Robin Li wants to change how the technology industry keeps score on artificial intelligence. In a corporate announcement distributed in late May 2026, the Baidu CEO introduced a metric he calls Daily Active Agents, or DAA, arguing that the real measure of AI progress is not how large a model is but how many autonomous software agents are completing tasks for people every single day.
The concept borrows directly from the mobile era’s Daily Active Users figure, the number that once determined app valuations, advertising budgets, and venture capital bets. Li’s pitch: swap “users” for “agents,” and you get a scoreboard better suited to a world where AI does not just answer questions but books flights, files reports, and negotiates supplier contracts on its own.
What Baidu actually announced
Baidu’s official statement, released through PR Newswire, describes the start of what the company calls an “agent era.” In this framing, the primary unit of AI value is no longer a large language model sitting behind a chat window. It is a task-executing application that initiates actions, adapts based on outcomes, and operates with enough autonomy that a user does not need to babysit every step.
The distinction matters because most consumers still experience AI as a question-and-answer tool. You type a prompt, a chatbot responds, and the loop resets. Baidu is describing something different: agents embedded across its search engine, productivity apps, cloud services, and its Apollo autonomous-driving platform that carry out multi-step workflows without waiting for fresh instructions at each stage.
By tying success to DAA, Li is making a strategic argument that favors Baidu’s own product roadmap. The company has spent the past two years building agent infrastructure on top of its Ernie foundation model, and redefining the scoreboard around agent engagement positions Baidu as a leader before rivals have agreed to play the same game.
Why the DAU-to-DAA analogy resonates
During the smartphone boom, DAU became the number that mattered most. A high DAU told investors that people returned to an app daily, which attracted advertisers, shaped feature priorities, and justified billion-dollar valuations. Li is betting that DAA can exert the same gravitational pull on AI development. If the industry adopts it, companies would be rewarded not for publishing impressive benchmark scores or training ever-larger models but for shipping agents that people actually rely on, day after day.
That logic has appeal beyond Baidu. Several enterprise software companies have already built platforms around the idea that AI should be measured by tasks resolved, not tokens generated. Venture investors have published frameworks arguing that AI agents represent the next major software cycle. Li’s DAA proposal fits squarely within that broader current, even if no Western competitor has publicly endorsed his specific metric.
What Baidu has not disclosed
For all the ambition behind the announcement, Baidu left out the numbers that would make DAA more than a branding exercise. The company did not reveal how many agents are active today, what internal DAA figures look like, or how its agent engagement compares with previous quarters. Without a baseline, investors and analysts have no way to track progress or benchmark Baidu against competitors. Baidu has also not disclosed updated monthly active user figures for Ernie Bot or provided any data on the size of its agent user base, leaving key questions about real-world adoption unanswered.
Equally important, Baidu has not published a methodology. The announcement does not define what counts as an “active” agent, whether a background process qualifies, or how long a task must run before it registers. Those details matter. If every company invents its own definition, DAA becomes as murky as the “monthly active user” figures that plagued social-media earnings reports for years. No independent analyst coverage available in public sources has yet evaluated whether DAA is a rigorous metric or primarily a marketing construct.
The “self-teaching” claim also lacks technical backing. Baidu describes its agents as capable of learning from interactions, but no whitepaper, architecture disclosure, or peer-reviewed research accompanied the release. Whether these agents improve through reinforcement learning from human feedback, retrieval-augmented generation, scripted decision trees, or some combination remains unclear from public materials alone.
The competitive landscape Baidu is trying to reshape
Baidu is not the only company racing toward autonomous agents. Google has demonstrated agent prototypes under its Gemini umbrella, including research efforts that let AI navigate web browsers and execute multi-step tasks. OpenAI has expanded its API with function-calling and tool-use capabilities designed to let developers build agent workflows. Anthropic has published research on “computer use” agents that interact with desktop software on a user’s behalf.
None of these rivals has proposed a single engagement metric equivalent to DAA, which gives Li’s framing a first-mover quality in the narrative war even if the underlying technology is contested. The risk for Baidu is that Western competitors, with larger foundation models and broader developer ecosystems, could adopt agent-centric metrics on their own terms and render DAA a footnote.
Inside China, the competitive picture is just as crowded. ByteDance, Alibaba, and Tencent have each launched agent-oriented products, and Beijing’s regulatory framework for generative AI, in effect since August 2023, imposes disclosure and safety requirements that could shape how any company reports agent activity. Baidu’s push for DAA will have to navigate those rules alongside market competition.
Where DAA stands without independent validation
If DAA gains traction, even informally, it could shift how AI products are designed. Instead of optimizing for chat quality or creative text output, development teams would focus on reliability, task completion rates, and the kind of trust that makes someone comfortable letting an agent handle a purchase or schedule a medical appointment without constant oversight.
For developers weighing whether to build agent-based products, the practical signal is straightforward: one of China’s largest technology companies is reallocating strategic resources toward autonomous agents and away from purely model-centric competition. That does not guarantee the “agent era” has arrived, but it does indicate where at least one major player with significant infrastructure sees commercial opportunity in the months ahead.
The gap between Li’s vision and current reality, however, remains wide. No public dataset tracks DAA across the industry. No independent auditor has validated Baidu’s agent capabilities. The PR Newswire release remains the sole primary source for the announcement, and Baidu has not supplemented it with earnings-call commentary, technical documentation, or third-party endorsements. And the core technical challenge of building agents that genuinely improve their own performance over time, rather than following pre-programmed paths, is still one of the hardest open problems in AI research. Until concrete data, shared standards, and external evaluations catch up, DAA is best understood as an ambitious proposal about how AI value might be measured in the coming months, not a proven scoreboard for today.
More from Morning Overview
*This article was researched with the help of AI, with human editors creating the final content.
