Hark Unveils Browser Agent to Automate Web Tasks

The system functions by parsing a site’s structure to determine when to click or input data, a process CEO Brett Adcock demonstrated by having the tool curate a custom floral bouquet. While the demo showcased the agent’s ability to handle ambiguous user requests, the company has yet to provide a full, unedited look at its real-world reliability. Unlike standard large language models that focus on token prediction, Hark claims its post-trained model is engineered specifically to anticipate UI actions such as mouse clicks or keystrokes.

Hark intends to transition to pre-trained models later this year to accelerate its data pipeline development. The startup faces stiff competition from established tech giants like Google, OpenAI, and Anthropic, alongside niche players such as Browser Use and Polar. To differentiate itself, the company asserts that Handoff offers superior speed and lower operational costs compared to current industry benchmarks. A waitlist is now live, with a public release scheduled for the end of summer.