{"html":"<p>At the start of the year, I made a decision: stop subscriptions. Thanks to AI, <a href=\"/article/je-tapprends-a-creer-une-application-de-a-a-z-sans-coder\">I know how to develop my own tools</a>, so why pay for yet another app?</p><p>Except that one app got me: <a href=\"https://wisprflow.ai\" target=\"_blank\" rel=\"noopener noreferrer\">Wispr Flow</a>.</p><p>The principle is simple. One touch on the Mac or iPhone, I speak, and the text appears, clean and directly usable, in an email, Slack, a code editor or a CRM. The tool removes hesitation and understands my jargon.</p><p>The figures convinced me more than any speech:</p><figure class=\"viz\"><ul class=\"viz-bars\"><li data-value=\"100\"><span>Very good typist</span> <strong>80 to 100 words/min</strong></li><li data-value=\"150\"><span>Competitive typist</span> <strong>150 words/min</strong></li><li data-value=\"175\" class=\"is-highlight\"><span>Me, after six weeks of dictation</span> <strong>175 words/min</strong></li></ul><figcaption>Writing speed in words per minute: the keyboard versus my dictation.</figcaption></figure><p>In six weeks, I had dictated 17,738 words, the equivalent of a book. It's not AI writing for me. These are my words, at the speed of my thought.</p><h2>Talking is faster than typing, and we think better out loud</h2><p>Since then, I dictate almost everything to the AI. Speed ​​isn't the only gain. Out loud, we structure our ideas differently: we explain, we reformulate, we follow through with the reasoning. The keyboard pushes you to cut it short.</p><h2>The machine now keeps pace</h2><p>For a long time, voice recognition was the weak link. This is no longer the case.</p><p>In May, <a href=\"https://platform.openai.com/docs/guides/realtime\" target=\"_blank\" rel=\"noopener noreferrer\">OpenAI</a> released a <a href=\"https://platform.openai.com/docs/guides/realtime\" target=\"_blank\" rel=\"noopener noreferrer\">new real-time audio suite</a>. I pushed it to its limits: I launched a YouTube video at x2 speed, or around 350 words per minute. Everything was transcribed, in real time, with a quality that largely holds up.</p><p>The real leap isn't just accuracy. It's latency. The window between what is said and what is produced is narrowing to a level I didn't expect so soon. And that opens up very concrete uses: multilingual meetings, live call analysis, training, accessibility.</p><figure class=\"article-inline-figure\"><img src=\"https://images.unsplash.com/photo-1617994452722-4145e196248b?crop=entropy&cs=tinysrgb&fit=max&fm=jpg&ixlib=rb-4.0.3&q=80&w=2000\" alt=\"Onde sonore bleue sur fond noir, comme une voix en train d'être transcrite\" loading=\"lazy\"><figcaption>The voice travels faster than the fingers: the text follows the wave, almost in real time.</figcaption></figure><h2>The end of forms</h2><p>I took advantage of a rainy Sunday to test this new vocal model. Result: <a href=\"/article/mon-ceo-est-une-ia\">an agent listening to a conversation</a>, extracts key information in real time, and fills out a CRM form without having to touch the keyboard.</p><figure class=\"viz\"><ol class=\"viz-flow\"><li><strong>Conversation</strong> <span>The agent listens to it</span></li><li><strong>Extraction</strong> <span>It pulls out the key information in real time</span></li><li><strong>CRM sheet</strong> <span>Filled without touching the keyboard</span></li></ol><figcaption>My test on a rainy Sunday: from conversation to CRM sheet, without forms.</figcaption></figure><p>The demo interests me less than the idea behind it: voice as a new interface for business applications. No more forms, menus, fields to fill out. We speak, the application understands, acts and structures.</p><ul><li>The field technician dictates his report when leaving the customer.</li><li>The caregiver completes their form during the consultation.</li><li>The manager on the move creates a brief out loud.</li></ul><h2>Transcribing is level zero</h2><p>A meeting without transcription is a lost meeting. But in 2026, transcribing is the minimum. What matters is what we do with it next.</p><p>In my team, each sales meeting transcript goes through an AI assistant which draws a summary: weak signals, who decides what, maturity of the prospect, next actions, follow-up email written. In seconds. We even made an app out of it, <a href=\"/projets/client-signal\">ClientSignal</a>, built in 72 hours during a buildathon <a href=\"https://lovable.dev\" target=\"_blank\" rel=\"noopener noreferrer\">Lovable</a> and <a href=\"https://www.granola.ai\" target=\"_blank\" rel=\"noopener noreferrer\">Granola</a>.</p><p>A good meeting from which nothing is extracted, it's just a forgotten conversation.</p><h2>The real brake is not technical</h2><p>If voice is so effective, why isn't everyone dictating already? Because in the office, no one wants to talk to their computer in front of their colleagues. The obstacle has never been technology. It’s how others see it.</p><figure class=\"article-inline-figure\"><img src=\"https://images.unsplash.com/photo-1556761175-4b46a572b786?crop=entropy&cs=tinysrgb&fit=max&fm=jpg&ixlib=rb-4.0.3&q=80&w=2000\" alt=\"Open space où plusieurs personnes travaillent côte à côte à leur bureau\" loading=\"lazy\"><figcaption>In an open space, talking to your machine is visible. This is often the real obstacle.</figcaption></figure><p>That's why the <a href=\"https://subvocal.company\" target=\"_blank\" rel=\"noopener noreferrer\">Subvocal</a> had such an impact on me. This startup passed by <a href=\"https://www.ycombinator.com\" target=\"_blank\" rel=\"noopener noreferrer\">Y Combinator</a> has designed glasses that turn words you don't speak into text. We articulate in silence, the system reads the micro-movements of the face and writes in real time.</p><p>And the use goes far beyond open space. For a person deprived of voice after an illness, an operation or an accident, articulation often remains possible even though the sound does not come out. Imagine: she articulates, the glasses transcribe, a synthesized voice speaks for her. His interlocutor has nothing to learn, nothing to install. He listens and responds.</p><p>We are still in the prototype phase, and nothing says that it will work for everyone. But for the first time, an interface designed for productivity could give voice to those who have lost it.</p><h2>Keep your feet on the ground</h2><p>Do not confuse demo and production. When <a href=\"https://elevenlabs.io\" target=\"_blank\" rel=\"noopener noreferrer\">ElevenLabs</a> showed a voice agent empathetically handling the cancellation of a flight, many cried out for the death of call centers. Watch the video a second time: it's a controlled demo, not a real call at 11 p.m. on a Friday night, facing a slow system.</p><p>Between the two, there is a canyon:</p><ul><li><strong>responsibility</strong>: If the agent grants a refund in error, who pays?</li><li><strong>integration</strong>: he needs real-time access to the CRM, reservations, prices;</li><li><strong>borderline cases</strong>: the client who negotiates, who threatens, who requests an exception.</li></ul><p>Appointment bookings and frequently asked questions can be automated now. Disputes and negotiations, not for several years. Call centers will shrink, not disappear tomorrow.</p><h2>What really changes</h2><p>The keyboard is not going to disappear overnight. But for the first time, voice is fast enough, precise enough, and well understood enough to become the default interface for many of our tools. What remains to be invented are the uses, and the way to talk to your machine without disturbing your neighbors.</p><p>And you, do you still only work on the keyboard?</p>","dek":"Six weeks of dictation with Wispr Flow, a test of OpenAI's new voice models and an observation: voice is becoming the interface of our tools, but the brake is not technical.","quickTake":["After six weeks with Wispr Flow, I dictate at 175 words per minute, compared to 80 to 100 for a very good typist. That’s 17,738 words, the equivalent of a book.","OpenAI's new real-time audio suite tracked a YouTube video at 2x speed, around 350 words per minute. The real progress is latency.","Voice can replace forms: in one Sunday, I set up an agent who fills out a CRM form while listening to a conversation.","The barrier is not technical: no one wants to talk to their computer in an open space.","A voice agent demo is not production. Appointments and frequently asked questions, yes; disputes and negotiations, not for several years."]}