May 20 was quite an interesting day. Kingdee released "Lingji," an enterprise AI operating system; on the same day, Yonyou opened exclusive invitation codes for "YonClaw." These two old rivals, locked in battle for thirty years, almost simultaneously showed their hands on the AI track.
Three weeks later, Tencent unveiled the WorkBuddy Enterprise Edition and Agent Suite at the Cloud AI Industry Application Conference, positioning it as an "AI-native enterprise collaboration platform." Going back half a month earlier, SAP announced at the China Summit that it had deployed 224 AI agents and 51 business assistants.
Looking at these things together, I believe the enterprise AI landscape in June 2026 is not about who wins or loses—but rather that the entire industry is undergoing a structural "time lag." Only by understanding this time lag can we grasp the direction of enterprise AI over the next two years.
The problems of Kingdee and Yonyou are not their own
Let me first talk about something many people haven't noticed: Yonyou's market value has shrunk by over 137.2 billion yuan from its 2020 peak, a decline of about 80%; Kingdee has shrunk by over 106.7 billion Hong Kong dollars from its peak, a decline of over 80%.
Yonyou
-137.2 billion market value shrinkage, a decline of approximately 80%
Kingdee
-106.7 billion Hong Kong dollars in market value shrinkage, a pullback of over 80%
The market cap curves of the two companies are almost identical—sliding all the way down from the 2021 peak, with the AI concept briefly giving a boost, but the capital market soon stopped buying the AI story.
This is quite worth pondering. In 2025, the AI-enhanced ERP market in China reached a scale of $315.7 million, a year-on-year increase of 96.1%. The manufacturing ERP market surpassed 26 billion yuan, with a quarter-on-quarter growth rate of 38%. The market is growing, but vendors' market values are falling—where does the problem lie?
The problem lies in the "triple jet lag."
Triple Time Lag: The Structural Dilemma of AI ERP
This is not a problem of poor management by Kingdee or Yonyou. On a global scale, SAP has invested 224 AI agents, yet the financial data has not fully reflected this yet. No manufacturer worldwide has completed an independent, closed-loop financial verification of an AI business model.
Key Points
- Signing→Delivery Time Gap: AI Agent deployment involves data cleaning, governance restructuring, model debugging, permission design, and multi-round validation. Complete delivery takes 6-12 months, and over 18 months for complex scenarios. This is not something you can "buy and use" like SaaS.
- Delivery→Confirmation Time Lag: The value of AI is difficult to accurately attribute. How much cost has been reduced, how much efficiency has been improved, how much risk has been controlled—these outcomes cannot be calculated independently. Many projects stall at the pilot stage, not because the technology is inadequate, but because it is impossible to clearly articulate how much it is actually worth.
- Confirmation→Profit Time Lag: Early-stage R&D and labor costs are high, making it difficult to amortize in the short term, continuously suppressing profits. Manufacturers are investing, but the return cycle is so long that the capital market has lost patience.
The triple time lag叠加 together results in: manufacturers are investing desperately, customers are trying slowly, and capital is losing confidence. This triangular dilemma is not unique to any single company but is industry-wide.
So you see Kingdee and Yonyou almost simultaneously showing their cards. This is not a coincidence, but a kind of anxiety: if they don't quickly secure a position on the AI track, they won't even have a chance to enter the game. However, between positioning and monetization lies this triple time lag.
The positioning of ERP is changing
Beyond these three time differences, an even deeper change is taking place: the positioning of ERP is shifting from a "recording system" to an "active management system."
What did the previous ERP do? After you completed a transaction, it helped you record it. Procurement inbound, sales outbound, financial accounting — post-event records, as long as you could find them when needed.
Now? Trend prediction, anomaly detection, resource scheduling, intelligent recommendation, and automated execution—ERP must evolve from "you do, I record" to "if you haven’t done it, I remind you to do it, or even do it for you."
| Dimension | Traditional ERP | AI-native ERP |
|---|---|---|
| Core Positioning | Record system | Active Management System |
| Data Role | Storage and Query | Input and Decision |
| Human role | Entry + Approval | Validation + Exception Handling |
| Value Embodiment | Process compliance | Efficiency improvement + risk reduction |
| Typical actions | Sales order placement → Inventory deduction → Financial accounting | Demand forecasting → automatic replenishment → anomaly warning → automatic allocation |
This positioning shift is not simply a matter of adding an AI dialog box. It requires that the underlying data architecture can support real-time analysis, the permission model can support AI proxy operations, and the audit system can track every automated decision made by the AI. In plain terms, this is not about putting an AI coat on an old system—it's about moving the foundation.
This also explains why the delivery cycle is so long. The 6-12 months is not about dragging things out; there really is a lot of foundational work to be done: data governance, permission restructuring, model debugging, and multiple rounds of validation. This kind of work cannot be rushed.
Tencent's approach to entry is different
In early June, Tencent released the WorkBuddy Enterprise Edition and Agent Suite at the Cloud AI Industry Application Conference. This is worth discussing separately because Tencent's approach differs from that of Kingdee and Yonyou.
Kingdee and Yonyou are adding AI to their own ERP systems—essentially still competing within the ERP track. Tencent is not. The positioning of WorkBuddy Enterprise Edition is an "AI-native enterprise collaboration platform," operating at the collaboration layer, not the ERP layer.
This is quite clever.
The reasoning is this: the switching cost of enterprise software is extremely high. If a company has been using Yonyou for ten years, can you make it switch to Kingdee? Not to mention data migration, just the user retraining alone can kill the project. But the collaboration layer is different—tools like WeCom, Feishu, and DingTalk have relatively much lower switching costs, and they themselves are not "systems of record" but "systems of collaboration."
Tencent's approach is clear: instead of competing with ERP vendors for that tough market, it focuses on building "AI-native" infrastructure at the collaboration layer. Leveraging the synergistic advantages of the WeChat ecosystem, Tencent Cloud, and WeCom, it prioritizes engineering and product capabilities over pure model capabilities.
The subtext of this line of thinking is: model capabilities will converge (today Fable 5 leads, but competitors will catch up in half a year), while product engineering and ecosystem barriers are long-term value. Tencent's accumulation in social networking and enterprise collaboration is more valuable than a few extra percentage points on benchmark test scores.
Of course, this path also has uncertainties. For enterprises, the difference between an enterprise collaboration platform building an AI Agent and an ERP system building an AI Agent lies in: the AI at the collaboration layer can handle information flow and decision-making suggestions, but to actually "take action"—automatically place purchase orders, automatically adjust inventory, automatically handle accounting—it still needs to go back to the ERP system. Whether Tencent's Agent suite can truly "move" the data within the ERP depends on the depth of its integration with various ERP systems. As for this matter, there is currently no clear answer.
SAP's 224 Intelligent Agents: Global Vendors Running in Sync
Zoom out a bit. On June 3, SAP announced at its China Summit that it has deployed 224 AI agents and 51 business assistants in core areas such as finance, supply chain, procurement, and human resources. KPMG has also signed a strategic cooperation agreement with SAP—ERP public cloud transformation + AI scenario implementation + compliance system.
This shows one thing: global manufacturers are investing synchronously in the AI race. It’s not just Chinese manufacturers feeling anxious; SAP is also pouring in resources. SAP’s advantage lies in its customer scale and global coverage—224 intelligent agents are not just a PowerPoint presentation; they are already operational. But it faces the same problem as Kingdee and Yonyou: investments are increasing, but returns have yet to be reflected in financial data.
The global ERP industry is undergoing the same transformation. The difference is only who runs out first, not who can avoid running.
Three Suggestions for Business Managers
After all this talk about manufacturers, what should business managers do? My advice is very specific—no more of that correct but empty talk about "embracing AI."
First, don't rush to switch systems. Among the enterprises I've worked with, quite a few are considering, "Since AI is so popular, shouldn't we switch to an AI-native ERP?" — my advice is, don't make a move for now. The reason is simple: the triple time lag means that no vendor's AI ERP has yet achieved a closed-loop financial validation. What you'd be switching to now is most likely a half-baked product. Moreover, the cost and risk of system migration itself are extremely high. An ERP migration project for a mid-to-large enterprise typically takes 12-18 months, during which business continuity can genuinely be compromised. Instead of switching systems, it's better to first conduct data governance and pilot AI scenarios on top of your existing system.
Second, organize your data before deploying AI. This may sound cliché, but it is an ironclad rule. The core capability of an AI Agent is "read data → make decisions → execute actions." If your master data is inaccurate, customer information is misaligned, or financial accounts are inconsistent—AI will read garbage data and produce garbage decisions. This has nothing to do with which model you use. So before signing any AI ERP contract, spend three months straightening out your data management. Only after completing a data accessibility audit should you select a solution; otherwise, no model change will help.
Third, when selecting a platform, focus on the depth of business process integration, not the model parameters. Many vendors are now saying, "I've integrated GPT-5" or "I'm using Claude Fable 5" — to be honest, model capabilities will converge within half a year. What you should really pay attention to is: how deeply is this platform integrated with your existing business processes? Can it directly read your ERP data? Can it automatically execute within your approval workflows? Can it link data across different systems? These are the key factors that determine whether AI is "usable" or "just for show" in your enterprise. The reason Tencent WorkBuddy's approach is worth noting is that it integrates at the collaboration layer — but how deep the integration actually goes depends on the product's implementation.
Core judgment
- The predicament of Kingdee and Yonyou is not unique to them—it is a structural issue across the entire ERP industry. The triple time lag will not disappear just because one company’s product is well-made.
- Tencent's entry posture is smarter than directly developing ERP, but the "last mile" from the collaboration layer to the ERP layer—automating core business data operations—still lacks a clear answer.
- For business managers, the most important thing right now is not to chase the AI story of any vendor, but to first sort out their own data and processes. If the foundation is not well laid, no matter how impressive the AI built on top is, it will be a castle in the air.
Thirty years ago, Kingdee and Yonyou started with financial software and fought their way to ERP. Thirty years later, they have almost simultaneously shown their cards on the AI track, but the capital market is no longer buying it. The key variable this time is not who releases the product first—it is who can first run through the chain of "signing → delivery → confirmation → profitability." Until then, everyone is waiting for the wind to come.
