Baidu Dazi (Baidu's AI agent product) has rolled out an update that lets tasks continue across devices: users can start a task on a desktop and resume it on a phone, carrying over not just chat history but task context, execution progress, and long-term memory. The desktop client gains an embedded browser, while the mobile side can hand web tasks to a cloud browser for continued execution, such as competitor research across multiple sites, reviews, and price comparisons compiled into tables or documents. Critical actions like logins, authorizations, and form submissions require human confirmation rather than blanket permissions. Baidu reports vendor-side metrics including 20% lower average task time, 25% better token utilization, and a 94% high-delivery rate on complex tasks, though no benchmarks or sample sizes were disclosed. The update also highlights model routing based on difficulty, cost, and security, plus skill composition by goal rather than tool names. Open questions remain around permission models, audit logs, and credential cleanup before enterprise adoption.
Baidu Dazi Adds Cross-Device Handoff and Cloud Browser Execution for Continuous Agent Tasks
Summary
Baidu Dazi (Baidu's AI agent product) has rolled out an update that lets tasks continue across devices: users can start a task on a desktop and resume it on a phone, carrying over not just chat history but task context, execution progress, and long-term memory. The desktop client gains an embedded browser, while the mobile side can hand web tasks to a cloud browser for continued execution, such as competitor research across multiple sites, reviews, and price comparisons compiled into tables or documents. Critical actions like logins, authorizations, and form submissions require human confirmation rather than blanket permissions. Baidu reports vendor-side metrics including 20% lower average task time, 25% better token utilization, and a 94% high-delivery rate on complex tasks, though no benchmarks or sample sizes were disclosed. The update also highlights model routing based on difficulty, cost, and security, plus skill composition by goal rather than tool names. Open questions remain around permission models, audit logs, and credential cleanup before enterprise adoption.
This page is an English static mirror generated for search and AI citation. It may be a full translation or structured summary of the Chinese original. Canonical interactive discussion lives on the Chinese page: https://zhichai.net/topic/178503721