From August 6 to 8, Yonyou held the 2026 Global Business Innovation Conference in Nanjing. When I went through the press release afterward, I was caught by an inconspicuous abbreviation — AIX.
In his opening speech, Wang Wenjing defined this acronym: enterprise software is moving from "XAI" to "AIX". XAI means "humans execute, AI assists"—you sit in the driver's seat while AI reads the navigation beside you; AIX means "humans authorize, AI executes"—you tell AI the destination, and it plans the route and steps on the accelerator itself, while you keep an eye on the dashboard.
The change in position of a single letter reflects the entire industry's redefinition of AI's role. Alongside the release of BIP6—63 vertical industry agents, 623 intelligent capabilities, and over 26,000 application interfaces—the message this round of releases aims to convey is clear: China's leading ERP vendors have officially embedded "AI that gets work done" into the product foundation.
I. A Shift of One Letter: From Co-pilot to Authorized Executive
Yonyou divides this path into three stages, which in plain language roughly means:
| Stage | The Role of AI | Who is working | Typical form |
|---|---|---|---|
| XAI (Copilot Era) | Assistant | Human-led, AI-assisted | Q&A, content generation, process guidance |
| AIX (Agentic Era) | Executor | AI-led, human-authorized + supervised | Autonomous planning, interface invocation, and execution of business closed-loop |
| AI-native era | Leader | AI-led, human-supervised | System-level intelligent agent autonomous operation |
This division aligns perfectly with the industry trends I have observed over the past month or so. The Odoo 20 roadmap has changed AI from a "standalone application" to task-level Agents embedded in "accounting, website, tickets, and timesheets"; Kingdee officially released the enterprise-level AIOS "Lingji" at the Hong Kong and Macau site on August 4, providing complete capabilities for agent development, orchestration, operation, and governance; Chanjet's dual-engine "Xiaochang + Changlongxia" released last week directly split "AI that can chat" and "AI that can work" into two products.
Different vendors call it by different names—Agentic, AIOS, dual-engine, AIX—but the core is the same: AI is no longer content with answering questions; it is beginning to take action and get things done.
II. What BIP6 includes: The logic behind five AI-native new products
Looking at the BIP6 release lineup, it's clear that Yonyou isn't just "stacking features" this time, but rebuilding a layer of architecture based on "data-model-platform-application":
- YonWork Enterprise AI Workbench — the unified AI entry point for employees, bringing work scattered across finance, HR, procurement, and supply chain into a single conversational interface;
- YonData Enterprise AI Data Agent — enabling AI to directly analyze and support decision-making on enterprise data assets, rather than working through a reporting tool layer;
- YonCode Enterprise-level AICoding Platform — AI on the development side, paired with "agent software," solving the question of "in the AI era, who writes the software and how to write it faster";
- BIP Penetrating Supervision Platform — this deserves a bit more explanation: it serves as a safety net for "AI execution" — what AI did, why it did it, and whether there is a trace left after it acts, all of this needs to be visible to someone.
- YonOnto Enterprise Ontology Platform — consolidates the enterprise's business objects, relationships, and standards into machine-understandable "ontologies," essentially providing AI with a corporate dictionary.
Looking at these five components together, the logic becomes clear: YonOnto handles "whether AI recognizes your enterprise," YonData handles "whether AI has data available," YonWork handles "how employees collaborate with AI," YonCode handles "development efficiency in the AI era," and penetrating supervision handles "whether AI execution can be audited."
The supporting implementation figures are also out: Baigong Steel, relying on BIP, achieved a cost reduction of 82 yuan per ton of steel, a 7% increase in job-based workforce efficiency, and an 18% increase in the sales ratio of specialty steel; Yonyou has helped over 300 large and medium-sized enterprises replace systems from international vendors, with migration cycles shortened by more than 85% and data accuracy at 99.9%.
My assessment is: the real highlight of BIP6 is not how smart a particular agent is, but the two infrastructure components—"penetrating supervision" and "enterprise ontology"—which answer the same question: no matter how fast AI executes, who takes responsibility when something goes wrong, and how can it be traced back. Without this layer, the AI execution capabilities mentioned earlier cannot be promoted within enterprises.
III. Not Just Yonyou: Global ERP Is Shifting at the Same Point in Time
Looking at the timeline, it becomes clear that this round of "executor pivot" was a collective action, not one company jumping ahead:
| Manufacturer | Recent Actions | Position on "AI Execution" |
|---|---|---|
| Yonyou BIP6 | Published August 6-8 | AIX Strategy: Human Authorization, AI Execution, Dual-Mode Coexistence |
| Kingdee Lingji | August 4th Hong Kong and Macau Station released AIOS | AI-native operating system, full lifecycle governance of intelligent agents |
| Odoo 20 | Released in Brussels, September 24-26 | AI shifts from being a feature to the underlying layer, with task-level Agents autonomously executing multiple steps. |
| Chanjet | August 6 Dual Intelligent Engine | Xiaochang (Dialogue) + Chang Lobster (Autonomous Execution) Dual Engine |
| Oracle / Epicor | Released successively in July-August | Agent embedded in finance and supply chain, natural language directly reaches business operations |
Five vendors, five product forms, all pointing to the same conclusion: In the second half of enterprise software competition, the battle is over "AI execution capability + governance capability." Previously, the competition was about who had more complete modules and a more user-friendly interface; now it is about whose intelligent agents can truly run end-to-end within business operations, while also leaving an audit-ready trail after execution.
IV. Three Words for Selectors and Implementers
First sentence: When evaluating AI capabilities, stop asking "what can it generate" and instead ask "what can it execute, and does it leave a trace." Have the vendor demonstrate an end-to-end business loop on the spot—such as automatically creating a quote, approval, shipment, and invoice from a sales opportunity—and see whether it actually calls business data and whether every step leaves an auditable record. If it cannot demonstrate the closed loop, no matter how strong its Q&A capability is, it is just a co-pilot.
Second sentence: Write the "data foundation" into the selection contract. In the AIX era, the cleanliness of master data directly determines whether AI agents dare to take on their roles. When selecting a vendor, the workload and responsibility boundaries for data cleaning, standardizing definitions, and ontology modeling must be clearly written out. This is not an area to skimp on—if you cut corners here, you will pay for it later in the delivery timeline.
Third sentence: Governance first, with permissions and auditing following. Before authorizing AI to execute, first define clearly: which actions AI can autonomously perform, and which require manual confirmation; whether every operation by AI has logs and can be traced back. It is recommended to start with small-scale, low-risk scenarios (data queries, report generation, document entry) as pilots, and only expand the scope of authorization after the governance mechanism is proven to work.
Also, a word of caution: don’t be carried away by the "fully automated, zero manual" hype. What enterprises need in the AIX stage is a stable triangle of "human authorization + AI execution + human supervision," not a hands-off approach. The so-called "AI-native" approach, when applied within an enterprise, first and foremost means "governance-native"—autonomous execution without governance is like installing an engine without brakes in a company. In Wang Wenjing’s speech, the concept of "intelligent dual-mode" is actually more pragmatic: processes with clear rules continue to use process software, while tasks requiring judgment and execution are handed over to intelligent agents, walking on two legs.
At the end, I want to say: from XAI to AIX, a shift of one letter is the industry's answer to "what role should AI play." For those in charge of enterprise digitalization, this is both an opportunity and a lesson—the opportunity lies in AI finally being able to transform from a "reporting tool" into a "worker"; the lesson lies in that before granting authorization, one must first nail down the three things: data, permissions, and auditing. The direction is already clear, and what comes next is a competition of who executes more steadily.
