"The measure of whether users will pay subscription fees for artificial intelligence (AI) services will be not 'what the AI knows,' but 'whether the AI can complete tasks.' This means AI is shifting from a conversational tool to an execution agent. That is a massive opportunity for both individuals and organizations."
Robin Li, founder and chief executive officer (CEO) of Baidu, known as China's Google, said on the morning of the 13th (local time), "In the past, the core of AI competition was who could build a smarter model, but now AI's execution capability matters more." He said, "The core of future AI will not be a single function, but the function that consolidates memory, permissions, and workflows in an integrated way." The center of AI competition is moving from model performance to actual work automation.
Baidu held Create 2026 on the morning of the same day at the National Convention Center in Chaoyang District, Beijing. It consisted of an opening ceremony and main forum, more than 40 specialized sessions and public lectures, and a large-scale exhibition hall, and runs for two days. This year's theme is "self-evolution," and Baidu projected, "As AI evolves, individuals' job performance capabilities are evolving, and the future form of organizations will evolve into one where AI collaboration is the default."
In the opening speech, CEO Li stressed that humanity is on the verge of major changes driven by the spread of AI agents. He said there will be significant changes in ▲ the form of AI products ▲ the relationship between humans and AI ▲ the structure of organizations. As AI becomes able to actually complete tasks, people will delegate more and more authority to AI, and as trust in AI grows, more tasks will be automated, he said. As a result, organizational efficiency will increase, and going forward, the core role of managers will be to coordinate the roles of people and AI, Li said.
He also proposed "daily active agents (DAA)" as a new metric for the AI era. Until now, daily active users (DAU) has been the most commonly used metric, but going forward, he argued, the focus should shift from how many users you have to how many agents are completing tasks for users. "How many users are consuming how many tokens only represents expense; it does not represent revenue or real value creation," he said, adding, "By contrast, DAA is closer to value and essence."
In line with the transition to the AI agent era, Baidu is also accelerating the buildout of an "agent-native" infrastructure spanning semiconductors, cloud, and large models. CEO Li said, "Existing foundational infrastructure was all designed around human users, but going forward the entire infrastructure must be rebuilt for the new principal, AI agents," adding, "It needs to be structured so that agents can easily call and utilize it." He continued, "To better support users, we will continue to strengthen agent infrastructure."
New AI product announcements also followed at the opening ceremony. First, in a performance update for the next-generation base model Wenxin 5.1 unveiled on the 9th, Baidu said the model achieved high performance with only about 6% of the pretraining expense compared with peer models. Baidu also said performance in AI agents, reasoning, and deep search has improved significantly, and its creation and reasoning capabilities are approaching the level of global front-running models.
Baidu also unveiled the AI agent mobile app DuMate (百度搭子). DuMate is a service with Baidu's core AI ecosystem functions built in, characterized by enhanced long-term task execution and autonomous decision-making. According to Baidu, DuMate currently records top-tier performance in several global AI agent evaluations.
In addition, ▲ the development platform "Miaoda," which can build web and mobile app programs using only natural language ▲ the comprehensive digital human platform "Baidu Yijing" ▲ and the AI decision-making system "Fanmou" were also introduced for the first time.