Former Meta Engineers Launch Jace - An Independent AI Agent

Zeta Labs, a startup based in London and founded by former Meta engineers Fryderyk Wiatrowski and Peter Albert, has launched Jace, a large language model (LLM)-powered AI agent capable of executing in-browser actions without user intervention.

Funding and Leadership

Zeta Labs recently secured $2.9 million in pre-seed funding, led by Daniel Gross, former head of AI at Y Combinator, and Nat Friedman, former CEO of GitHub. Several other notable investors, including Shawn Wang, Bartek Pucek, and Mati Staniszewski, founder of ElevenLabs, also contributed to the round.

Autonomous AI Agents

The AI landscape has seen various autonomous agents, with Cognition’s Devin being a notable example. However, Zeta Labs claims Jace stands out due to its complete independence, requiring no user guidance. Users simply instruct Jace on tasks, and it takes over, eliminating the need for constant user input.

Currently, Zeta Labs is working with early partners to refine Jace’s capabilities, aiming to make it more reliable and efficient for complex consumer and business tasks.

Capabilities of Jace

Peter Albert initially conceived the idea of an AI agent eight years ago while working in ecommerce, where repetitive operational tasks were common. With advancements in AI, Albert collaborated with Fryderyk Wiatrowski to develop Jace at Zeta Labs.

Jace functions similarly to other web agents, such as ChatGPT. Users provide task instructions through a chat interface, and Jace’s underlying models create a plan, gather information, and take action within the browser.

Tasks Jace can perform include:

• Booking hotels and handling payments

• Searching for flights

• Responding to emails

• Setting up recruitment pipelines on LinkedIn

• Managing inventory

• Launching advertising campaigns

In one case, Jace was even capable of creating a company, complete with a business plan, registration, and initial client acquisition.

Technology Behind Jace

Jace utilizes a combination of models to operate effectively. One model handles chat-based interactions, while Zeta Labs’ proprietary AWA-1 (Autonomous Web Agent-1) model converts these plans into browser actions. This approach ensures the AI agent can manage the complexities of web interfaces.

“Our core model is based on an open-source model, enhanced through reinforcement learning from AI feedback (RLAIF),” explained Wiatrowski. Extensive simulated interactions and synthetic data ensure Jace can handle multi-step tasks efficiently.

To prevent common issues with web agents, such as looping during complex tasks, Jace incorporates reasoning systems to verify task completion. Additionally, user credentials are securely stored in an encrypted format.

Future Development and Monetization

Zeta Labs is refining Jace with design partners, aiming to prepare the AI agent for a broader release. The next iteration of the AWA model will be larger, faster, and better equipped to handle complex tasks, including those requiring visual interactions.

Most of the pre-seed funding will support these advancements and hiring efforts.

Market Potential

Zeta Labs envisions Jace as a valuable tool for consumers and small businesses, automating repetitive browser-based tasks in sectors such as recruiting, ecommerce, marketing, and sales. The company plans to offer a free plan with limited messages, with a subscription fee of $45 per month for additional usage.

“There is significant demand among small businesses,” noted Wiatrowski. For instance, recruiters currently perform manual data sourcing and management tasks that Jace can automate entirely. By streamlining such processes, Jace aims to save time and increase efficiency.