Lexy.fm
A voice-first media platform, built by a solo founder and a newsroom of agents
There are roughly three billion people who can make a phone call but can't download an app. They have phones, just not smartphones. They have minutes, just not data plans. Every voice platform of the past decade assumed WiFi, touchscreens, and app stores, and left that audience out of the conversation. The phone call remains the most universal interface ever built. It works on every handset made in the last thirty years, without literacy, without bandwidth, without an account. Lexy is built on the conviction that the best interface is the one that was always there. You just have to dial it.
I've been circling this idea for twenty years. In 2004 I built Yeowsa, an audio publishing platform for feature phones, which became Syndout, where readers could call a number and hear articles read aloud. At Al Jazeera I built IVR applications for markets where phones were everywhere but internet wasn't. Ghana Voices launched during the 2012 elections and drew a thousand listeners and 428 citizen journalist voice reports in its first week. Kenya Voices followed with six thousand listeners. In 2017 my World Quiz game won the Actions on Google developer challenge in two categories, and then Google shut the platform down and years of work disappeared with a decision made in a room I'd never enter. That experience settled it. The telephone network is the only voice platform no one can shut down. Lexy is the assembly of every one of those experiments, finally built in an era where the AI, the voices, and the browser could keep up.
The proof point was trivia. Lexy hosts a marketplace of more than 2,000 multilingual voice games fronted by Lexy Qulha, an AI personality who speaks English, Spanish, Turkish, and Arabic. You call a number, she asks questions, you answer by speaking. The same experience runs entirely in the browser, where the page captures your voice, transcribes it, and scores your answer with AI in real time, with nothing to install. Building it taught me that invisible design is harder than visible design. Every job a screen would do, the confirmation, the encouragement, the sense of momentum, has to be carried by tone, pacing, and word choice instead. What looks effortless requires the most effort.
The newsroom without the newsroom
Trivia proved the interface. News revealed the platform. At TRT World in 2017 I ran a daily news brief and quiz that took a fifteen-person team six hours to produce one edition in one language. Lexy now produces the same product as a virtual newsroom of AI agents working against a written editorial standard. A scout searches the local press of 48 countries across the Global South, translating at ingest so a story from a Chadian or Nepali outlet competes on equal footing. An editor selects three stories against criteria for consequence, diversity, and originality, surfacing news from the places the headlines skip. A writer works only from verified facts, a fact-checking pass validates every claim against the source articles, and a deterministic lint gate blocks anything off-standard from publishing. One editorial act fans out into twelve briefs a day, three editions in four languages, for less compute cost per month than one hour of the old team's payroll.
The hardest role to automate was never the writer or the narrator. It was the executive producer, the person whose judgment becomes the team's operating system. Lexy encodes that judgment in an editorial architecture document, and the agents execute against it the way a well-trained newsroom executes against an EP's vision. That's the difference between automation and replacement. Automation removes a task. Replacement encodes the judgment that made the role necessary.
Create once, publish everywhere
The pipeline never treats a finished MP3 as the product. Each edition is a structured object, verified facts, locked scripts, narration clips with word-level timestamps, quiz questions with accepted answers. Every distribution surface is just a rendering of it. The same edition plays as an embeddable widget on publisher pages, answers the phone line, syncs a live call to a browser with karaoke-style highlighting, launches from a QR code on a printed page, ships as RSS and a podcast feed, and runs as an interactive Alexa skill in three languages. Adding Alexa took days, not months, because it consumed assets the newsroom already produced. Each new surface is cheaper than the last.
That architecture is also the business. Publishers embed Lexy to turn a finished read into one more interaction, a narrated game on the story the reader just finished. Brands get participation instead of exposure, a campaign message people answer out loud rather than scroll past. And Lexy Originals gives publishers a daily multilingual news product, newsroom-grade coverage without a newsroom's payroll.
Lexy is what happens when twenty years of voice-first conviction meets the moment when one person can orchestrate the work of an organization. The domain expertise didn't get replaced by the AI. It got concentrated into the two things that matter, deciding what to build and encoding the judgment that ensures it gets built right. The goal hasn't changed since the feature phone era. Make voice interaction a utility rather than a luxury, and bring the spoken web to the people the app economy left behind.