At Baidu’s annual Create 2026 developer conference in Beijing, CEO Robin Li delivered a keynote that crystallized a shift many in the industry have been sensing: the era of competing on model size is over — the era of agent competition has begun.
”Thinking Is No Longer Enough”
Li’s central argument was blunt. Users don’t care about benchmark scores, parameter counts, or reasoning chain lengths. What they want is task completion — and that requires a fundamentally different product paradigm.
“Users are no longer impressed by models that think. They want systems that act — that book flights, negotiate contracts, debug code, and manage supply chains without human intervention.”
Baidu’s Agent Stack
Baidu unveiled a comprehensive agent development platform at the conference:
| Component | Purpose |
|---|---|
| AgentBuilder 3.0 | Low-code platform for creating multi-step agents |
| Ernie Agent Runtime | Execution environment with tool calling, memory, and planning |
| Agent Store | Marketplace for pre-built enterprise agents |
| Agent Monitor | Observability suite for tracking agent behavior and costs |
| WuKong Enterprise | Full-stack agentic workflow engine for large organizations |
The Agent-First Thesis
Li outlined why he believes agents — not models — will define the next wave of AI value:
- Models are commoditizing: Performance gaps between top LLMs are narrowing rapidly
- Agents create lock-in: Once embedded in workflows, agents are harder to replace than model APIs
- Revenue per user is higher: Agents that perform tasks command transaction-based pricing, not just API token fees
- Enterprise readiness: Businesses care about outcomes, not inference quality in isolation
China’s Agent Landscape
Baidu’s pivot reflects a broader trend in China’s AI ecosystem. While US companies have led in foundational model development, Chinese firms are increasingly focused on deployment-first strategies:
- Alibaba’s WuKong agents are managing logistics for millions of parcels daily
- Tencent has deployed agent-based customer service across WeChat’s entire ecosystem
- ByteDance is using internal agents for content moderation at scale
- Baidu itself claims over 200 million monthly active agent interactions through its Ernie ecosystem
The Global Implications
Li’s keynote arrives at a pivotal moment. Just days before a high-stakes Trump-Xi summit, the AI competition narrative is shifting from “who has the biggest model” to “who has the most effective autonomous systems.” This reframing has significant implications for:
- Enterprise procurement: Companies may increasingly evaluate AI vendors on agent capability rather than model benchmarks
- Regulation: Autonomous agents acting in the real world raise different policy questions than chatbots
- Workforce impact: Agent deployment at scale directly displaces specific job functions, not just augments them
The Bottom Line
Baidu’s Create 2026 message is clear: the foundational model race was the qualifier. The agent race is the main event. And China intends to compete aggressively.
Source: TechNode, South China Morning Post, Baidu
