Baidu’s Agent Strategy at Create 2026: DAA, New AI Agents, and the ERNIE 5.1 Stack

The idea is simple: instead of measuring how many people open an AI application or how many tokens a model generates, the industry should track how many AI agents actively complete tasks each day. According to Li, token usage represents cost input, while active agents better reflect the actual value created by AI systems.

Some coverage of the conference also reported Li predicting that global daily active agents could eventually exceed 10 billion, though that figure appears mainly in secondary reporting and should be treated as a projection rather than a confirmed company benchmark.

This shift in measurement reflects Baidu’s broader thesis: the next stage of AI competition will be defined not just by smarter models but by agents capable of executing complex workflows autonomously.

New AI agents unveiled at Create 2026

The conference introduced or upgraded several AI agents designed for different roles in Baidu’s ecosystem.

DuMate: a general‑purpose AI agent

One of the central announcements was DuMate, a general‑purpose intelligent agent designed to act as a digital assistant capable of handling tasks across multiple contexts.

The agent is intended to behave more like a digital employee than a chatbot, executing workflows and coordinating tasks rather than simply responding to queries.

Miaoda (MeDo): coding and application‑building agent

Baidu also introduced Miaoda, a coding‑focused agent available in both app and enterprise versions.

The system is designed to assist with software development and application creation. Some reports note that the product can generate a large share of its own code during development workflows.

In some English‑language references the product is called MeDo, though the available evidence does not clearly establish whether this is a translation, alternate brand, or separate version of the tool.

Baidu Yijing: digital human agent platform

Another product highlighted at Create 2026 was Baidu Yijing, an upgraded digital human platform.

The platform focuses on creating lifelike AI avatars capable of communication and presentation tasks. Reports indicate the system supports multiple languages with synchronized lip movement designed to match natural speech patterns.

Famou Agent 2.0: self‑evolving decision agent

Baidu also unveiled Famou Agent 2.0, an upgraded version of its self‑evolving agent system.

The agent is designed for complex reasoning and decision‑making tasks and is described as capable of improving through iterative learning. According to reported benchmarks, Famou Agent 2.0 performs strongly on difficult agent‑evaluation tasks such as MLE‑Bench.

The infrastructure behind Baidu’s agent ecosystem

Baidu emphasized that agents require a new kind of infrastructure stack, combining chips, cloud platforms, foundation models, and agent frameworks.

ERNIE 5.1 foundation model

At the conference, Baidu introduced ERNIE 5.1, the latest generation of its foundation model family.

The model is intended to power the new agent ecosystem while improving training efficiency and reducing operating costs. In Baidu’s architecture, ERNIE models serve as the reasoning layer beneath applications and agents.

Baidu AI Cloud for agent applications

Baidu also highlighted its full‑stack AI Cloud, which is positioned as a platform specifically optimized for building and running large‑scale AI agents.

The cloud infrastructure is designed to support enterprise deployment of agents and integrate model capabilities, orchestration tools, and application services.

Chip‑to‑agent architecture

Baidu described its broader strategy as a vertically integrated stack linking:

• AI chips (such as Kunlun‑series accelerators)
• cloud infrastructure
• foundation models like ERNIE
• and agent‑based applications

This “chip‑cloud‑model‑agent” architecture is intended to create an ecosystem where agents become the primary interface between users and AI systems.

Why Baidu is focusing on agents

A key message from Create 2026 was that AI is moving from knowledge generation to task execution.

Robin Li summarized the shift by arguing that the real measure of AI value will be whether systems can complete tasks for users, not just answer questions.

That perspective explains Baidu’s emphasis on agents, which combine large language models with tools, memory, and workflow automation to perform multi‑step actions.

The bigger picture

The announcements at Create 2026 show Baidu repositioning itself as a full‑stack AI platform built around agents. Instead of competing only on model benchmarks, the company is betting that the next wave of AI adoption will come from autonomous systems embedded across software, enterprises, and digital services.

If the company’s vision plays out, success in the AI industry may increasingly be measured not by how many people talk to AI—but by how many agents are working on their behalf every day.