Hark Unveils Game-Changing AI Browser Agent for Automated Tasks
Hark, a startup with $700 million in Series A funding, has unveiled Hark Handoff, a new browser-based AI agent capable of efficiently completing tasks on diverse websites, including those without official APIs. Handoff leverages visual data to navigate and act, positioning itself as a faster and more cost-effective alternative to competitors. The company is currently accepting waitlist sign-ups ahead of its summer release.
Hark, a burgeoning startup that successfully secured $700 million in Series A funding in May, has officially launched its innovative agent, Hark Handoff. This new agent is designed to efficiently complete a variety of tasks by leveraging a web browser, even on websites that lack official Application Programming Interfaces (APIs). Hark claims that Handoff possesses the capability to seamlessly navigate complex online platforms, including major retailers like Target and Walmart, as well as service providers such as OpenTable and LinkedIn.
The core functionality of Hark Handoff relies on its ability to analyze website structure and visual data, enabling it to discern when and where to perform actions like clicking buttons or inputting information. This operational premise is consistent with numerous other browser-based agents that have been previously introduced. Users can issue commands to Handoff, and it will execute diverse tasks, ranging from ordering food or coffee, booking travel tickets, filing returns, and shopping for essential items, to reserving tables at restaurants or conducting research by synthesizing information from multiple online sources.
During a video demonstration, Brett Adcock, Hark’s CEO, showcased the agent's capabilities by illustrating how it could fulfill a command to assemble a flower bouquet. Notably, Handoff demonstrated flexibility in accommodating 'fuzzy' terms, such as including 'some of the florist’s choice' within the user's specifications. However, the company acknowledged that the video only presented a partial view of the process, making it difficult to fully assess its comprehensive effectiveness.
For its initial release, Hark Handoff utilizes a post-trained model, with plans to transition to a pre-trained model later in the year. This strategic approach allows Hark to accelerate the refinement of its data pipeline, training infrastructure, and various techniques. A key differentiator highlighted by Hark is that its model predicts the 'next action' – such as a click or a keyboard input at a precise location – in contrast to large language models (LLMs) that typically predict the 'next token'.
The market for computer-use agents is a competitive landscape, with established players like Google, OpenAI, and Anthropic actively developing solutions. Additionally, numerous VC-funded startups, including Browser Use, Polar, Strawberry, and Aside, are focusing on browser-based task automation. Hark asserts that its Handoff agent offers superior speed and significantly lower operational costs compared to rival models like GPT 5.5 and Opus 4.8. The startup has commenced a waitlist for its platform, with a full release anticipated by the end of the summer.