Baidu's "ShengSuan" Platform: Are AI Agents Finally Entering Enterprise Core Business?
At Create2026 Baidu AI Developer Conference (May 13-14), Li Yanhong introduced DAA as the AI era's new metric and launched the enterprise data intelligence platform "Baidu ShengSuan" — targeting the ultimate bottleneck preventing AI Agents from entering core enterprise operations.
DAA: The "Daily Active Users" Moment for AI
Li Yanhong introduced a new concept at the Create2026 opening ceremony: DAA (Daily Active Agents).
His argument: Token consumption only measures cost input, not actual value output. The real metric for an AI platform's success should be how many Agents are actually completing tasks and delivering results.
Analogous to DAU (Daily Active Users) in the mobile internet era, DAA aims to be the corresponding metric for the AI era. Li Yanhong even predicted that global DAA could exceed 10 billion in the future.
The logic is clear: when Agents evolve from "chat tools" to "task executors," their value is no longer measured by how many times they're called, but by how many effective tasks they complete.
What Exactly Is "Baidu ShengSuan"?
During the conference, Baidu Intelligent Cloud officially launched its enterprise data intelligence platform — "Baidu ShengSuan" (which literally means "winning calculation").
The name is telling: "ShengSuan" = a strategy with a good chance of winning. The platform's goal is equally direct — solving the bottleneck that prevents AI Agents from handling mission-critical enterprise decisions: insufficient accuracy and inability to participate in core decision-making.
Core Capabilities
Built on enterprise business ontology, Baidu ShengSuan currently covers 20+ industries with 370+ relational and multimodal operators.
In deployed core business scenarios, Agent decision accuracy in complex situations has improved to 99% — a critical figure. Enterprise core operations (financial risk control, medical diagnostic assistance) have zero tolerance for errors.
Three Business Graphs
The platform's technical innovation can be summarized as the "Three Business Graphs":
| Graph | Function | Problem Solved |
|---|---|---|
| Business One-Map | Structures core business objects (orders, inventory, products) | Enables LLMs to understand enterprise entity relationships |
| Business Logic One-Map | Converts expert decision-making experience into executable rules | Teaches Agents "how an experienced employee works" |
| Business Execution One-Map | Directly executes operations (orders, returns) via APIs | Evolves Agents from "advisory only" to "direct action" |
The third graph is particularly important — it suppresses LLM hallucinations at the source. Agents no longer just provide recommendations; they can directly invoke enterprise systems to complete operations, with auditable execution records at every step.
"Chip-Cloud-Model-Agent": From Model Company to AI Infrastructure Provider
Create2026 revealed another significant signal: Baidu is systematically repositioning itself as an AI infrastructure provider, not just an LLM company.
This is reflected in the "Chip-Cloud-Model-Agent" four-layer architecture:
Chip Layer: Kunlun Core P800 (scale-verified)
↓
Cloud Infrastructure Layer: Tianchi 256 Super Node (launching June)
↓
Model Layer: Wenxin 5.1 (trained at only 6% of industry cost)
↓
Agent Layer: DuMate, Miaoda, Baidu Yijing, Famou — full-stack Agent product matrix
Each layer prepares for Agent-scale deployment. Chips ensure independent and controllable computing power; cloud platforms ensure elastic inference; models ensure capability foundation; the Agent layer directly faces application scenarios.
The Controversy: Real Breakthrough or Another "Concept Product"?
Industry reactions to Baidu ShengSuan's launch have been polarized.
Supporters argue: - 99% accuracy + Three Business Graphs represents a critical breakthrough moving Agents from "demo" to "production" - "Chip-Cloud-Model-Agent" full-stack layout distinguishes Baidu from competitors who only do models or only do applications - If the DAA metric gains industry adoption, Baidu controls the discourse on "what defines success in the AI era"
Skeptics counter: - 370 operators sounds impressive, but do they truly adapt to complex, variable enterprise scenarios across China? - 99% accuracy is measured in "already deployed scenarios" — whether it generalizes remains unknown - Baidu has launched many enterprise AI products before — what's different this time?
One deeper question: Are enterprises willing to map their core operational logic onto Baidu's platform? The essence of business graphs is structuring an enterprise's most valuable operational knowledge — both a value proposition and a risk for many companies.
Implications for Nizwo Users
Baidu ShengSuan and Nizwo are not competitors — they are complementary.
Baidu ShengSuan addresses cloud-based, enterprise-grade, complex business decision scenarios; Nizwo addresses local, private, data-secure scenarios.
A typical enterprise AI deployment might look like this: - Core business decisions (requiring multi-system integration, historical data training) → Baidu ShengSuan (cloud) - Daily office automation, local knowledge base Q&A, sensitive data processing → Nizwo (local)
The value of local deployment becomes even more apparent as cloud platforms like Baidu ShengSuan grow more powerful — the more enterprises rely on AI for core decisions, the more they need to ensure "critical data never leaves the premises."
In one sentence: Whether Baidu ShengSuan can actually "win" depends less on how impressive the launch event is, and more on how many enterprises are genuinely running their core business operations on it a year from now. If DAA grows, then Baidu truly has a winning calculation.
The AI Frontier column continuously tracks the latest AI industry developments, decoding the business logic behind technology.