Hark previews its browser use agent for completing tasks
https://techcrunch.com/2026/08/05/hark-previews-its-browser-use-agent-for-completing-tasks/📌 【Hark 新品預覽】不用 API 也能自動操作，Hark Handoff 試圖改變瀏覽器代理的競爭格局
TL;DR：Hark 推出 Handoff 代理，透過視覺與結構數據直接操作無 API 網站，並宣稱比現有 LLM 更快且更便宜。
當我們談論自動化時，通常受限於網站是否提供 API。但如果 AI 可以像人類一樣，直接「看」著網頁並點擊按鈕呢？
🤔 跳脫 API 限制，直接與網頁視覺互動
Hark 推出名為 Hark Handoff 的代理程式，其核心能力在於能夠高效使用瀏覽器來完成任務。與傳統依賴官方 API 的自動化流程不同，Handoff 能夠瀏覽並操作如 Target、Walmart、OpenTable 與 LinkedIn 等沒有提供官方 API 的網站。
🧩 從預測「下一個字」轉向預測「下一個動作」
Hark 在技術路徑上提出了一個關鍵差異：
- 傳統 LLM：預測下一個 Token（字元/詞元）。
- Hark 模型：預測「下一個動作」— 這可能包含在特定位置觸發時鐘動作或鍵盤輸入。
為了加速開發，Hark 目前使用「後訓練模型」（post-trained model），並計畫在今年年底進行預訓練（pre-train），藉此更快速地優化其資料流水線（data pipeline）與訓練基礎設施。
📊 執行任務與成本優勢
在實際應用場景中，Handoff 可以處理包含訂購食物、預訂旅遊票券、處理退貨、購物、預訂餐廳，甚至是根據模糊指令（例如：「隨花藝師的選擇」）來製作花束等研究任務。
根據 Hark 宣稱的表現：
- 執行速度：比目前的競爭對手更快。
- 成本效益：成本遠低於 GPT 5.5 與 Opus 4.8 等模型。
⚠️ 技術實作細節仍待觀察
儘管 Hark 宣稱其效能優越，但目前的展示影片僅呈現了流程的一部分，開發者與研究者仍難以從中判斷其在複雜環境下的真實有效性。目前該平臺已開放候補名單，預計將於夏季結束前正式發佈。
🎯 實務啟示
對於需要處理大量非結構化網頁資料的自動化流程，Hark 展現了「視覺驅動型代理」的發展趨勢。如果該技術能如其宣稱般降低成本並提高速度，未來處理複雜網頁操作的門檻將大幅降低。
🔗 來源
- 標題：Hark previews its browser use agent for completing tasks
- 作者／機構：Ivan Mehta @ TechCrunch AI
- 連結：https://techcrunch.com/2026/08/05/hark-previews-its-browser-use-agent-for-completing-tasks/
#AI #BrowserAgent #Automation #Hark #MachineLearning #ComputerUse #LLM #TechNews #WebAutomation #AIStartup
原始資料 TechCrunch AI · 收集於 2026-08-06
摘要原文
Hark , a startup that raised $700 million in Series A funding in May , today launched its agent Hark Handoff, which can use a browser efficiently to complete tasks. The company claims that Handoff can easily navigate websites that have no official APIs, including Target, Walmart, OpenTable, and LinkedIn. It said that the Handoff agent looks at website structure and visual data to understand if it needs to click buttons or type information. The premise is similar to tons of browser-based agents released before. Issue a command, and it will complete tasks for you, including ordering food or coffee, booking travel tickets, filing returns, shopping for essentials, reserving a table at a restaurant, or researching on your behalf by looking at various sources. In a video demo, the company’s CEO, Brett Adcock, showed that the assistant can take a command to build a bouquet with flowers users specify, along with accommodating fuzzy terms like “some of the florist’s choice.” Notably, the video shows only part of the process, so we can’t really gauge its effectiveness. The company said that for this release, it’s using a post-trained model, and plans to pre-train later this year. Hark said that, with this approach, it can refine its data pipeline, training infrastructure, and techniques more quickly. Hark also claims that, unlike large language models (LLMs) predicting the next token, its model can predict the next action — which could be a clock or a keyboard input at a specific place. There are many companies working on computer-use agents, including Google, OpenAI, and Anthropic, with VC-funded startups working on browser-based task automation, such as Browser Use , Polar , Strawberry, and Aside. Hark said that its Handoff agent is faster than the competition and also costs much less than other models like GPT 5.5 and Opus 4.8. The startup has opened a waitlist for its platform and plans to release it by the end of the summer.
由 tencent/hy3:free 自動生成