Hark just stepped into one of the most crowded races in AI. The startup, which raised $700 million in Series A funding back in May, today launched Hark Handoff, an agent that uses a web browser to complete tasks on your behalf, according to TechCrunch AI. The pitch is simple: issue a command, and Handoff does the clicking, typing, and navigating for you.
What stands out is where Hark says it can operate. TechCrunch AI reports that Handoff can work across sites with no official APIs, including Target, Walmart, OpenTable, and LinkedIn. Instead of plugging into a backend, the agent reads a website’s structure and visual layout to figure out whether it needs to press a button or fill in a field. That’s the same core idea behind most browser agents, but the target list is ambitious.
What Handoff can do
Hark says the agent handles the kind of errands people actually put off. Based on the TechCrunch AI report, that includes:
- Shopping and essentials. Ordering food or coffee and buying everyday items.
- Travel booking. Reserving tickets and sorting out trip logistics.
- Returns and admin. Filing returns and working through tedious forms.
- Restaurant reservations. Booking a table through sites like OpenTable.
- Research. Pulling information from multiple sources and reporting back.
In a video demo, CEO Brett Adcock showed Handoff building a flower bouquet from a user’s spec, including loose instructions like “some of the florist’s choice.” Handling fuzzy language is genuinely hard for agents, so that’s a smart thing to show off. One caveat worth flagging: the video only shows part of the process, so as TechCrunch AI notes, there’s no real way to judge how well it works end to end yet.
The technical angle
Here’s the part I find most interesting. Hark says this release runs on a post-trained model, with pre-training planned for later this year. The company frames that order as a feature, not a shortcut, arguing it can refine its data pipeline, training setup, and techniques faster this way.
Hark also claims its model works differently from a standard large language model. Where an LLM predicts the next token, Hark says its system predicts the next action, like a click or a keystroke at a specific spot on the page. If that holds up in practice, it’s a meaningful design choice for agents that live inside a browser rather than a chat window.
How it stacks up
Hark is not alone here, and TechCrunch AI is direct about that. Google, OpenAI, and Anthropic are all building computer-use agents, and a wave of VC-backed startups, including Browser Use, Polar, Strawberry, and Aside, are chasing browser-based automation specifically.
Hark’s differentiation comes down to two claims: speed and cost. The company says Handoff is faster than the competition and runs much cheaper than models like GPT 5.5 and Opus 4.8. Those are the kinds of claims that only matter once independent testing confirms them, so treat them as marketing until proven.
Availability
Hark has opened a waitlist for the platform and plans to ship it by the end of summer. There’s no public pricing yet, and no open access, so for now this is a preview backed by a big funding round and a partial demo.
This matters because browser agents are shaping up to be the next real battleground in applied AI. The technology is close enough to work that everyone from the biggest labs to nine-figure startups is pushing on it, but nobody has nailed reliability across messy, real-world websites. Hark’s action-prediction approach and cost claims give it a story. Whether Handoff can actually finish the job, without a human quietly fixing things off-camera, is what the summer release will reveal. You can find the full details at the original TechCrunch AI report.