Datost Bet on the Answer. Then It Bet on the Glasses You Read It In.

The clearest way to understand Datost is to watch what it did after it won. In early 2026, the two-person startup shipped an AI data analyst that lived inside Slack. On BIRD-Interact, a text-to-SQL benchmark built to be nasty - 600 deliberately ambiguous questions across 22 messy Postgres databases - it scored 75.2%. A leading frontier model, working alone, scored 33%. That is the kind of number you put on a slide and ride for a year.

Datost did not ride it. By mid-year, the company had reframed itself as an applied research lab and pointed the same two engineers at a different question: not "how do we answer faster," but "why are you still staring at a screen to read the answer." The result is ED1TH - a 25-gram pair of AI glasses with no cameras, a display hidden in the lens, and a small steel ring to control it. Same founders. Same YC batch. A very different bet.

Two engineers who kept following the same instinct

Datost was founded by Maceo Cardinale Kwik, its CEO, and Jason Wang. Kwik studied computer science at Virginia Tech, graduating summa cum laude, and worked at SimonComputing before joining Traba as a software engineer. There he built a data-analyst agent that generated research reports inside Slack - essentially a prototype of what Datost first shipped. Wang studied computer and software engineering with an AI focus at the University of Toronto, then spent more than two years at Confluent working on the Flink control plane, the plumbing behind large-scale streaming data, before also landing at Traba.

So the pair share a resume line and a habit: build the tool you wish existed at your day job, watch it work, then leave to build the real version. It explains the first product. It also explains the willingness to walk away from it.

"We design the object, optics, software, and intelligence as one behavior - then remove everything that asks for attention without earning it." Datost, on its design approach

A data analyst that knew what your questions meant

The original Datost was a Slack-native AI data analyst. You @mentioned it the way you'd ping a colleague, and it answered in plain English, with the query it ran attached. The trick was not raw model horsepower. It was a semantic layer: a running model of what your company's words actually mean. When someone asks about "active users" or "churn risk," those terms rarely map cleanly onto a database schema. Datost read Slack history, docs, the CRM, and even the codebase to learn how a metric was really calculated before writing a line of SQL.

It connected to the usual warehouses - Postgres, MySQL, BigQuery, Snowflake, Databricks - and to tools like Datadog, Sentry, PostHog, Notion, and GitHub through MCP. A second model checked answers before they went out, and corrections were remembered so the same mistake did not repeat. On the analytical slice of the benchmark, it reached about 91%.

Why walk away from a number you're winning on

Benchmarks win demos. They do not always win markets. Datost's read was that faster answers still left people tethered to a rectangle of glass. The company's public thesis became a single sentence you can repeat: it is building "the computers that come after the screen." That is a claim you can agree with or roll your eyes at, but it is compressible, which is more than most startups can say about their roadmap.

ED1TH: Jarvis, minus the recording light

ED1TH is the lab's first public system - a personal AI that lives in a pair of glasses and, per the company, reads your email, runs your calendar, and puts directions in your eye. The most striking spec is the one that is absent: no cameras. In a category racing to strap lenses onto your face, leaving them off is a product decision dressed as a trust decision. It means you can wear the glasses at dinner and nobody has to flinch.

The rest of the hardware is built around a single obsession: weight. Smart glasses fail when they are heavy, hot, and taken off within the hour. At 25 grams, ED1TH is lighter than a AA battery. The display is a green Micro-LED tucked into the lens, running 640x480 at 60Hz and pushing 2,000 nits so it stays legible in direct sun. Power comes from hot-swap cells in the temples, backed by a charging case. And instead of tapping the frame, you control it with a companion steel ring.

Hardware up front, software over time

The model is the familiar device-plus-subscription shape. ED1TH reserves for a $50 refundable deposit against a launch price around $500, with prescription lenses as a paid add-on. The AI runs on tiers - a free level, with paid plans starting near $29 a month. A refundable deposit is a small confidence signal: it only works if you believe buyers will keep the thing rather than ask for their money back.

"A personal AI that lives in a pair of glasses. 25 grams, no cameras, all day." ED1TH, in one line

The camera-free corner of a crowded market

Smart glasses are no longer a fringe idea - Meta's Ray-Ban Display, Even Realities, Brilliant Labs, XREAL, and Rokid are all pushing on the same face-worn future, and Apple looms behind the category. Datost's wager is that the winning version is the discreet one: light enough to forget, private enough to trust, useful enough to keep on. That is a narrow lane, but a real one. And it is backed by an unusual pair of names for a two-person shop - Y Combinator, whose Spring 2026 (P26) batch it joined with David Lieb as partner, and OpenAI's startup program. A reported $130,000 seed landed around March 2026.

Whether ED1TH ships on its promises is the open question. What is already clear is the company's instinct: when a product is winning on the wrong axis, change the axis. Datost has done it once in public. The glasses are the second act.