May 13, 2026, Orlando, Florida, USA. At Sapphire 2026, SAP announced that enterprise AI has moved from the “assistive” stage to the “autonomous execution” stage.
The weight of this announcement has only gradually become apparent today (May 22)—because immediately afterward, Google I/O 2026 took the stage in the early hours of May 20, upgrading Gemini from a "chat assistant" to an "around-the-clock action agent." One week apart, the two companies independently shifted AI's role in the enterprise from a "tool that helps you do things" to an "agent that directly completes tasks for you."
This is not an isolated action by two companies, but a signal of the collective transformation of the enterprise software industry in 2026.
What exactly does SAP's “autonomous enterprise” mean?
SAP Global CEO Christian Klein said something at the meeting that is worth recording in full:
For customers' critical business processes, 'almost right' is far from enough. By combining the SAP Business AI platform with the SAP Autonomous Management Suite, we deeply embed AI agents into business processes, data, and governance systems to ensure they deliver accurate, compliant, and secure results.
The key phrases in this sentence are “deep embedding” and “direct execution.” What SAP released this time is not a specific AI feature, but a complete “autonomous enterprise” architecture, consisting of three pillars:
| pillar | Content | Enterprise Value |
|---|---|---|
| Unified AI Platform | SAP Business AI Platform | Build and manage AI agents to provide business scenarios |
| Self-Management Suite | Execute end-to-end core business processes | AI no longer just suggests; it directly executes |
| New user experience | Joule Work (New Interaction Method) | The user describes the goal, and Joule automatically executes the entire process. |
50 smart assistants + 200 professional agents—what does this mean?
SAP has released more than 50 Joule intelligent assistants focused on specific business areas, covering finance, supply chain, procurement, human resources, and customer experience. Under each assistant, there are more than 200 specialized AI agents responsible for specific tasks.
50 + 200—this number sounds like marketing. But there is a concrete case that can verify whether it is real: Autonomous Order Closing Assistant.
The role of this assistant is to autonomously handle journal entries and reconciliations, and to detect and automatically correct errors throughout the entire financial close process. The result is that the financial close cycle is shortened from weeks to days.
This case has been validated by a customer: RWE (a European energy giant) is already using SAP's autonomous asset management agent to analyze historical incident data and automatically generate work orders containing tool recommendations and maintenance plans.
This is not a proof of concept in a demo environment. RWE is a real customer, offshore wind turbines are a real scenario, and reducing unplanned downtime is real business value.
Noteworthy: SAP also announced that the workload for customers migrating to SAP ERP Cloud can be reduced by more than 35%. This is a direct result of the Agent-Driven Transformation toolkit—AI automatically completes system analysis, code remediation, configuration, and testing.
Google I/O 2026: Gemini Transforms from Assistant into an "Operating System Layer"
In the early hours of May 20, Google I/O 2026 conference. This timing falls exactly one week after SAP Sapphire, and this is no coincidence.
The core of this Google release is: Gemini is no longer an assistant that you open a dialog box to ask questions. It has become an AI layer at the Android system level (Gemini Intelligence), and has launched Gemini Spark—a persistent action agent that can run in the background around the clock and continue working even when the computer is turned off.
Let's take a closer look at a few important releases:
| Product | Core Competencies | Corporate Meaning |
|---|---|---|
| Gemini 3.5 Flash | 4 times faster, with costs continuing to decline | Economic Feasibility of Large Enterprise Deployment |
| Gemini Spark | Runs continuously in the background, and can work even if the computer is shut down | AI truly becomes a “digital employee,” rather than a “tool” |
| Antigravity 2.0 | AI programming platform ($1,000 to build an operating system) | Reduce enterprise custom development costs |
| AI Search Agentification | Search has changed from "returning links" to "directly executing tasks" | The Paradigm Shift in Enterprise Knowledge Management |
| US$180 billion to US$190 billion in infrastructure investment | Google's largest-ever AI infrastructure investment | Long-term service capability guarantee |
The Gemini Spark product is particularly worth noting. Its core design is: you give Gemini a goal, and it can continue operating in the background without you needing to stay online. This means the usage model of AI shifts from "conversational interaction" to "task delegation".
This is essentially the same direction as SAP's Joule Work: users describe the desired outcome, and AI autonomously orchestrates and executes the process.
Microsoft, SAP, Google: The Three-Way Battle for Enterprise AI Control
Looking at the events of the past month, the strategies of the three parties have become very clear:
Three-Way Rivalry Among Enterprise AI Platforms (Situation as of May 2026)
| Manufacturer | Core Products | Governance Strategy | Differentiation |
|---|---|---|---|
| Microsoft | Agent 365 (Observe/Govern/Secure) | Independent control plane, non-human entity identity management | Deep integration with M365, $15 per user per month |
| SAP | Business AI Platform + Joule Work | Data governance system embedded in business processes | Deep expertise in core ERP processes and industry solutions (7 major industries) |
| Gemini Intelligence + Spark + AI Control Center | Embedded into everyday Workspace tools (Gmail/Drive/Meet) | System-level integration, Android ecosystem |
The approaches of these three companies are completely different: Microsoft focuses on the "control plane," emphasizing governance; SAP focuses on "business process embedding," emphasizing accuracy and compliance; Google focuses on "system-level integration," emphasizing a seamless experience.
For enterprise customers, this means the choice is not just about "which model to use," but rather "what is your AI governance philosophy": Do you want strict control (Microsoft), deep business integration (SAP), or seamless daily integration (Google)?
Can domestic manufacturers keep up?
Shift the focus back to China. Kingdee held an AI summit on May 20, while Yonyou released its BIP enterprise AI product matrix in March; both are advancing their respective AI-native strategies.
But there is an insurmountable gap: behind SAP's more than 200 professional agents lies 40 years of accumulated accounting standards, tax rules, and industry best practices from 142 countries. This data moat cannot be caught up with merely by training a large model.
The breakthrough point for domestic manufacturers lies in:
- Industry depth:Kingdee has localized professional capabilities in manufacturing, while Yonyou has this advantage in the financial field—these are precisely the areas where SAP lacks accumulated experience. If an AI-native architecture can solidify this localized knowledge into intelligent agents, it will create differentiated value.
- Price: SAP Joule agents are billed based on usage, which poses a high barrier for small and medium-sized enterprises. If local vendors can reduce the price to one-third of SAP's, they can capture a large portion of the mid-range market.
- Compliance: Once the national AI agent evaluation standard (submitted for review on May 21) is released, local vendors will complete compliance alignment faster than SAP.
This is not a narrative of “domestic substitution,” but a real evolution of the market landscape. SAP’s advantage in the high-end market (large groups, multinational corporations) will not be shaken in the short term; however, competition in the mid-market will become very intense in the second half of 2026.
Practical Advice for Enterprise CIOs
First, distinguish between "AI demos" and "AI implementation."SAP’s checkout assistant has RWE validation; Google’s Gemini Spark has just been released, and no customer cases have been seen yet; while Kingdee’s AI-native product has just held a launch event. The three differ in maturity. The specific question to ask during procurement is: "Are there customers in the same industry who are already running this intelligent agent in a production environment?"
Second, governance is more important than functionality.IDC predicts that by 2028, enterprises will have 1.3 billion intelligent agents. By then, enterprises without governance systems will have 1.3 billion uncontrolled intelligent agents. When choosing an AI platform, governance tools (Microsoft's Agent 365, SAP's Business AI Platform governance framework, Google's AI Control Center) must become one of the decision factors.
Third, do not wait for the "perfect timing". The benchmark for 2026 is "whether it can replace a functional department of an enterprise" (as Kai-Fu Lee said at the AMD Developer Conference on May 19). If your AI deployment is still stuck at "improving efficiency" and has not touched business process restructuring, you are already falling behind.
What to watch for next
Over the next three months, there are a few things worth keeping an eye on:
- Timeline for the official release of the national AI agent evaluation standards (expected June-July) — this will directly affect domestic enterprises' AI procurement standards
- Among SAP RISE/GROW customers, the actual activation rate of the Joule intelligent assistant (currently free for the first year)—regardless of whether the data is true or false—depends on the activation rate.
- Customer feedback on Kingdee's AI-native product — at the summit, it was mentioned that the "on-site experience" has opened, meaning the product is already ready for delivery, and the first batch of customer cases is expected to be released around August.
- Microsoft Agent 365 actual adoption rate — officially launched on May 1; the first batch of usage data should be announced before August
In 2026, competition in enterprise AI has shifted from "whose model is stronger" to "whose intelligent agents are actually running business operations." This is a shift from a technology race to a value race. Those who run fast will eat those who run slow.
