Computer-use models and autonomous AI agents
Yutori
Yutori builds AI models and agents that operate browsers and computers to complete digital tasks. Its products include Delegate, a personal agent for handing off work, and Scouts, always-on agents that monitor news, prices, reservations, tickets and leads. Developers can build on its technology through APIs. The Navigator model family powers computer use; Navigator n2 extends beyond browser tasks to operate Linux, macOS and Windows desktops, combining visual interaction, command-line operations and code.
Founded in 2024 by Abhishek Das, Devi Parikh and Dhruv Batra, Yutori is led by Das and Parikh as co-CEOs, with Batra serving as chief scientist. The company develops its own models alongside the applications they power. Its Navigator training approach uses agents to generate and test tasks, with verifiers checking the resulting environment rather than accepting an agent’s account of success. Supervised fine-tuning and reinforcement learning develop the model’s ability to navigate applications and complete longer workflows.
Scouts applies this technology to continuous monitoring. Its multi-agent architecture distributes research among specialized agents while an orchestrator filters findings for relevance. Agents search an archive of past reports to contextualize new findings and avoid repeating old information. Users can refine monitoring through feedback on individual findings, whole reports or the Scout’s overall behavior.
2 talks
Newest firstThe Bitter Layout or: How I Learned to Love the Model Picker
AI Engineer World's Fair 2025 · 14:24
2 speakers at AIE
Affiliations reflect their AIE appearances, not necessarily current employment.
Start here
- Computer-use models will agentify the web, not APIs
Start here for practical browser-agent examples, from image-based restaurant menus to procurement documents, and an introduction to evaluation alongside latency and cost per task.
Dhruv BatraAI Engineer World's Fair 2026
- The Bitter Layout or: How I Learned to Love the Model Picker
Choose this talk for a design perspective on the recurring chat-input, turn-by-turn, model-picker layout and the case for reconsidering interfaces around higher-level design goals.
Maximillian PirasAI Engineer World's Fair 2025
Messages from the stage
Rendered state matters for browser agents
Batra explains why text extraction can miss information visible only after a page renders. His demonstrations combine screenshot grounding with JavaScript execution.
Flexibility sustains imperfect interfaces
Piras contrasts criticism of chat usability with widespread adoption, arguing that conversational flexibility helps interfaces accommodate changing models. Model selectors introduce a separate difficulty: troublesome interface modes.
Affiliations reflect each recorded session, not necessarily current employment.
Company sources · checked 2026-08-28
- Yutori — AI agents that autonomously browse and act on the web
- Company — Yutori
- Blog — Yutori
- Yutori achieved 2–3x faster task completion with lower inference costs
- Yutori Launches from Stealth with $15M Seed Funding to Build Consumer AI Assistants Capable of Everyday Tasks on the Web
- Building the proactive, multi-agent architecture powering Scouts
- Navigator n2: frontier computer use at a fraction of the cost

