Exclusive: Former Meta engineers launch Jace, an AI agent that works independently

The founders of Zeta Labs are Fryderyk Wiatrowski and Peter Albert1. Both of them have a tech-savvy background. Fryderyk Wiatrowski is a former Meta engineer, while Peter Albert has experience in e-commerce and also worked at Meta1. They combined their respective backgrounds to drive the initiative of creating Zeta Labs, setting it on a path to redefine the potentials of AI-driven assistance1.

The Jace AI agent, developed by Zeta Labs, is designed to perform tasks in the browser by leveraging a combination of models. At its core, Jace utilizes a regular Large Language Model (LLM) which handles chat-based interactions, captures the required information, and creates a plan of action. The LLM is used to understand the user's instructions and generate a sequence of steps to accomplish the given task.

In addition to the LLM, Jace also employs Zeta Labs' proprietary web-interaction model, AWA-1 (Autonomous Web Agent-1). The AWA-1 model is responsible for converting the plan generated by the LLM into browser actions. This model effectively handles the challenges and inconsistencies commonly found in web interfaces, allowing Jace to interact with websites and perform tasks autonomously.

To achieve its capabilities, Jace leverages extensive simulated interactions and synthetic data for rapid self-improvement. The AI agent learns from these interactions and adapts its behavior to handle a wide array of tasks, ranging from simple activities like making restaurant reservations to complex projects such as managing recruitment pipelines or launching businesses.

Jace operates by understanding the user's task instructions and breaking them down into manageable steps. It then utilizes its internal models and algorithms to execute the necessary actions in the browser, such as searching for information, filling out forms, or navigating through web pages3. The AI agent can perform tasks that require reasoning and decision-making, making it a valuable tool for automating repetitive browser-based tasks.

Overall, the Jace AI agent combines the power of LLMs with its proprietary web-interaction model to perform tasks autonomously in the browser. It offers a user-friendly interface for users to provide task instructions and provides a scalable solution for automating a wide range of tasks.