What this is
Alibaba Qwen themed this week's livestream "Agent First" — redefining multimodal from "can understand" to "tool that gets work done." This is the first time a major Chinese tech company has explicitly followed the global Agent (AI capable of autonomously executing multi-step tasks) pivot in its product narrative. The livestream centered on a 27B-parameter model, emphasizing multimodal connecting directly to tasks rather than remaining at the conversational level.
Industry view
This warrants broader context. Since the second half of 2024, global foundation model companies have pivoted from "answering fluently" to "completing tasks": OpenAI launched Operator, Anthropic launched Computer Use, Google launched Gemini Agent. Qwen's explicit Agent-first positioning is the first time a Chinese major has aligned with this judgment in product narrative.
But there are cautious voices. The real-world reliability of multimodal agents has long been in question — pretty demos, poor deployment was common throughout 2024. Whether 27B can support complex task chains is a skepticism held by many engineers.
Impact on regular people
- For enterprise IT: When selecting models, distinguish between "chat-capable multimodal" and "work-capable multimodal" — the latter's deployment cost will be significantly higher than the former.
- For individual professionals: Those who can run workflows through AI Agents will hold a meaningful efficiency edge over those who only know how to prompt.
- For the consumer market: No major C-end changes short-term. Multimodal agents need to prove themselves in B-end scenarios first.