Baidu Deploys AI Agents to 700M Search Users

The underlying technology for Baidu's agent is the ERNIE (Enhanced Representation through Knowledge Integration) series of models. The latest versions, ERNIE 4.5 and X1, feature a multimodal architecture that processes text, images, audio, and video jointly, improving reasoning and efficiency. This is part of a broader company strategy focused on four agent categories: Search, "Digital human" agents, Coding agents, and "Self-evolving" AI. The move reflects an intensifying AI agent race in China, with competitors like Alibaba and Tencent also developing agent platforms such as Tongyi Lab and Yuanqi, respectively. The Chinese AI agents market, which generated over $577 billion in revenue in 2025, is projected to grow at a CAGR of 50.8% through 2033. This domestic competition is fostering a new wave of autonomous agents, like Manus AI, designed to execute complex workflows with minimal human input. For orchestrating such complex, multi-agent systems, open-source frameworks like Microsoft's AutoGen, CrewAI, and LangGraph are becoming critical infrastructure. AutoGen excels at managing conversational, asynchronous interactions between agents, while CrewAI offers a higher-level, role-based abstraction for faster prototyping. These frameworks provide essential components for memory management, structured workflows, and deterministic control, which are vital for reliability at scale. Recent research in AI agent architecture focuses on enabling more autonomous and robust behaviors. Papers on "Self-Evolving Agents" explore how agents can learn from feedback and experience to improve their skills over time. Other key research areas include dynamic role discovery, where agents can autonomously assign tasks based on evolving needs, and developing unified memory systems to manage both short-term context and long-term knowledge. From a user experience perspective, designing for agents requires a paradigm shift from direct manipulation to delegation and supervision. Key UX patterns include providing transparency into the agent's reasoning ("footprints"), ensuring users have ultimate control with overrides, and designing flexible interfaces that support continuous human-agent interaction across multiple stages of a task. The goal is to make the agent's complex behavior feel predictable and trustworthy to the end-user. Scaling the engineering teams required to build these systems introduces its own set of challenges. A primary responsibility for a CTO is aligning AI initiatives directly with business problems and defining success metrics beyond just model accuracy, focusing instead on operational efficiency or user adoption. As teams grow, establishing a solid operational baseline with a comprehensive service catalog, clear ownership, and robust monitoring is crucial to prevent AI from amplifying chaos.