AI Enters the Era of 'Self-Evolution', Li Yanhong First Proposes 'DAA' as the Measurement for the AI Era

05/13 2026 509

From May 13 to May 14, the Create2026 Baidu AI Developer Conference was held in Beijing. Baidu founder Li Yanhong, at the opening ceremony, first proposed the 'measurement' for the AI era—Daily Active Agents (DAA). In Li's view, Tokens do not necessarily represent the final outcome; they only represent costs, not benefits, measuring input rather than output. DAA roughly corresponds to the commonly used metric in the mobile internet era, Daily Active Users (DAU). 'To measure the prosperity of a platform and ecosystem, we should focus more on the DAA metric, paying attention to how many Agents are working for humans and delivering results,' Li said. Li also predicted that the global number of Daily Active Agents (DAA) could exceed 10 billion in the future.

At the opening ceremony, Li Yanhong proposed the 'AI Evolution Theory.' Its meanings are: first, the self-evolution of intelligent agents, from passive responses to continuously improving themselves by absorbing nutrients from the environment and taking proactive actions; second, the self-evolution of individual humans, from ordinary individuals to super individuals, learning to coexist with AI; third, the self-evolution of corporate organizations, from division of labor and collaboration among humans to mixed teams of humans and intelligent agents, becoming super organizations.

'AI entry points are being rebuilt, human-machine relationships are being reshaped, and organizational forms are being reconstructed. We are entering a new era of productivity,' Li said. The emergence of intelligent agents represents that AI development is moving from the model stage to the application stage, 'for the first time, the protagonist of AI is not the model but the application.'

Focusing on the surge in intelligent agent applications, Baidu collectively released and upgraded a series of products, including the general-purpose intelligent agent DuMate, the code intelligent agent Miaoda, the digital human intelligent agent Baidu Yijing, and the self-evolving decision-making intelligent agent Famou. Baidu Smart Cloud was also upgraded to a new full-stack AI cloud for large-scale intelligent agent applications. Li emphasized, 'As a platform company, Baidu has built full-stack capabilities of 'chip-cloud-model-agent' to support the surge in intelligent agent applications.'

Multiple Intelligent Agents Launched, Initiating a Personal Creation Revolution

In Li Yanhong's view, driven by the wave of intelligent agents, individual humans are also undergoing self-evolution, with Builders, Founders, and Creators achieving a trinity—every developer is also an entrepreneur and even more so a creator. At this conference, Baidu released three intelligent agent products for individual users.

The general-purpose intelligent agent DuMate (Chinese name 'Baidu Dazi') officially debuted. Since the release of its MVP version, the product has rapidly iterated and can now handle a variety of complex tasks. Currently, DuMate has achieved SOTA levels in multiple international authoritative Agent Benchmark evaluations. The DuMate mobile app is also launched, enabling real-time remote interconnection with the PC version.

Li said that DuMate is a unified entry point, a general-purpose intelligent agent. 'It seamlessly integrates Baidu's search AI API, Miaoda, Famou, and other capabilities into the same entry point. In this way, various tasks such as searching, coding, in-depth research, data analysis, and application creation can be collaboratively completed within a single system.'

The code intelligent agent Miaoda released the Miaoda App and an enterprise version. The Miaoda App is the first app dedicated to creating applications, with 90% of its code automatically generated by Miaoda. 'Code intelligent agents like Miaoda allow people who completely do not understand coding to directly generate applications, eliminating the threshold. At the same time, development costs are being compressed to extremely low levels, even close to zero, eliminating costs as well,' Li said. The software industry is being redefined, with 'one-time software' or 'daily disposable software' becoming reasonable. 'This market is likely to be amplified by 10 times as a result.'

In terms of digital human intelligent agents, Baidu Huiboxing has been upgraded to 'Baidu Yijing.' 'Digital humans are 'visible' intelligent agents,' Li said. Baidu Yijing has evolved from a digital human mainly used in live-streaming e-commerce scenarios to a full-scenario digital human platform featuring digital humans and including multiple forms such as live streaming, video, and real-time interactions, capable of generating highly interactive content lasting from a few minutes to several hours. The overseas version of Yijing is also launched simultaneously to assist Chinese companies in going global.

AI Empowers Corporate Organizations to Proactively Self-Evolve

'For AI to get things done is just the first step; whether it can judge whether it has done things well is the key to truly unlocking efficiency.' In Li's view, if intelligent agents can verify and form closed loops, they can evolve infinitely.

At this conference, the self-evolving decision-making intelligent agent Famou was upgraded to version 2.0, further directly targeting business experts and focusing on three major scenarios: production scheduling, process optimization, and logistics planning. Taking the logistics planning field as an example, in automated terminals, Famou helped the world's first set of terminal intelligent control systems, A-TOS, achieve a 10.21% absolute performance improvement. Previously, Baidu's Famou Agent 2.0 had topped the authoritative machine learning benchmark MLE-Bench, securing first place in 9 out of 15 of the most difficult questions, surpassing similar intelligent agents equipped with mainstream large models such as Claude-Opus-4.6.

Li said that AI will force changes in how companies operate. He proposed four points of reflection on the self-evolution of corporate organizations: first, more authorization and less control; second, faster alignment and fewer hierarchies; third, higher talent density and fewer mass tactics; fourth, more tasks and less division of labor.

The underlying capabilities supporting the aforementioned intelligent agent applications were also upgraded simultaneously. Previously, on May 9, Baidu released the new-generation foundational large model, ERNIE Large Model 5.1, which topped the domestic rankings on the LMArena text and search benchmarks. At this conference, Baidu Smart Cloud was comprehensively upgraded to a new full-stack AI cloud for large-scale intelligent agent applications, with comprehensive upgrades to the two major infrastructures of Agent Infra and AI Infra. It was also reported that the Kunlunxin fully domestic cluster has successfully supported the research and development of the ERNIE 5.1 series models. Meanwhile, the Tianchi 256-card super node based on Kunlunxin has been lit up and will be officially launched in June.

'Self-evolution is a systematic transformation for the AI era. Only those companies that dare to break inertia and continuously reshape themselves will have the opportunity to truly transcend cycles and establish new competitive advantages,' Li said. 'There are no bystanders in this era; we are all creators!'

Solemnly declare: the copyright of this article belongs to the original author. The reprinted article is only for the purpose of spreading more information. If the author's information is marked incorrectly, please contact us immediately to modify or delete it. Thank you.