Hark Launches Browser Agent Handoff for Task Automation

Hark, a startup that secured $700 million in Series A funding in May, has introduced its agent, Hark Handoff, capable of completing tasks via a browser. The company asserts that Handoff can navigate sites without official APIs, such as Target, Walmart, OpenTable, and LinkedIn, by analyzing their structure and visual data to perform actions like clicking buttons or entering information. The agent can handle tasks including ordering food, booking travel, returning items, shopping, reserving tables, and conducting research. In a demonstration, CEO Brett Adcock showcased the assistant creating a bouquet based on user preferences, including vague requests like 'some of the florist's choice.' Hark employs a post-trained model for this version and aims to pre-train later this year to enhance its data pipeline and training framework. Unlike LLMs that predict subsequent tokens, Hark's model anticipates the next action, such as a click or keystroke. The company claims Handoff is more efficient and cost-effective than rivals like GPT 5.5 and Opus 4.8. A waitlist has been established, with plans to launch the agent by summer's end. The browser-use agent market features competitors including Google, OpenAI, Anthropic, and startups like Browser Use, Polar, Strawberry, and Aside.