Baidu Unveils DuMate, Miaoda Agents for "AI Agent Era" Shift

Baidu has introduced new AI tools called DuMate and Miaoda, aiming to move from just building bigger language models to creating smart AI agents that can complete tasks. CEO Robin Li said that in the future, people may pay for AI based on what it can do, not just what it knows. Analysts suggest this shift could change how tech companies are valued and how they spend money. Li also suggested a new way to measure success in AI, focusing on how many tasks these agents finish daily, but the timeline for reaching these goals is still unclear. Whether Baidu's new approach will work worldwide may depend on how many people use it, government rules, and how competitors respond.

Baidu is ushering in an "AI agent era" with the unveiling of its DuMate and Miaoda AI agents, a strategic pivot announced by CEO Robin Li at the Baidu Create 2026 conference. This move shifts the industry's focus from building larger models to creating autonomous agents that execute tangible tasks. Li argued that AI's value will now be measured by what it can do, not just what it knows, according to the TechNode report.

From Models to Agents

Baidu's strategic shift from models to agents prioritizes creating autonomous AI that can execute complex tasks from start to finish. Instead of competing on large language model parameters, the focus is now on building practical tools that learn, verify results, and can refine their own workflows.

New Toolkits: DuMate and Miaoda

To realize this vision, Baidu launched a portfolio of agent-native tools built on its full-stack AI Cloud and ERNIE 5.1 model. The new offerings include:
- DuMate: A general-purpose agent for mobile and desktop that can read screens, operate software, and integrate with business systems with real-time synchronization.
- Miaoda: A coding agent that reportedly generated 90% of its own mobile app code, enabling non-developers to build software.
- Baidu Yijing: A digital human platform.
- Famou Agent 2.0: A tool for industrial optimization.

Daily Active Agents Metric

Reflecting this new paradigm, Li proposed "Daily Active Agents" (DAA) as a key performance indicator to succeed Daily Active Users (DAU). He argued that while tokens measure cost, DAA measures tangible output. Li projects significant growth in agent adoption, suggesting platform success will soon be judged by the number of tasks completed by autonomous agents.

Implications for Industry Observers

This agent-centric approach could potentially disrupt the creator and low-code markets. DuMate's unified gateway competes with subscription assistants, while Miaoda challenges no-code platforms by making disposable software economically feasible. However, Li warned that the primary hurdle for enterprise adoption is data integration. Success requires creating a closed, verifiable loop for agents to operate within - a significant infrastructural challenge that suggests the next AI race will be won by measurable task completion, not parameter counts.

What is Baidu's "AI Agent Era" and why does it matter for creators?

At Baidu Create 2026 CEO Robin Li declared the industry is moving beyond the "model race."
The new battlefield is agent execution - whether an AI can finish tasks, not just talk about them.
Baidu therefore launched four agents, led by DuMate (general-purpose desktop + mobile agent) and Miaoda (zero-code app generator).
According to industry reports, early adopters have begun monetizing Miaoda-built tools, hinting at the potential size of the emerging "disposable-software" market.

How do DuMate and Miaoda actually work?

DuMate
- Reads your screen, drives Office, browsers and line-of-business apps end-to-end
- 24/7 mobile-PC sync with real-time synchronization
- Integrates Search, coding, data analysis and file management in one chat thread
- Demonstrates strong performance on agent benchmarks

Miaoda
- Type a sentence → runnable mobile or web app; 90% of its own codebase was self-generated
- Two editions: Miaoda App (consumers) and Miaoda Enterprise (teams)
- Significantly reduced development costs, letting non-developers create micro-apps

Will these tools compete with existing creator platforms?

Yes, on three fronts at once:
1. Low-code market (Bubble, Webflow, Zapier) - Miaoda removes code entirely
2. AI-assistant subscriptions (ChatGPT Plus, Claude Pro) - DuMate bundles search, coding and automation for a single login
3. Content & live-commerce suites (YouTube, TikTok) - Baidu Yijing digital-human studio now supports 12 languages and script-to-video workflows

Analysts expect Daily Active Agents (DAA) - Baidu's proposed success metric - to replace "tokens" the way DAU replaced page-views, forcing rivals to prove real task completion rather than chat volume.

What infrastructure changes are needed for an "agent-native" world?

Robin Li warns that clouds, semiconductors and even security stacks were built for human clicks, not machine loops.
Baidu is therefore rebuilding its full stack - from ERNIE 5.1 foundation model to custom silicon - so agents can call memory, permissions and workflows without human hand-offs.
Companies that want similar speed will need closed-loop data, verifiable outputs and APIs that agents can trust - a bigger integration lift than simply buying a model license.

When can global users try the products and what comes next?

DuMate mobile & desktop and Miaoda App launched in China on May 13, 2026
International availability and specific global rollout plans remain to be announced
Li proposed Daily Active Agents (DAA) as a defining metric for the industry's future growth

For creators, the takeaway is simple: if your workflow still moves through separate search, code, edit and upload steps, an all-in-one agent may soon do it in one prompt while you sleep.