.
Read these for what they are. They are one executive's self-reported estimates, recalled after the fact, not instrumented measurements, and the seventy to eighty percent of lost ideas is by definition an estimate of things that were never recorded. What makes the case persuasive is not the precision of any single number. It is that the mechanism is legible. Speaking has always been faster than typing, and the only thing that stopped people exploiting the difference was error rate.
When one person's speed becomes the team's
The part of Bartlett's story most readers skip is what happened after him. Eighty-nine percent of his team adopted Flow once they saw the difference, and it spread across the organization.
That belongs in a separate category from personal productivity, because latency inside a company is not an individual property. A reply that takes you thirty seconds to write also costs the person waiting on it, and everything routed through the slowest hop inherits its speed. When Bartlett got faster, the queue behind him got shorter. One person's speed became the team's operating standard.
It also arrived the right way round. Nobody mandated it. They watched it work and copied it, which is the only adoption pattern for a personal tool that reliably holds.
What to do with this
If you want to run the same experiment, the sequence matters more than the tool.
- Time yourself honestly on one ordinary message before you change anything. Without a baseline you will never know whether you got faster or merely felt faster.
- Judge any dictation tool on edit rate, not word rate. If you are still correcting the output, you are still typing, just later and in a worse mood.
- Install it everywhere at once, phone included, not only at the desk. Partial deployment is what kills these habits in week one.
- Point it first at the work you already resent. The long reply, the message you have been avoiding for two days, the thought you had on a walk and lost by the time you sat down.
- Watch whether anyone copies you. If your own speed improves and nothing downstream changes, you were not the bottleneck, and something else is.
For anyone already dictating and unimpressed, the diagnosis in Bartlett's case is the useful part. The failure mode is rarely that speaking does not work. It is that a tool with a ten percent error rate trained you to distrust it, and you never stopped proofreading long enough to collect the gain.
How much faster is speaking than typing?
The comparison AI Central uses is roughly sixty words per minute typed against about two hundred and twenty spoken, close to four times. Bartlett's practical gain on short messages was larger than that ratio, thirty seconds down to five, which suggests the raw word rate understates what changes when the tool is trusted.
Why did dictation not work for me before?
Bartlett's answer is accuracy. Even a ten percent error rate meant re-editing every message, so he went back to typing and stayed slow. If you have to check every sentence, the speed gain is cancelled by the correction pass, and most people quit at that point instead of diagnosing it.
Does this only work for people who talk for a living?
Bartlett is a podcast host, so the objection is fair, and the mechanism does not depend on it. The claim in his case is not that he speaks unusually well. It is that his hands were slower than his head, which is true of almost everyone with a keyboard.
Should I roll this out to my whole team?
In Bartlett's case it was not rolled out, it spread. Eighty-nine percent of the team adopted it after seeing the difference in him. That order is the part worth copying: prove it on your own throughput first, let people ask, then standardize. A mandated input method is a policy people resent. A copied one is a habit.
What does Wispr Flow actually do?
It turns speech into text across the apps you already work in, which is why Bartlett was able to make it his primary input on every device rather than a tool he opened on purpose. What separates it in his account is not a new capability, it is accuracy. Ninety percent of what he says needs zero edits, and that is the threshold at which a person stops proofreading and starts actually saving time.