The Agentic Data Co.
Specialized audio datasets for training speech models.
The founders
MV
The company
The Agentic Data Co. designs and records high-quality conversational audio datasets to train speech models, capturing not just words but identity, emotion, intent, hesitation, turn-taking and the acoustic texture of a room. Every hour is consented, licensed and paid for at the source.
Voice interfaces are moving from novelty to default input for the next wave of computing, and frontier speech models are bottlenecked by realistic, messy, multilingual conversational data rather than clean studio reads.
Founded in San Francisco in 2026 by Christian Vestergaard and Matias Voldby Drejer, the company is tackling the difficult data problems that speech-model teams keep running into.
The founding team is building for frontier AI and speech-model teams and is part of Y Combinator's Fall 2026 batch, backing a bet that the ethics and provenance of audio data will matter as much as its volume.
Who else is building in this space
Companies working on the same thing, at every stage.
AI agent that models blast radius and gates infrastructure change safely
Real-time AI personalisation engine for modern commerce
Agentic voice AI for consumer research insights
AI sales agent for live, multilingual software demos
AI growth platform for professional service firms
AI video engagement layer for websites and products
Thousands more founded every month.
Dealroom tracks new companies as they appear — founders, funding and signals, over the API or in the platform.