Over the past two decades, the enterprise software industry has focused on one thing: finding ways to keep users within their own systems.

Whether it's Salesforce's CRM, Oracle's ERP, or domestic players like Yonyou and Kingdee, the business logic is largely the same: migrate enterprise business processes into software, get employees accustomed to using their products, and then charge through accounts, feature modules, and projects.

But in 2026, the most critical survival model for the SaaS industry is crumbling.

At September's Dreamforce conference in San Francisco, Salesforce officially launched AIforce, opening up customer data, business processes, and permission capabilities to external agents via MCP, API, and other methods. Enterprise employees can now complete business operations through AI tools without even opening Salesforce's webpage.

A month later, another software giant, Workday, announced plans to advance its "headless" agent platform and partnered with Anthropic, enabling employees to directly query HR and financial data—or even initiate approval processes—within Claude.

Now, a question is spreading across the enterprise services industry: If AI becomes the unified gateway for all enterprise software, is there still a need for SaaS?

Interfaces are disappearing, but SaaS's moat remains

To understand this shift, we must first answer a fundamental question: Why are enterprises willing to keep paying for SaaS?

For many, the answers might be functionality, data, and user habits.

A company uses Salesforce for customer management, Workday for employee management, and Oracle for financial management. Over time, these tools become integral to daily operations: employees learn to operate them, managers adapt their processes, and IT departments maintain the systems.

The longer they're used, the higher the replacement cost.

A representative story circulates in China's enterprise services sector.

A major bank had used Oracle ERP for years, with employees constantly complaining about its complexity. Eventually, the bank decided to develop its own ERP system. But when the new project was initiated, the first requirement from the business department was: "Make the new interface as similar to Oracle as possible."

The reason was simple.

While everyone found the old system cumbersome, they had grown accustomed to its operation.

But when employees no longer need to operate the system themselves—letting AI handle tasks instead—the friendliness of the interface or the complexity of menus becomes irrelevant. Customers now care about only three things: Can it connect to the system? Can it retrieve the needed data? Can it complete the specified operations?

Does this mean the SaaS era is over? The answer is no.

Consider a simple example.

An employee can ask AI to query this month's procurement budget or generate a purchasing recommendation based on historical data. But if they ask AI to procure equipment worth 1 million yuan, the situation changes entirely: Who has the authority to initiate the purchase? Who approves it? Is the budget sufficient? Does the supplier meet requirements? What payment processes are involved?

Enterprises must have clear identity authentication, permission management, business rules, and audit mechanisms—exactly the core capabilities that Salesforce, Workday, and others have long built.

For instance, in Workday's collaboration with Anthropic, it was explicitly stated that requests initiated through Claude must still adhere to Workday's original permission and business rules, with any modifications requiring user confirmation.

This explains why Salesforce dares to open up its capabilities: What it's surrendering is the operational gateway, but what it retains is control over data, rules, and business execution. In fact, AI is helping SaaS establish a new form of customer binding.

In the past, enterprise software relied on user habits for stickiness; in the future, it may rely on the call (translation note: " call " means "invocation" or "integration" in this context, referring to how agents interact with business systems) relationship between agents and business systems. Once enterprises entrust core processes like sales, finance, and HR to agents, the stability of underlying software, data quality, and rule systems will become increasingly critical.

So, AI isn't simply eliminating SaaS—it's forcing SaaS companies to answer: When users no longer need your interface, what value can you still create for enterprises?

From selling services to selling results: SaaS business models are being reshaped

If the disappearance of interfaces transforms enterprise software products, another AI-driven change is directly impacting SaaS business models.

The most typical SaaS pricing model charges per account or per seat.

A company with 100 salespeople buys 100 CRM licenses; one with 500 employees purchases an HR system at corresponding scale.

Agents are disrupting this premise.

Suppose a company previously needed 100 customer service agents to handle inquiries but now, with AI agents, only 20 employees are needed for complex issues and manual reviews. Does this company still need to purchase 100 customer service software licenses?

For SaaS companies relying on seat-based pricing, this is clearly bad news.

Capital market sentiment reflects this concern: In mid-April, the U.S. software industry index IGV fell nearly 37% from its peak, as investors worried that AI would not only replace some enterprise employees but also reduce the number of software licenses companies purchase.

But subsequent developments reshaped market perceptions.

Salesforce's latest quarterly revenue reached $11.35 billion, up 11% year-over-year. Its Agentforce and Data 360 platform now generate nearly $3.9 billion in annualized recurring revenue. Meanwhile, enterprise giant ServiceNow's AI business annual contract value surpassed $1 billion, with its full-year target raised to $1.5 billion.

These numbers make one thing clear:

Enterprise software hasn't lost all commercial value due to AI adoption. Instead, some SaaS companies are already generating new revenue from AI applications.

More critically, the billing targets are shifting.

Enterprises used to pay for the right to use software; now they're increasingly willing to pay for the work the software actually completes.

Salesforce now measures AI workload using "agent work units," with 3.2 billion units delivered in the second quarter, up 97% sequentially.

In customer service, the shift is even more intuitive (translation note: " intuitive " means "visible" or "apparent," referring to how clearly the change can be seen).

Companies like Zendesk and Intercom are exploring pricing based on problem resolution counts rather than simply the number of customer service accounts. For example, an AI customer service system might charge $1–2 per successfully resolved issue.

In other words, enterprise software is gradually shifting from per-user pricing to pricing based on workload, tasks, or even outcomes.

Of course, this transformation is still in its early stages.

If an AI customer service agent answers a user's question but the user is unsatisfied with the result, does that count as a successful resolution? If AI completes a procurement but causes losses due to incorrect judgment, who bears responsibility?

But the industry direction is becoming clearer.

Future competition in enterprise software will no longer hinge solely on feature count or interface friendliness but on who can complete business tasks more efficiently and reliably—measured by AI's ability to truly understand enterprises, follow rules, and get work done.

The AI gap between China and the U.S. is shifting from model capabilities to ecosystems

Why have U.S. SaaS companies returned to growth trajectories while China's SaaS sector remains pessimistic? The issue isn't that SaaS lacks value but that China's underdeveloped SaaS ecosystem may struggle to support AI's ambition to dominate work interfaces.

At the product level, China's AI ecosystem hasn't fallen behind.

Tencent launched WorkBuddy, ByteDance offers TRAE, Alibaba has Qoder, and platforms like Feishu and DingTalk continue integrating AI capabilities. All are transforming AI from simple Q&A tools into work assistants capable of handling documents, invoking tools, and executing tasks.

But when we examine the tools behind these agents, clear differences emerge between China and the U.S.

Take overseas AI tools like Claude and Codex: users can already collaborate with third-party services like Notion, Figma, GitHub, and Jira through plugins, MCP, or other connections. AI is no longer just an isolated chat window but evolving into a work platform that connects different software.

In contrast, many domestic agent products remain confined to their own ecosystems.

While platforms like Feishu, DingTalk, and WorkBuddy are embracing open protocols like MCP, open interfaces don't equate to open ecosystems. Tencent's products more easily integrate with Tencent's services, Alibaba's products tightly couple with Alibaba Cloud and DingTalk, and ByteDance builds capabilities around Feishu and its own tools.

The question remains: How many mature, standardized, and authorized business capabilities can AI invoke across different vendors?

After two decades of development, the U.S. SaaS market has formed a mature ecosystem of specialized software. Enterprises can choose Salesforce for customer management, Workday for HR, and ServiceNow for IT services, then connect these systems through various interfaces.

These tools weren't designed for the AI era, but their long-term standardization, cloudification, and API development have given them a foundation for agent invocation. AI adds a new interaction layer atop this existing software ecosystem.

China's enterprise software market followed a different path.

Many large enterprises prefer custom software tailored to their organizational structures and management processes: the enterprise specifies requirements, the software vendor handles development; when business changes, the system is adjusted through secondary development. This meets customized needs but results in software capabilities trapped in individual projects rather than standardized products that can be resold or reused across clients.

It's not that Chinese SaaS companies haven't recognized these changes.

Kingdee launched its Lingji agent platform to convert enterprise business data, management knowledge, and rules into AI-understandable contexts. Yonyou is advancing AI integration with enterprise management software... The challenge isn't launching an agent product but standardizing underlying business capabilities.

Relatively optimistically, while traditional software required new feature development, interface adaptation, and system integration for each business need, large models' natural language understanding, code generation, and tool invocation capabilities now offer the chance to transform custom development work into standardized tool combinations invoked by agents.

As agents become new work gateways, demand for professional services around data governance, identity authentication, business rules, cross-system connections, and execution audits will likely emerge, reshaping business models long reliant on project-based customization and human delivery.

Final Thoughts

A mature enterprise AI ecosystem requires not just intelligent models but also abundant software services that can be securely invoked to continuously create business value.

Models determine how smart AI can be; software ecosystems determine how much work AI can actually do.

As more enterprises let AI directly participate in business execution, the focus shifts beyond who creates the next chat interface to who becomes the indispensable provider of business capabilities behind that new gateway.

For China's SaaS industry, AI has exposed years of accumulated weaknesses but may also offer a reshuffling opportunity.