Artificial intelligence research startup Zeta AI Inc., better known as Zeta Labs, said today it has closed on $2.9 million in pre-seed funding to accelerate the development of its autonomous generative AI agent, called JACE.

Today’s round was led by Daniel Gross, former head of AI at Y Combinator, and former GitHub Inc. Chief Executive Nat Friedman. Earlybird VC, Kaya VC and a number of angel investors also participated in the round.

Zeta Labs announced the round alongside the debut of its “groundbreaking” digital assistant JACE, which some believe represents the next evolution of AI agents. It far exceeds the capabilities of existing chatbots such as ChatGPT because it does much more than just generate text.

Instead, it’s focused more on taking action, with an ability to use a web browser just as a human can and perform a wide array of online tasks. Whether it’s finding an apartment online, shopping for groceries or even registering a company, Zeta Labs claims JACE has it covered.

The startup says JACE can do this thanks to its cognitive architecture, which it says is far more complex than standard chatbots. Its secret sauce is its ability to control every aspect of the web browser, allowing it to perform all of the same tasks as a human could using that tool. JACE can do this because it has been trained using Zeta Labs’ proprietary web-interaction large language model, called Autonomous Web Agent-1, or AWA-1.

Zeta Labs co-founder Peter Albert said the company’s aim is to enable AI models to interact in the digital world they live in.

“The development of our AWA-1 model allows JACE the ability to control a browser, essentially giving the AI assistant the digital equivalent of arms and legs,” he explained. “You can’t ask any current chatbot on the market to book a trip, pay an invoice or set up a job post. With JACE, you can.”

Although some of the more recent AI agents have impressed with their ability to do things such as writing code – such as Cognition’s Devin – the difference with JACE is that it doesn’t require any guidance at all. As a result, users won’t have to even sit in front of a computer to keep an eye on it as it goes about completing tasks on their behalf. All they have to do is tell it what must be done.

The range of tasks JACE is capable of is impressive. Zeta Labs said JACE showed it was able to create a company fully autonomously. Beginning with a simple prompt, the agent went ahead and created a business plan for a math tutoring company, registered it officially, found its first client and began making revenue, all within just one week.

The team says JACE can emulate a “substantial portion” of day-to-day office work and to prove this, it underwent a number of benchmark tests focused on common web-based tasks, such as ordering a pizza and finding two apartments in London that meet specific criteria. It showed an impressive performance, delivering an 89% task completion success rate, far ahead of GPT-4o’s 68% success rate.

Albert explained that he was inspired to create JACE after his experience when he launched his first e-commerce business eight years ago, and realized the extensive amount of time his staff spent on repetitive and mundane tasks. Back then, he wished for a way to automate that work, so his staff could focus on more interesting and creative tasks. JACE is the result of that wish.

Zeta Labs’ other co-founder, Fryderyk Wiatrowski, describes JACE as a “meta-aggregator” for web interfaces. “Why learn to navigate all the different user interfaces when you can have a universal UI for everything?” he said. “With JACE, you simply say what you need, and if you forget, it will ask you all the necessary questions up front. Imagine a universal service desk with a consultant who knows you personally and never needs to sleep. JACE will be that for you, just sitting there, ready to serve you at all times.”

Lead investor Gross said the web browser is a legacy technology that was invented 34 years ago, yet today’s AI models still can’t use it. “The next paradigm shift in AI is unlocking that behavior, and Zeta Labs is poised to make that happen,” he said.

According to Holger Mueller, an analyst with Constellation Research Inc., Zeta Lab’s JACE proves that generative AI’s destiny lies not just in content summarization and generation, but in taking actions in the real world. “Zeta Labs has demonstrated this in an impressive way with its first prototype, JACE, operating a company it created by itself,” Mueller said. “The challenge now is to scale JACE’s services, and show it can create repetitive and reliable value.”

Zeta Labs said it is already working with some early partners, and will use the money from today’s round to improve JACE’s capabilities, with plans to improve its speed and reliability. It will also seek to enable JACE to perform even more complex tasks that consumers and businesses are likely to need doing. To do this, it’s building a new iteration of the AWA model, which will be much larger and faster, and better at handling tasks that require visual recognition, for example interacting with maps, the company said.

Eventually, Zeta Labs hopes to commercialize JACE by packaging it as a kind of virtual assistant for businesses needing to automate repetitive, browser-based tasks such as e-commerce, marketing, sales and recruitment. It will also offer a consumer version, with a free plan that places limits on the number of messages per month. For those who want unlimited access, the startup will offer a subscription, likely priced at about $10 per month.

Image: Zeta Labs