Innovative AI Assistant Jace Launched by Former Meta Engineers

Zeta Labs, a startup based in London and founded by former Meta engineers Fryderyk Wiatrowski and Peter Albert, has unveiled Jace, a new AI assistant capable of performing in-browser actions independently. The announcement was accompanied by news of a $2.9 million pre-seed funding round, led by Y Combinator's former head of AI Daniel Gross and former GitHub CEO Nat Friedman.

According to Zeta Labs, Jace stands out in the crowded AI agent market by not requiring any user guidance. Users simply specify what needs to be done, and the AI agent handles the rest. This innovation aims to save users from the need to remain in front of their computers for mundane tasks. The startup plans to utilize the pre-seed funds to enhance Jace's capabilities, making it more reliable and efficient in handling complex tasks for both consumers and businesses.

Albert first identified the need for such an AI assistant eight years ago while managing an e-commerce business, where repetitive operational tasks like data transfer were commonplace. With the advent of mature language models in the GPT era, Albert and Wiatrowski collaborated to create Zeta Labs and its flagship product, Jace. The AI agent functions similarly to ChatGPT, where users interact with a chatbot, detailing the tasks that need to be accomplished. Once the instructions are given, Jace's underlying models plan, provide information, and execute actions within the browser.

Jace can perform a wide range of tasks, from simple ones like booking a hotel or replying to emails, to more complex operations such as setting up recruitment pipelines on LinkedIn, managing inventory, and launching ad campaigns. In one notable instance, Jace was even able to establish a company, complete with a business plan, registration, and acquiring its first client. Users can also monitor how the AI agent operates within the browser through a changeable interface.

The capabilities of Jace are driven by a combination of models. One model handles chat-based interactions and action planning, while Zeta Labs' proprietary web-interaction model AWA-1 (Autonomous Web Agent-1) translates these plans into browser actions, effectively navigating the challenges and inconsistencies typical of web interfaces.

"Our core model is based on an open-source framework," explained Wiatrowski. "We applied our dataset to reinforcement learning from AI feedback (RLAIF) and fine-tuned it accordingly." This process involved extensive simulated interactions and synthetic data to ensure the model could handle multi-step web tasks. Jace also incorporates reasoning systems to verify the execution of plans, preventing the agent from getting stuck in loops during tasks with multiple steps.

Security is another key feature of Jace. The AI agent stores user credentials for specific tasks, such as LinkedIn job postings, in an encrypted format akin to password storage solutions.

Although Jace is already capable of handling various tasks, Zeta Labs has not yet monetized the product. The company is refining the AI agent in collaboration with several design partners, aiming for a general release. The next iteration of the AWA model is in development, promising greater speed, size, and the ability to manage more complex tasks, including those requiring visual interaction like working with maps. Most of the pre-seed funding will support these development efforts, alongside some strategic hiring.

Ultimately, Zeta Labs envisions Jace as a valuable tool for consumers and small businesses looking to automate repetitive, browser-based tasks in areas such as recruitment, e-commerce, marketing, and sales. There will be a free plan with limited usage, and a paid subscription model priced at $45 per month.

"On the business side, especially with small businesses, we see a massive demand," Wiatrowski noted. "A great example is recruiters who want to source from LinkedIn and move data to Airtable. Currently, the process is manual. Jace can automate this entire pipeline. You just have to ask."