A new AI startup called Hark has entered the browser-agent race with a product that may change how businesses think about automating web-based work.
On August 5, Hark announced a technical preview of its first product, Handoff, an AI agent built to browse the web and complete real tasks on a user’s behalf. The company, founded by serial entrepreneur Brett Adcock, backed the announcement with a bold claim: Handoff achieved the highest-ever score on the Online-Mind2Web (OM2W) benchmark, the industry-standard measure for web-browsing AI performance, scoring 97.7 compared to OpenAI’s ChatGPT 5.4 at 92.8 and Anthropic’s Claude Opus 4.8 at 84.1.
Hark raised a $700 million Series A round at a $6 billion valuation in May 2026. Adcock previously co-founded Figure AI, one of the most-watched humanoid robotics companies, and Archer Aviation. Handoff is the company’s first publicly announced product.
How Handoff Works
Handoff operates through a dedicated virtual computer that Hark runs on its own infrastructure. The agent sees the screen, clicks buttons, fills in forms, and navigates websites just like a person would, which means it can handle tasks on sites that have no official API.
That distinction matters for practical business use. Most AI automation tools require a clean data structure or a developer to build an integration before anything works. Handoff bypasses that by reading and interacting with the interface directly, the same way a human employee would.
Hark says the agent can handle tasks including ordering food end-to-end through DoorDash and Uber Eats, price-comparing and booking flights across United, Delta, and American Airlines, researching and messaging candidates on LinkedIn, and multi-source research projects that would take hours to complete manually.
The company also claims a pricing advantage: less than one-tenth the token cost of comparable products from OpenAI and Anthropic.
A broader commercial launch is expected before the end of summer 2026. The current phase is an application-based beta through a waitlist.
What This Means for Business
The arrival of affordable, high-performing web agents marks a real inflection point for business operations.
For most of the past two years, deploying AI agents against real browser-based workflows required expensive integrations, a technical team, and models that were either too slow, too costly, or too unreliable for production use. The combination of benchmark performance and low pricing that Hark is positioning around starts to change that calculus.
Think about the kinds of tasks that eat time in every business: checking competitor pricing, monitoring product availability, submitting data to government portals, pulling information from vendor websites, researching candidates before a meeting. These are not complex decisions, but they are real work that falls to your team by default.
A browser agent that can reliably complete those tasks, at low cost, running in the background, is the kind of tool that makes sense to wire into operations as soon as it becomes stable enough to trust.
There are still real questions about Handoff specifically. Hark has yet to address how enterprise security teams should think about a virtual machine that has full browser access, including sessions, passwords, and connected accounts. For most businesses, those questions need answers before this goes anywhere near production.
But the broader trend is clear. Computer-use AI, which lets agents interact with software through its interface rather than through code, is maturing quickly. Google’s Gemini Spark recently gained the ability to control a logged-in Chrome session. Anthropic’s Claude Computer Use has been in enterprise testing for months. OpenAI’s Codex runs code and files directly on a user’s machine.
Hark is positioning Handoff as the fastest and most affordable option in this emerging category, at a price point that could bring browser automation to businesses that currently assume it is out of reach.
The full launch will be the real test, but the benchmark results and the funding behind the company suggest this is a product worth watching.
Enterprise DNA helps businesses build the data skills and AI strategies needed to get ahead of changes like this. Explore EDNA Learn for practical training in AI and data tools, or book a discovery call to talk through how AI agents fit into your operations.
Source
TechCrunch
Free Resource
Going deeper with Claude?
Get the free 32-page implementation guide for ANZ teams.
Your guide is ready
Check your downloads folder. If it did not open automatically, use the button below.
Download the GuideWant this working inside your business?
See what's possible