Hark previews a browser-use AI agent for everyday tasks

Hark, a startup that raised $700 million in a Series A round in May, today launched an AI agent called Hark Handoff that it says can navigate a browser to complete tasks on its own. The company claims Handoff can navigate websites with no official APIs, including Target, Walmart, OpenTable, and LinkedIn, by reading a page’s structure and visual layout to figure out whether it needs to click a button or type information.

The premise is similar to many browser-based agents released before it: issue a command, and it completes the task, whether that’s ordering food or coffee, booking travel tickets, filing returns, shopping for essentials, reserving a restaurant table, or researching a topic across multiple sources. In a video demo, Hark CEO Brett Adcock showed the assistant taking a command to build a bouquet with flowers the user specified, including handling fuzzy instructions like “some of the florist’s choice.” The video showed only part of the process, though, making it hard to independently gauge how well the agent actually performs.

Hark said it used a post-trained model for this release and plans to pre-train its own model later this year, an approach it says will let it refine its data pipeline, training infrastructure, and techniques more quickly. The company also claims that, unlike large language models that predict the next token, its model predicts the next action — which could be a click or a keystroke at a specific point on the screen.

Hark is entering a field where Google, OpenAI, and Anthropic are all building computer-use agents, alongside venture-funded startups such as Browser Use, Polar, Strawberry, and Aside. Hark says its Handoff agent is faster than the competition and costs much less to run than models like GPT-5.5 and Opus 4.8. The startup has opened a waitlist for the platform and plans to release it by the end of summer.

Why it matters

With $700 million already raised and a claim of being cheaper and faster than rival frontier models, Hark is betting it can undercut incumbents on cost in the crowded and still largely unproven computer-use agent market — though the unverified demo leaves the real-world performance claim untested.