Hark, an AI startup that raised $700 million in Series A funding in May 2026, launched its browser-based agent called Hark Handoff on August 5, 2026. The tool is designed to complete tasks on behalf of users by navigating websites directly, without relying on official APIs.
According to the company, Handoff can operate across sites including Target, Walmart, OpenTable, and LinkedIn. It does so by analyzing website structure and visual data to determine when to click buttons or enter text. Supported tasks include ordering food or coffee, booking travel, filing returns, shopping, making restaurant reservations, and conducting research across multiple sources.
In a video demo, CEO Brett Adcock showed the agent handling a command to build a flower bouquet, including accommodating vague instructions such as “some of the florist’s choice.” The demo shows only part of the process, leaving the agent’s full effectiveness difficult to assess.
Hark says Handoff currently runs on a post-trained model, with plans to move to pre-training later in 2026. The company says this approach allows it to refine its data pipeline, training infrastructure, and techniques more quickly. Hark also distinguishes its model from standard large language models by claiming it predicts the next action — such as a click or a keyboard input at a specific location — rather than the next token.
The company states that Handoff is faster and cheaper than competing products, citing GPT 5.5 and Opus 4.8 as cost comparisons. Hark is not alone in this space; Google, OpenAI, Anthropic, and startups including Browser Use, Polar, Strawberry, and Aside are also working on computer-use or browser-based agents.
Hark has opened a waitlist for Handoff and plans to release the platform publicly by the end of summer 2026.
Source: TechCrunch