Hark Previews Handoff Browser Agent After $700M Series A

Summary

On August 5, 2026, Hark announced Hark Handoff, an AI agent designed to carry out tasks through a web browser. The startup, which raised $700 million in Series A financing in May, says Handoff can operate websites even when they do not provide official APIs. Hark has opened a waitlist and says the platform will become available by the end of the summer. Because the product is not yet generally available, the verifiable development is its announcement and preview; its performance claims have not been independently demonstrated in the article.

Handoff is intended to interpret both the structure and visual presentation of a webpage, then decide where to click or enter information. Hark specifically named Target, Walmart, OpenTable and LinkedIn as sites the agent can navigate. Proposed uses include ordering food or coffee, buying household essentials, booking travel, filing returns, reserving restaurant tables and researching information across multiple sources. These functions place Handoff in the broader category of “computer-use” agents, which attempt to complete multi-step workflows rather than merely answer questions.

CEO Brett Adcock demonstrated a prompt asking the agent to assemble a bouquet from specified flowers while also handling an imprecise instruction such as including selections chosen by the florist. That example suggests Hark wants the system to combine explicit constraints with discretionary requests. However, TechCrunch noted that the video showed only part of the transaction. It therefore did not establish whether Handoff could reliably complete the full workflow, recover from errors or handle unexpected changes on the website.

Hark says this release uses a post-trained model and that it intends to begin pre-training later in 2026. The company argues that its present approach lets it improve its data pipeline, training systems and methods more quickly. It also characterizes the model as predicting the next interface action, such as a mouse click or keyboard input at a particular location, rather than predicting the next text token in the manner associated with conventional large language models. The article does not provide technical documentation or independent testing to validate that distinction.

The competitive field is already crowded. Google, OpenAI and Anthropic are developing computer-use technology, while venture-backed companies including Browser Use, Polar, Strawberry and Aside are pursuing browser automation. Hark claims Handoff is faster and substantially cheaper than models such as GPT 5.5 and Opus 4.8, but it disclosed no benchmark results, pricing figures or comparison methodology in the article. If the claims hold, Handoff could make browser automation useful on services that lack integrations, affecting consumers and businesses with repetitive online workflows. The next milestone is the promised end-of-summer release; reliability, access terms, pricing and real-world performance remain uncertain.