Hark Unveils Browser Agent for Hands-Off Web Task Automation

Hark, which landed a $700 million Series A in May, has now revealed Hark Handoff—an AI-powered agent designed to traverse the web autonomously and handle everyday tasks that normally require manual interaction.

How It Works#

The agent approaches web automation differently from traditional browser bots. Rather than relying on official APIs or pre-built integrations, Handoff analyzes website layouts and visual elements to determine what needs clicking or typing. The system can work with major retail and service platforms including Target, Walmart, OpenTable, and LinkedIn—sites that lack standardized automation channels.

According to Hark, the underlying model doesn't simply predict the next word like typical language models. Instead, it forecasts the next action—whether that's a mouse click at a precise coordinate or a keyboard keystroke—allowing for finer-grained control over web interactions.

Capabilities and Demo#

In a demonstration, Hark CEO Brett Adcock illustrated the agent handling a flower arrangement task, accepting specific blooms plus looser instructions like "some of the florist's choice." The system can manage a range of workflows: purchasing goods, ordering food or beverages, booking travel, filing tax returns, making restaurant reservations, and gathering information across multiple sources.

The company emphasized that its current release relies on post-training techniques, with plans to shift to pre-training later in the year. This phased approach lets Hark refine its data pipelines and training infrastructure more rapidly.

Competitive Landscape and Performance Claims#

Hark enters a crowded field where tech giants like Google, OpenAI, and Anthropic are all developing computer-use agents. Venture-backed startups including Browser Use, Polar, Strawberry, and Aside are pursuing similar browser automation play.

Hark claims its Handoff agent outpaces rivals in both speed and cost-efficiency compared to models such as GPT 5.5 and Opus 4.8. The startup has opened a waitlist and plans a full launch by summer's end.