The concept of the "AI Agent" continues to gain significant traction. The upcoming 2026 World Artificial Intelligence Conference is set to feature a high concentration of these agents, and a preview visit today (16th) revealed that this futuristic concept is already becoming part of everyday life. Leading companies' exhibition booths are almost universally showcasing their own AI agent products. While past events focused on comparing AI parameters and computing power, the industry's focus has shifted this year towards real-world application scenarios. AI agents have evolved from simple conversational chatbots into "digital employees" capable of independently performing tasks. Many companies are calling this the inaugural year for practical AI agent deployment.
During a demonstration, a command was issued: "Help me order an errand service from point A to point B." The Baidu "Buddy" agent application automatically accessed an integrated "Meituan Paotui" tool and proceeded to place the order. "As long as you state your need, this agent can autonomously understand the task, break down the execution steps, and call upon the appropriate tools," explained Li Cong, the product operations lead at the booth. He added that it possesses continuous memory and learning capabilities, recording users' work habits over time, and is always available on-demand, helping users reduce meaningless waiting periods.
Li Cong shared a recent example from Beijing's rainy weather last week. Staff set a task for the agent to book a taxi from point A to point B only when it was not raining. Instead of responding immediately, the agent monitored the weather in real-time. When the forecast indicated no rain for the next hour, it automatically completed the ride-hailing via a Didi Chuxing interface and sent a notification to the user, who could then prepare to leave while waiting for the car to arrive. This product is the only general-purpose AI agent selected among the conference's ten "Treasures of the Hall," boasting unique advantages.
Li Cong stated that while many general-purpose agents exist, the differentiating strength of Baidu's "Buddy" lies in its underlying self-developed engine, which ensures more stable execution performance. For users, this translates to better output quality and higher security. It is reported that on core task delivery metrics for general-purpose agents, "Buddy" ranks within the global top tier, with end-to-end task completion rates and complex task completion rates reaching internationally leading levels. It demonstrates superior performance in practical tests for general capabilities like web operations and information retrieval, as well as in high-frequency work scenarios such as report writing, proposal drafting, copywriting, and marketing content creation. In authoritative evaluations, "Buddy" topped the "PinchBench" public agent benchmark with a score of 93.3%.
Evolving User Habits
An interesting observation is that the evolution of AI agents reflects profound changes in user behavior. At the previous World AI Conference, the showcased AI assistant "Lingxi" was an embedded plugin within WPS Office software. This year, it has become a fully independent AI agent product. "'Lingxi' is no longer just an AI feature layered onto traditional software. This change stems from our observation of shifting user behavior patterns," said Liu Tuochen, product lead for Beijing Kingsoft Office's Lingxi AI. He noted that users now tend to open a dedicated AI application first to handle an entire workflow—like understanding a concept or creating a presentation—rather than repeatedly editing a single document within a traditional program.
The standalone "Lingxi" AI-native office agent features a new "Automatic Dreaming" function, or knowledge extraction capability. This function can automatically extract valuable insights from a user's cloud documents, local files, and recent conversation history, gradually building a deep understanding of the user's personal knowledge system over time. Once activated, it systematically records past projects, writing styles, topics of interest, and mentioned concepts, and can then automatically generate text tailored to the user's preferences based on new instructions.
From Conversation to Execution
Tencent also debuted its "WorkBuddy" (AI Office Executor) at the event. Built on Tencent's agent architecture, it allows users to create, invoke, and coordinate multiple agents using natural language to complete complex office tasks, upgrading the experience from "conversational AI" to "executive AI." Through one-click task decomposition and multi-agent collaboration, employees can simply state a requirement, and the AI will autonomously run through the entire compliant office process in the background.
Core Industrial Applications
As a global manufacturing powerhouse, China's vast industrial sector presents a core application arena for AI agents to demonstrate their capabilities. At the JD.com booth, a staff member demonstrated a use case: she took a photo of a worn equipment nameplate with her phone. The AI agent automatically identified the model, and within seconds, product links and three alternative options appeared on the screen. A click confirmed the selection and placed the order with one tap.
Wang Chen, an operations staffer for the AI Intelligent Procurement Steward product, explained that the team visited numerous manufacturing firms during development and identified common procurement pain points. When a part on the production line fails, an engineer sends a photo to procurement, who then must identify the model from the image, seek quotes from various sources, and compare parameters—a process that can take days for a single part. For procurement lists with over 100 line items, manually searching for each one is particularly tedious and inefficient for many manufacturers. AI can automate all of this.
Wang Chen introduced that using the AI Intelligent Procurement Steward, a buyer simply states their need in natural language—via text, voice, a photo, or a blueprint—and the AI handles the understanding and execution. The process spans from image recognition and sourcing, parsing bills of materials, and intelligently matching substitute parts, to real-time price comparison and sourcing, one-click ordering, and order tracking. This shifts procurement decisions from "experience-driven" to "data-driven," reducing selection and decision-making time by 70% and lowering the risk of mismatched parts by 60%.
Extending to Design
The application extends beyond procurement into the design phase. A selection agent can start from technical drawings, automatically reason and decompose tasks, bridging design and procurement to shorten design cycles by over 50%. For instance, AI design visualization projects enable "design as selection, selection as procurement." By generating 3D models through AI dialogue and completing parts list matching, it significantly lowers the barrier to design, accelerates the journey from design to production, reduces overall design costs, and automatically matches the most cost-effective parts solutions.
Comments