Toko.

It's late in Taipei, and someone is practicing a job interview with nobody

The phone is propped against a mug of cooling tea. The apartment is quiet. And a voice - patient, unbothered, a little uncanny - asks, "So, tell me about yourself." The person answering has read thousands of English sentences and spoken almost none. Tonight that changes, one stumble at a time, because the thing listening will not laugh, will not sigh, and will happily hear the same wobbly sentence four times until it lands.

That thing is Toko. It is an AI English tutor built for the enormous, mostly silent population of people who can decode a paragraph but seize up when asked to speak. Founded in 2021 by two MIT engineers and launched into Y Combinator's Winter 2022 batch, Toko made a deceptively simple wager: reading English and speaking English are different sports, and almost every language app trains the wrong one.

The premise sounds obvious once said out loud, which is the mark of most good ideas. A human tutor is the gold standard for speaking practice. A human tutor is also expensive, scheduled, and - this is the part the founders understood - a little terrifying. Nobody wants to fumble in front of a stranger they are paying by the hour. So Toko removed the stranger and kept the practice.

Short conversations. Instant fixes. No word limits.

Open the app and you pick a scenario - ordering coffee, a work stand-up, small talk at a wedding, a TOEFL prompt. Then you just talk. Toko replies like a person would, and quietly flags the grammar and phrasing you got wrong, in the moment, without stopping the flow. Sessions run five to ten minutes, which is roughly the attention span of a tired adult after work.

Two engineers who'd already shipped learning to millions

Toko isn't a first rodeo. Both founders came out of MIT and out of companies where teaching-at-scale was the day job - which is why the product feels less like a demo and more like something that quietly knows what it's doing.

Seed-stage, and deliberately lean

Figures vary across public trackers; treat the exact total as approximate. What's clear is the shape: a small, engineering-heavy team that stretched a modest seed into seven-figure download numbers.

Why "just talk" beats "just tap"

Most language apps optimize for streaks and vocabulary recall. Toko optimizes for the thing learners actually panic about: opening their mouth. Here's the gap it aims at.

Illustrative - not survey data. The point is directional: the demand Toko serves is the space between how much English people can read and how little they'll risk saying.

A short, fast timeline

The tea has gone cold. The interview went fine.

By the fourth attempt, "tell me about yourself" no longer triggers a flinch. The sentences come out crooked, then straighter, then almost easy. Nobody witnessed the fumbling except an AI that has already forgotten it. Tomorrow the same person will say the same words to a real interviewer, and the room won't feel quite so large.

That's the whole trick, and it's smaller than it sounds. Toko didn't invent conversation or feedback. It removed the audience from practice and left only the practice - so that the first time you're brave in English isn't the first time it counts. A million cold mugs of tea later, that quiet late-night apartment has a lot of company.