In a bold move that underscores the shifting landscape of artificial intelligence, three seasoned veterans from Meta have launched a new startup, Yutori, aimed at redefining how people interact with AI personal assistants. On March 27, 2025, Yutori announced a $15 million funding round led by Rob Toews at Radical Ventures, with participation from top-tier investors like Felicis, Fei-Fei Li, and Jeff Dean, chief scientist at Google DeepMind.
Yutori: Built by AI Pioneers
The co-founders—Abhishek Das, Devi Parikh, and Dhruv Batra—bring a formidable pedigree. Parikh previously led multimodal AI research at Meta, focusing on systems that understand and process different types of data like text, images, and video. Batra, on the other hand, led Meta’s embodied AI research, working on models that could interact with and understand the 3D physical world. Das has extensive experience in AI systems and product development.
Rather than taking the typical chatbot route, Yutori wants to go deeper. The team doesn’t intend to create another assistant that merely responds to questions. They aim to build an autonomous agent—an AI system that can independently carry out complex, multi-step tasks on behalf of users.
Moving Beyond Chatbots
In an interview, Parikh pointed out the limitations of current AI tools. “Right now, there’s a lot happening with chatbots,” she said, “but they’re not doing things for you in a way that can take work off your plate.”
Yutori focuses on automating real-world tasks such as ordering food online, booking complex travel arrangements, and managing personal schedules. These are tasks people typically handle manually, jumping between apps and filling out forms. Yutori wants to eliminate that friction, offering a system that performs these jobs with minimal oversight.
Why the Timing Matters
AI has reached an inflection point. Thanks to advancements in large language models (LLMs) and machine learning, systems can now handle long sequences of actions with accuracy and reliability. Executives across the tech world, including OpenAI’s CFO Sarah Friar, have emphasized that autonomous agents represent the next wave of AI innovation.
Yutori sees an opportunity in this momentum. Rather than relying solely on foundational models like GPT or LLaMA, the startup emphasizes post-training—a method where AI models undergo fine-tuning after their initial large-scale training. This technique helps them become more specialized and effective for real-world use cases.
Focus on Post-Training and Multimodal Systems
Yutori doesn’t start from scratch. The founders and early team members played key roles in post-training the LLaMA 3 and 4 models—Meta’s flagship open-source language models. Post-training allows developers to refine a model’s behavior using targeted data, resulting in improved performance for specific tasks like web navigation, scheduling, or decision-making.
This emphasis on multimodal post-training gives Yutori an edge. By building assistants that can understand and act on text, visual, and contextual data, the team can create AI agents capable of interacting with today’s complex digital environments.
For instance, a user could ask Yutori to “book a flight to Tokyo next week, but only if there’s a window seat in business class and the total travel time is under 14 hours.” Yutori’s assistant would not only search for flights but filter them intelligently, make the booking, and update the user’s calendar—all autonomously.
Industry Leaders Back the Vision
Yutori’s seed funding round reflects widespread confidence in its vision. Rob Toews of Radical Ventures, known for backing cutting-edge AI startups, led the funding. The round also attracted Fei-Fei Li, a leading voice in AI and Stanford professor often dubbed the “godmother of AI.” Her endorsement alone signals Yutori’s potential impact.
Jeff Dean, a titan in the AI world and chief scientist at DeepMind, also joined the investor list. His support hints at the broader industry belief that next-gen AI assistants will shape how humans live and work in the near future.
Competitive Landscape and Differentiation
Yutori enters a competitive space. Companies like OpenAI, Anthropic, and Adept have made similar moves toward AI agents. However, Yutori stands out through its deep integration of multimodal AI, post-training optimization, and embodied cognition.
Where most rivals still treat the assistant as a glorified search engine or basic task executor, Yutori approaches it as a context-aware, interactive entity. Their system learns user preferences over time, anticipates needs, and executes tasks proactively.
Instead of asking users to fill out long forms for travel or deliveries, the assistant might suggest what needs to be done next or act without prompting—if permissions allow.
Real-World Use Cases: Beyond Convenience
Yutori doesn’t just focus on convenience. The startup aims to unlock productivity in both personal and professional settings. For instance, an executive assistant bot could manage meetings, coordinate travel across departments, order office supplies, and even evaluate vendor reviews.
In daily life, users could rely on Yutori for home management—automating groceries, scheduling kids’ classes, or organizing repair services based on smart home feedback.
The team believes that AI agents should feel less like software and more like a helpful colleague—a shift that could fundamentally reshape digital interactions.
What’s Next for Yutori
With $15 million in hand, Yutori plans to scale its engineering team, expand product development, and begin private beta testing later this year. The company will also continue to refine its post-training techniques and integrate more real-time data capabilities.
Parikh and Batra have made it clear: they want to change how people use the internet. Not just to search or chat, but to get things done—without lifting a finger.
Final Thoughts
Yutori represents more than just another AI startup. It symbolizes a broader movement within the tech community: the evolution of artificial intelligence from reactive tools to proactive agents.
By harnessing the power of multimodal systems and post-training optimization, the Yutori team is building a smarter, more helpful AI—one that doesn’t just talk, but acts.
The road ahead will bring challenges, from data privacy to trust-building with users. But if anyone has the experience and ambition to navigate this space, it’s the team behind Yutori. Their journey could very well mark the beginning of a new era in intelligent automation—one where AI doesn’t just assist, but anticipates.