Humans spoke for tens of thousands of years before anyone picked up a pen. We communicated through voice for millennia before the first keyboard was invented. And yet, for the past fifty years, we have been forcing ourselves to interact with computers through the least natural interface imaginable: typing on a grid of tiny buttons.
I expect that to change soon. My view comes from building a voice AI product, selling it to paying customers, and using it every day.
The Siri Problem
It is worth being clear about where voice technology has been. For most people, voice AI means Siri, Alexa, or Google Assistant. And for most people, those experiences have been underwhelming. You ask Siri a question and get a web search. You tell Alexa to play a song and it plays the wrong one. You try to dictate a message and spend more time correcting errors than you saved by speaking.
These early voice assistants were built on rule based systems and limited speech recognition models. They worked well for simple, predictable commands: set a timer, turn off the lights, what is the weather. But the moment you tried to use natural language, the way you actually speak to another human, they failed.
This taught a generation that voice interfaces are unreliable toys and that typing will always be faster. I understand that skepticism, and I shared it. The technology has changed a great deal since then, and most people have not caught up.
What Changed
Two developments changed voice AI: OpenAI released Whisper, and large language models became practical.
Whisper is an open source speech recognition model trained on 680,000 hours of multilingual data. Beyond recognizing words, it understands context, handles accents, manages background noise, and works across more than 90 languages. When I first tested Whisper, I spoke naturally, the way I would in a conversation, complete with pauses, restarts, and filler words. It transcribed everything accurately. That was when I knew the technology had reached a usable level.
Large language models added the second piece. Beyond transcribing speech, you can now understand intent, generate responses, summarize conversations, extract action items, and translate in real time. Voice is now a full interface layer, more than an input method.
The keyboard was an adaptation to the machine. Voice is the machine adapting to us, and that changes how we should design products.
Why I Built TAWK
TAWK started because I had a simple problem. I write a lot. Emails, documents, strategy memos, Slack messages, social media posts. I noticed that my biggest bottleneck was typing rather than thinking. My brain could formulate thoughts far faster than my fingers could capture them.
I tried existing dictation tools. Most were cloud based, meaning my words were being sent to someone else's server. Many had noticeable latency. Some required subscriptions. None of them felt like they were built for someone who wanted voice to text as a serious daily tool rather than an occasional novelty.
So I built TAWK. It runs locally on your Mac using Whisper's small model. Your audio never leaves your computer. It processes in seconds. It costs $59 once, with no subscription. And it works everywhere: any text field, any application. You press a keyboard shortcut, speak, and the text appears.
I use TAWK for hours every day. It has changed how I work. I draft emails by speaking. I write first drafts of documents by talking through my ideas. I capture meeting notes in real time. It is three to four times faster than typing, and the quality of the output is often better because I am thinking out loud rather than editing as I type.
Voice as a Daily Productivity Tool
The mistake most people make with voice to text is treating it as a replacement for the keyboard. It is a different tool for different moments. I still type when I am editing, when I need precision formatting, or when I am in a quiet environment where speaking is not appropriate.
But for first drafts, brainstorming, quick messages, and capturing ideas, voice is better in almost every way, and the reason is neurological. When you type, you engage a different cognitive process than when you speak. Typing encourages editing in real time. You write a sentence, delete half of it, rewrite it, then move on. Speaking encourages flow. You express complete thoughts because that is how speech works. Your brain is well suited to this, having done it for a hundred thousand years.
The practical result is that voice to text produces more natural, conversational writing. It captures ideas that might get lost in the editing loop of typing. And it reduces the physical strain of hours at a keyboard. I have spoken to TAWK users who tell me it changed how they write, how they think, and how much they can produce in a day. That is a lasting change in how they work.
Where Voice AI is Heading
Tools like TAWK are an early stage. Here is where I see voice AI going in the next three to five years.
Real Time Translation
Speak in English, and the person on the other end hears your words in Mandarin, in your voice, with your intonation. The individual components already exist. Whisper handles speech to text. LLMs handle translation. Voice synthesis handles text to speech. Connecting them in real time with low latency is an engineering problem more than a research one. When this works reliably, it will remove most of the language barrier.
Voice First Applications
Today, most apps are designed around screens and touch. Voice is bolted on as an afterthought. The next generation of applications will be designed voice first. Think about a project management tool where you say "move the design review to Thursday and tell the team" and it is done. Or a CRM where you debrief after a sales call by speaking and the system automatically updates the contact record, creates follow up tasks, and drafts the next email.
Multimodal Interaction
The most capable interfaces will combine voice with vision. Point your phone at a piece of equipment and say "what is wrong with this?" and get a diagnosis. Look at a dashboard and ask "why did revenue drop last week?" and hear an analysis. This is the convergence of voice AI, computer vision, and language models, and it is closer than most people think.
Ambient Computing
As voice AI improves, the devices around us become more responsive. Your home, your car, your office respond to natural speech without wake words or rigid command structures. You talk, and the environment responds. The interface almost disappears.
The best interface is the one you barely notice. Voice gets closer to that than a screen can.
Why Now Matters
Every major interface shift creates a large wave of new products and companies. The mouse created the desktop software industry. Touch created the mobile app economy. I believe voice AI will create the next equivalent wave, and we are at the very beginning of it.
The builders who understand voice now, who build for it today, will have a large advantage. Voice will not replace screens, but I expect it to become the primary way we interact with AI powered systems. And AI powered systems are becoming the primary way we interact with everything.
When I built TAWK, I built the simplest possible version of this future: speak and your words become text. But that simplicity is what makes it useful every day. The best products in any new paradigm start simple and expand. The iPhone started as a phone and became a computer in your pocket. Voice AI products will follow the same arc.
We spent fifty years adapting ourselves to machines, and I expect the next fifty to go the other way, with voice as a large part of that shift. For anyone building products now, it is one of the best opportunities available.