Baidu's Robin Li: DAA Replaces Token Count as AI Metric

Baidu founder Robin Li has declared that the artificial intelligence industry must abandon 'token consumption' as its primary metric of success. He argues that Daily Active Agents (DAA) is the true measure of an AI ecosystem’s prosperity and utility.

This pivotal statement was made during the opening ceremony of the Create 2026 Baidu AI Developer Conference in Beijing on May 13. The shift in perspective marks a significant evolution in how tech giants evaluate the impact of generative AI technologies.

Why Token Metrics Fail to Capture Value

The current AI landscape heavily relies on tracking the volume of tokens processed by large language models. Companies often boast about processing trillions of tokens to demonstrate scale. However, Li contends that this metric fundamentally measures cost rather than value. It reflects input expenditure, not the actual output or benefit delivered to users.

Token consumption is a proxy for computational load. It does not indicate whether the AI successfully solved a problem or completed a task. A model can consume vast amounts of tokens while generating irrelevant or incorrect information. Therefore, focusing solely on tokens provides a distorted view of technological progress.

Li emphasizes that the end game of AI is not merely processing text but delivering actionable results. The transition from chat-based interfaces to autonomous agents requires a new yardstick. This new standard must quantify how many intelligent systems are actively working on behalf of humans. It shifts the focus from raw computation to tangible productivity.

Defining the New Standard: Daily Active Agents

Daily Active Agents (DAA) serves as the AI-era equivalent of Daily Active Users (DAU) in the mobile internet age. In the mobile era, DAU measured how many people engaged with an app daily. Similarly, DAA measures how many autonomous agents are executing tasks and delivering outcomes every day.

An agent differs from a simple chatbot. It possesses the ability to plan, execute, and verify actions across various digital platforms. When an agent books a flight, writes code, or analyzes data without constant human intervention, it counts toward DAA. This metric captures the autonomy and utility of the system.

Li argues that DAA is closer to the essence of economic value. It represents the number of digital workers contributing to human goals. A high DAA indicates a thriving ecosystem where AI tools are integrated into daily workflows. It signals that users trust these systems enough to delegate critical tasks to them.

The Evolution of the Human-AI Partnership

The rise of autonomous agents is driving a profound change in the role of human developers. Li describes a convergence of three distinct roles: Builder, Founder, and Creator. This triad represents the new identity of the modern technologist in the AI age.

Previously, these roles were often separate. Developers built tools, entrepreneurs founded companies, and artists created content. Today, AI lowers the barriers to entry for all three activities. A single developer can now build complex applications, launch a startup, and generate creative assets simultaneously.

This self-evolution empowers individuals to achieve more with fewer resources. AI acts as a force multiplier, allowing small teams to compete with large corporations. The distinction between coding and product management blurs as natural language becomes the primary interface for building software.

Key Takeaways from the Create 2026 Conference

• Metric Shift: The industry should prioritize DAA over token count to measure true utility.
• Agent Autonomy: Success is defined by agents completing tasks and delivering results independently.
• Role Convergence: Developers are becoming builders, founders, and creators simultaneously.
• Ecosystem Health: A high DAA indicates a robust and valuable AI platform for users.
• Cost vs. Value: Token usage measures input costs, whereas DAA measures output value.
• Industry Consensus: Major players like Horizon Robotics CEO Yu Kai are aligning with this vision.

Industry Context and Competitive Landscape

The push for DAA reflects a broader trend among Western and Chinese tech leaders. Companies like OpenAI, Anthropic, and Microsoft are increasingly focusing on agentic workflows. They recognize that future revenue will come from automated services, not just API calls for text generation.

In the United States, enterprises are struggling to integrate LLMs into core business processes. Many pilot projects fail because they do not deliver measurable ROI. By shifting the metric to DAA, Baidu aims to highlight practical applications. This approach encourages developers to build tools that solve real-world problems.

Western competitors are also exploring similar concepts. For instance, Microsoft’s Copilot ecosystem focuses on user engagement and task completion rates. However, Baidu’s explicit framing of DAA provides a clear strategic direction. It challenges other firms to redefine their success metrics beyond mere compute usage.

This competition drives innovation in agent reliability and safety. As agents take on more responsibility, the need for robust guardrails increases. The industry must ensure that these digital workers operate within ethical and legal boundaries. Trust becomes the most valuable currency in the agent economy.

Practical Implications for Developers and Businesses

For software engineers, the shift to DAA means changing how they design applications. Instead of optimizing for low latency or minimal token usage, developers should focus on task completion rates. How often does the agent successfully finish the job? How much human oversight is required?

Businesses must also adapt their key performance indicators. Marketing teams should track how many customers interact with autonomous support agents. Product teams should measure the frequency of agent-initiated actions. These metrics provide deeper insights into customer satisfaction and operational efficiency.

Investors should look for startups with high DAA growth. A company with millions of active agents demonstrates strong product-market fit. It suggests that users rely on the technology for critical daily tasks. This reliance creates sticky ecosystems and recurring revenue streams.

Looking Ahead: The Future of Agentic AI

The transition to an agent-centric model will accelerate over the next 5 years. We can expect to see more sophisticated agents capable of handling complex, multi-step workflows. These systems will integrate seamlessly with existing enterprise software stacks.

Regulators will likely begin to scrutinize agent activity. Questions around liability and accountability will arise as agents make autonomous decisions. Clear guidelines will be necessary to ensure safe deployment.

Ultimately, the goal is a symbiotic relationship between humans and machines. Humans provide intent and creativity, while agents handle execution and analysis. By adopting DAA as the standard metric, the industry can better track progress toward this vision. It ensures that AI development remains focused on delivering genuine human value.

The Create 2026 conference highlighted that the era of passive AI interaction is ending. We are entering a phase where AI works alongside us, actively shaping our digital and physical worlds. Embracing DAA is the first step toward measuring this new reality accurately.