Ex-Meta Leaders Launch Yutori Following $15M Raise To Build AI Agents That Act as ‘Chief of Staff’ 

Key Takeaways

• A group of former Meta executives have founded Yutori.
• Yutori’s aim is to build a “multi-agent system” that acts as an “AI chief of staff.”
• Startups increasingly use multiple AI models from different providers to optimize their products.

Three former Meta employees who worked across the company’s AI portfolio have raised $15 million to found Yutori.

In a press release on Thursday, March 27, the new startup announced it will focus on building AI assistants that can act as a “chief of staff” for online tasks.

Enhancing AI Usefulness

Yutori was founded by three former Meta AI researchers: Abhishek Das, Devi Parikh, and Dhruv Batra.

“Yutori was founded to build personal AI assistants that are so reliable that they become the primary drivers of action on the web,” Thursday’s release stated.

To achieve this end, Yutori’s founders are taking a full-stack approach to AI development.

“To raise both the floor and ceiling of human productivity on the web, you need to innovate across the stack in a coherent and tightly coupled way,” explained Co-Founder Abhishek Das.

“You can’t be focused just on the AI models or just on the orchestration or application layers,” he added.

Beyond Single-Model AI

While Yutori hasn’t revealed which models it plans to use for its AI agent platform, startups increasingly leverage a competitive field of available options.

Rather than building everything from scratch, the company will follow in the footsteps of the viral Chinese AI agent Manus, which deploys Anthropic’s Claude 3.5 Sonnet and fine-tuned versions of Alibaba’s Qwen.

Described by Yutori as a “multi-agent system,” this approach has the advantage of pulling together the best solutions from different providers. For example, the platform may select one model for web navigation, one for translation and another for reasoning.

Open-Source AI for Enhanced Functionality

Despite promising the technology for over a year now, OpenAI and other major AI developers have been slow to deliver web-browsing capabilities.

Even advanced platforms like OpenAI Deep Research occasionally trip up on simple tasks like downloading documents.

On their own, the open-source models Yutori plans to work with aren’t much better. But the startup said it plans to use various post-training techniques to improve their agentic functionality.