Hark, the startup that raised $700 million in Series A funding in May, announced Hark Handoff, an agent designed to perform tasks by operating a web browser. The company says Handoff can navigate sites that lack official APIs, including Target, Walmart, OpenTable, and LinkedIn.
How it works
According to Hark, the Handoff agent inspects page structure and visual cues to determine whether to click buttons or type information. Users issue commands and the agent carries out tasks such as ordering food or coffee, booking travel tickets, filing returns, shopping for essentials, reserving a restaurant table, or conducting research across multiple sources.
Demo and limitations
In a demo video, Hark CEO Brett Adcock showed the assistant assembling a bouquet based on user-specified flowers and handling fuzzy instructions like “the florist’s choice.” The company’s video covers only part of the process, so the demonstration does not allow external observers to fully assess real-world effectiveness.
Model approach and roadmap
For this release, Hark said it is using a post-trained model and plans to move to pre-training later this year. The company argues that this approach lets it more rapidly refine its data pipeline, training infrastructure, and techniques. Hark also claims that, unlike large language models (LLMs) which predict the next token, its model predicts the next action — for example, performing a click or a keyboard input at a specific location.
Competition and pricing claims
Multiple companies are developing computer-use agents, including Google, OpenAI, and Anthropic, as well as venture-backed startups focused on browser-based task automation such as Browser Use, Polar, Strawberry, and Aside. Hark says Handoff is faster than competitors and significantly cheaper than models like GPT 5.5 and Opus 4.8.
Availability
Hark has opened a waitlist for the platform and plans to release the service by the end of the summer.
