A startup that raised $700 million in a Series A round just three months ago is already shipping product. Hark launched its browser agent called Handoff, a tool designed to complete real-world tasks on the web without relying on official APIs. That funding figure alone makes this worth watching, even if the demo leaves some questions open.
The pitch is familiar. Give Handoff a command, and it handles the execution. That means ordering food, booking flights, filing tax returns, shopping for groceries, or reserving a restaurant table. The agent reads website structure and visual data to figure out where to click or what to type. According to Hark, it works on sites like Target, Walmart, OpenTable, and LinkedIn, all of which have no public APIs for third-party automation.
CEO Brett Adcock showed a demo where the agent builds a custom flower bouquet, handling vague instructions like “some of the florist’s choice.” It’s a smart choice of demo because it shows flexibility with language. But the video only covers part of the task flow, so it’s hard to judge reliability in real conditions. That gap matters a lot for browser agents, which historically fall apart on edge cases and dynamic page layouts.
The technical story is interesting. Hark says its model doesn’t predict the next token like a standard large language model. Instead, it predicts the next action, whether that’s a mouse click or a keyboard input at a specific location. The company is currently using a post-trained model and plans to move to pre-training later this year, arguing this lets it iterate faster on data pipelines and training methods before committing to a full pre-training run. That’s a reasonable approach for an early-stage team moving fast.
Still, the competitive pressure here is real. This space includes Google, OpenAI, and Anthropic on the model side, plus VC-backed startups specifically focused on browser automation:
- Browser Use, an open-source framework with strong developer traction
- Polar, focused on structured web workflows
- Strawberry, targeting enterprise task automation
- Aside, building assistants for knowledge workers
Hark claims Handoff is faster and cheaper than GPT 5.5 and Opus 4.8. Those are meaningful benchmarks if they hold up under real workloads. Speed and cost are exactly what enterprise buyers and developers look at when they’re choosing between agents that otherwise look similar on the surface. The company has opened a waitlist and is targeting a full release before the end of summer. Whether the product lives up to the funding round is the only question that matters now.
