From Token to DAA: Baidu Recalculates the Agent Equation

The AI industry is searching for new metrics.

As agents gradually enter real-world business scenarios, the industry is beginning to focus on how many agents are truly being utilized and continuously creating value.

Two months ago, Baidu founder Robin Li first proposed the metric DAA (Daily Active Agents), aiming to measure the development of AI applications by the number of agents that are genuinely active, complete tasks, and generate value on a daily basis.

On July 17, at the 2026 World Artificial Intelligence Conference (WAIC), IDC released the "DAA Research Report" for the first time, further systematizing this metric.

The report shows that the global number of active agents is expected to grow from 28.6 million in 2025 to 79.4 million in 2026, reaching 2.216 billion by 2030.

This also serves as a central theme behind Baidu's showcase of multiple agent products at this year's WAIC. From Baidu Dazi, designed for personal office productivity, to products like Wenku, Netdisk, and Miaoda, Baidu is seeking more high-frequency entry points for agents.

This is only one aspect of DAA.

For enterprises, the standard for measuring an agent's value is more direct: Can it integrate into core business operations and deliver calculable efficiency improvements or operational gains?

The Decision-Making Agent Fa Mou 2.0 represents Baidu's exploration of how industries can apply AI.

Unlike writing, summarization, and information organization in office scenarios, Fa Mou targets complex decision-making scenarios such as enterprise production scheduling, process optimization, and logistics dispatching. It aims to help enterprises find optimal solutions under numerous constraints.

Currently, Fa Mou has covered industries including ports, logistics, industrial manufacturing, chemicals, energy, and finance, while also entering AI for Science scenarios such as agricultural breeding and electromagnetic research.

According to Li Annan, head of Baidu Intelligent Cloud's Fa Mou product, who spoke to Wallstreetcn, Baidu's Fa Mou currently primarily serves two types of enterprises:

One type is large key account (KA) clients, including major enterprises in manufacturing, energy, and ports. These clients have large-scale operations, where even a few percentage points improvement in production efficiency can yield significant economic benefits. Therefore, they mostly adopt private deployment models, typically engaging in project-based collaborations with a standard service cycle of approximately three years per project.

The other type consists of small industry leaders and regional leading enterprises. These enterprises usually have annual outputs between RMB 1 billion and RMB 10 billion, making them better suited for public cloud subscription models. Compared to private deployment, public cloud offers lower costs and reduces the barrier for medium-sized enterprises to experiment with agents.

"Medium-sized enterprises are showing good acceptance now, because public cloud pay-as-you-go pricing is much cheaper than private deployment, and results are visible," Li Annan pointed out.

Once agents enter enterprise production systems, their value must be measured by efficiency improvements, cost reductions, and business gains.

Whether DAA can become the new metric for the AI era is drawing significant attention.