Tech Giants Roll Out New AI Agents
Major technology firms are accelerating the deployment of AI agents for conversational commerce. Meta's Manus division launched AI agents on Telegram with plans for WhatsApp and Slack, while Alibaba introduced its Qwen 3.5 model with advanced agent capabilities. The rapid evolution of these platforms is expected to reshape customer engagement and automation in the retail sector.
- Meta's acquisition of the Manus agent technology was valued at over $2 billion, bringing the Chinese-founded AI startup under its umbrella to accelerate its conversational commerce ambitions. - The Manus agent on Telegram offers two distinct models: "Manus 1.6 Max" for complex reasoning and creative tasks, and a faster "Manus 1.6 Lite" for everyday queries, which users can switch between within the chat interface. - Alibaba's Qwen 3.5 model utilizes a "Mixture-of-Experts" architecture with 397 billion total parameters, but only activates a fraction (17 billion) for any given task, a design intended to make it 60% cheaper and eight times more efficient than its predecessor. - Beyond chat, Alibaba's AI has a deep role in its supply chain operations; in 2025, the company reported that the integration of predictive AI tools resulted in a 27% reduction in stock-out incidents and a 19% improvement in delivery efficiency. - The global conversational commerce market is expected to expand from $12.94 billion in 2025 to $14.11 billion in 2026, indicating rapid industry investment in these agent-driven platforms. - These AI agents are designed to go beyond simple chatbots by executing multi-step tasks such as data analysis, file processing, and integrating with productivity tools like calendars, directly within messaging apps. - The broader retail application of such AI agents includes dynamic pricing, hyper-personalized product recommendations based on real-time behavior, and predictive demand forecasting, which has been shown to reduce logistics errors by up to 50%. - Alibaba's model is natively multimodal, meaning it can process and analyze text, images, and video content up to two hours in length and supports over 200 languages, expanding its utility for global vendor and customer interactions.