Startup Hark has introduced Hark Handoff, a browser-use agent designed to complete digital tasks by navigating websites on its own. Backed by a $700 million Series A round announced in May, the company is positioning the product as a practical step toward more capable AI assistants.
According to Hark, Handoff can work across sites that do not offer official APIs, including retail, booking, and professional platforms. The system analyzes page structure and visual cues to decide when to click, type, or move to the next step, aiming to handle workflows that usually require manual input.
The assistant is built for everyday actions such as shopping, reserving tables, arranging travel, or gathering information from multiple web sources. In a demo, CEO Brett Adcock showed the agent assembling a flower bouquet based on user preferences, including flexible instructions like selecting the florist's choice.
For this release, Hark says it is using a post-trained model and plans to move toward pre-training later this year. The company says this approach will help it improve its data pipeline, training systems, and model behavior more rapidly. It also describes Handoff as an action-predicting system rather than a next-token model, with outputs focused on the next step in a task.
Hark enters a fast-moving field that includes major AI developers and emerging startups building computer-use agents. The company says Handoff is designed to be faster and more affordable than competing systems, and it has opened a waitlist ahead of a planned launch by the end of the summer. As browser agents mature, they may reshape how people delegate digital work in the years ahead.
