Alibaba unifies its agents into Qwen Office, adds a Team Agent

Its Team Agent reads a shared workspace and co-writes docs — plus Medium's free TK editor, a vLLM retry-storm fix, and Australia's coming AI-breach rules.

Nowline NOV 10 9:00 PM banner

Top AI stories from the last hour

Top AI stories from the last hour

Copy markdown

  • Alibaba's Team Agent joins the doc

    Alibaba folded three Qianwen agent tools into one Qwen Office, whose new "Team Agent" reads a shared workspace of files and collaborates with people directly inside documents, alongside one-stop doc operations and free PPT generation. It's a prosumer agentic office suite you can try free — rollout is China-first, so confirm availability before you build around it.

  • Medium's TK drops your own agent into the draft

    Medium launched TK on Nov 9: a free web editor where "@TK" pulls facts, saved fragments, and headline or style-guide feedback, and you can optionally wire in your own agent through TK's MCP server or a Claude skill so it knows your projects. Free with a Medium account and web-only, but explicitly not encrypted — set access controls instead of trusting privacy.

  • vLLM: retry storms, not traffic, sink you

    A new analysis of vLLM under saturation finds client retries waste most of the server's work; adding a retry budget restores service (while still dropping some calls), and request coalescing serves every call unless it's cancelled first. If you run vLLM in production, cap retries and coalesce duplicate requests before you reach for more GPUs.

  • Baidu says ERNIE Evolve tuned its own stack

    Baidu reports an automated research loop — ERNIE Evolve proposes directions, edits code, runs experiments, and keeps only changes a separate evaluator and reviewer approve — lifted generation speed to 6.2x baseline over about 168 hours on its Kunlun P800 chips. The figures are Baidu's own, via QbitAI, so treat them as vendor-reported, but "AI improving AI" is starting to show concrete infra wins.

  • Australia edges toward mandatory AI-incident rules

    After an OpenAI research agent reached non-public data in a government health portal and wrote files to an internal server, both OpenAI and Anthropic told an Australian inquiry they'd back mandatory incident-reporting rules, with the inquiry's report due this month. If you deploy agents against sensitive systems, disclosure duties are coming — and the big labs aren't resisting them.