Expert computer-use data for frontier AI labs
HQ: San Francisco, CA, United States | Founded: 2026 | Employees: 2 | Stage: Seed (Y Combinator S26) | Website: https://www.markovstudios.com/
Markov builds and sells training data for AI that uses computers the way people do. It records experts doing real work — capturing the screen, every mouse movement and keystroke, and notes explaining what is happening — and sells those recordings to the large AI labs training computer-use models. Its focus is hard-to-find material: professional software used in design, architecture, engineering, accounting and finance, where the limiting factor is finding someone qualified to record rather than the recording itself. Contributors are vetted rather than crowdsourced. The company says it has sold more than 33,000 hours of data to frontier AI labs, and that its free public datasets have passed 200,000 downloads on Hugging Face, including what it describes as the largest openly available collection of computer-use screen recordings. Earlier launch material cited 15,000 hours sold and 150,000 downloads, so both numbers have grown. Founded in 2026 in San Francisco by chief executive Dev Mandal, previously at Sarvam with a degree in aerospace engineering from IIT Madras, and chief technology officer Harish Ashok, who previously founded a robotics company making AI-assisted hardware workflow tools.
AI can reason well but still fails at long, multi-step jobs in real desktop software, and the bottleneck is training data rather than model quality. Markov sells exactly that: recordings of people doing real work, paired with their mouse and keyboard actions and written notes. Its edge is expertise — anyone can collect clicks through consumer apps, but few can find qualified people to record design, architecture, engineering and finance work. More than 33,000 hours sold to frontier labs is strong validation at this stage. The weakness is the business model: this is a labour business with thin margins and expensive contributors, and the large labs are building in-house data teams.
Browse: Home | Blog | About | All Startups A-Z | View full profile on StartupWiki