Past AI News

All

1. JPMorgan research: Muse passes 10M downloads with ~25% of users shopping; agentic shopping may feed card networksUnconfirmed

   User and size data for agentic shopping.

   @ArtofSpecuyckyChinese-language13 likes · 0 reposts · 5 replies · 6,529 viewsView topic

   ExpandCollapseJPMorgan research: Muse passes 10M downloads with ~25% of users shopping; agentic shopping may feed card networks

   The post relays JPMorgan research: Meta and Stripe launched card-based agent payments in the Muse AI assistant, which has passed 10M downloads since its September launch; about 25% of users shop with it, with average transactions above $200. Agents use Stripe Link to create single-use virtual cards and pay on existing merchant sites without per-merchant integration. About 30% of consumers are open to agentic shopping to some degree; about half of them prefer digital wallets, led by PayPal at 32%. JPMorgan estimates that even at 40% US penetration and $6,000 per user per year, volume would be about $240B, roughly 2.2% of Visa and Mastercard US purchase volume.

2. Alibaba's Qwen Go adds "My Apps": build apps by describing them, with authorized access to Taobao, Amap and Damai data

   Bringing e-commerce data into AI agents.

   @ppmmzzcnChinese-language1 likes · 0 reposts · 0 replies · 106 viewsView topic

   ExpandCollapseAlibaba's Qwen Go adds "My Apps": build apps by describing them, with authorized access to Taobao, Amap and Damai data

   Alibaba's AI agent Qwen Go launched a "My Apps" feature: users can build life, work and study apps by describing what they need, without coding, and share apps that keep updating their data. It runs on Alibaba Cloud's app platform Meoo, which handles building, database setup and deployment automatically. With user authorization, these apps can draw on personal data from Taobao, Amap, Damai and Quark Drive.

3. Paper: in retail, airline and telecom customer-service simulations, multi-agent setups lost to single agents in 12 of 13 comparisons at ~5.5x cost

   A reference for retail customer-service agent design.

   @vintcessunChinese-language2 likes · 0 reposts · 5 replies · 159 viewsView topic

   ExpandCollapsePaper: in retail, airline and telecom customer-service simulations, multi-agent setups lost to single agents in 12 of 13 comparisons at ~5.5x cost

   The post covers an arXiv paper (2610.27655) that compared centralized multi-agent systems with a single agent, using the same model, in τ²-bench's airline, retail and telecom customer-service scenarios across about 7,950 simulated conversations. In 13 valid model-domain comparisons, multi-agent setups were worse 12 times and tied once; even without a budget cap they at best tied, at about 5.5x the cost. The authors note the results are limited to the systems and simulator tested. For retail customer service, splitting work across more agents is not automatically better.

4. Naver's AI shopping agent adds payments in a year-end pilot for select categories; daily transactions are up about 4x since its March launch

   Shopping agents are moving from recommendations to checkout.

   @Aju Pressno post-level likes, reposts, replies, or viewsView topic

   ExpandCollapseNaver's AI shopping agent adds payments in a year-end pilot for select categories; daily transactions are up about 4x since its March launch

   According to Aju Press (Oct 25), Naver announced it will pilot payments inside its AI shopping agent for select categories by year-end and expand the agent to the PC web. Since the agent launched in March, monthly active users and daily conversations had grown about 5x by August, and daily transaction volume about 4x. Digital and electronics products account for 40% of agent transactions, and Naver Plus members for 78%. Naver plans to draw on blog and Cafe community content plus purchase reviews to strengthen the agent's recommendations.