Editorial illustration for AI startup Hark launches its first product, the Hark Handoff computer use agent
Hark Launches Handoff, AI Agent for Web Tasks
AI startup Hark launches its first product, the Hark Handoff computer use agent
Brett Adcock's latest company came out of stealth today. Hark, founded earlier this year by the entrepreneur behind Figure and Adaptive Shipyards, unveiled its first product: a "computer use agent" called Handoff, built to browse the open web and complete tasks on a person's behalf. Order dinner through DoorDash, book a flight on United, message a candidate on LinkedIn, the pitch is that Handoff does it end-to-end without a human clicking through each step.
Sign-ups open today at hark.com, with the product itself rolling out later this month as part of Hark's first software release. The company is leaning hard on benchmark numbers to make its case, claiming a top score on Online-Mind2Web, a third-party leaderboard that grades web agents with human evaluators, and pricing well below what OpenAI, Anthropic and Google charge for their frontier models. Each Handoff session spins up its own virtual computer, complete with browser, file system and terminal, and users can link existing accounts so the agent can log in with saved addresses and payment details already on file.
Why this matters
Adcock has a habit of launching bold companies (Vettery, Archer Aviation, Figure AI) and this one arrives with the same confident framing but a thinner data trail. Hark wants Handoff judged as a top performer at browsing the open web, yet the benchmark table it published undercuts that pitch: GPT 5.5 beats it 72.3 on WebTailBench v2, and Hark wouldn't say whether it plans to test against newer competing models at all. For developers and founders evaluating computer use agents for real workflows, that's the detail to sit with, not the sign-up page.
Autonomous agents booking flights or messaging candidates end-to-end sound useful right up until they fail silently on an edge case nobody benchmarked. We'd rather see Hark publish head-to-head numbers against whatever OpenAI, Anthropic, or Google ship next month than lean on a comparison table it already knows has holes. Until then, treat "among the best in the world" the way you'd treat any founder's launch-day claim: interesting, unverified, worth checking again once real users start filing bug reports.
Related Reading
Business & Startups Nordic pilot adds Gemini for Education, NotebookLM to boost AI literacy
Business & Startups Kling launches Video O1, all-in-one model with MVL bridge using transformer
Business & Startups DeepSeek Seeks More Capital Weeks After USD 7B Funding Round
From Archives Anthropic is building a team to design its own AI chips
From Archives Court Overturns Ban on Perplexity's AI Shopping Tool for Amazon
Common Questions Answered
What is Hark Handoff and what tasks can it perform?
Hark Handoff is a computer use agent developed by AI startup Hark that can browse the open web and complete tasks on a person's behalf without requiring manual clicks through each step. Examples of tasks it can handle include ordering dinner through DoorDash, booking flights on United, and messaging candidates on LinkedIn, automating end-to-end workflows that would normally require human intervention.
How does Hark Handoff's performance compare to competitors on the OM2W benchmark?
Hark Handoff recorded a score of 97.7 on the Online-Mind2Web (OM2W) benchmark, outperforming OpenAI's GPT 5.4 (92.8), Anthropic's Claude Opus 4.8 (84.1), and Google's Gemini 2.5 Pro (69). This top-ever score on the third-party human-evaluated leaderboard positions Handoff as the leading web agent on this particular benchmark.
Who founded Hark and what is his background?
Hark was founded by Brett Adcock, an entrepreneur known for launching several bold companies including Vettery, Archer Aviation, and Figure AI. Adcock has a track record of launching confident, ambitious ventures, and Hark represents his latest venture in the AI space.
What concerns exist about Hark Handoff's benchmark performance claims?
While Handoff leads on the OM2W benchmark, it underperforms on other metrics, with GPT 5.5 scoring 72.3 compared to Hark's score on WebTailBench v2. Additionally, Hark has not committed to testing Handoff against newer competing models, which raises questions about the comprehensiveness of its performance evaluation.
Further Reading
- Figure AI Founder Brett Adcock Launches Hark: A New Lab Building True Personal AGI - Quasa
- Computer-Using Agent - OpenAI
- Handoff Agent Orchestration - Microsoft Learn
- Handoffs - OpenAI Agents SDK - OpenAI
- Documentation | Handoff - Handoff
