"The future belongs to Agents, and multi-Agent interaction will undoubtedly shape the landscape. However, before we reach that point, 'SaaS+AI' will serve as a bridge, as any TO B service necessitates industry knowledge, processes, permissions, and security—pillars that remain indispensable."

Author | DouDou

Editor | PiYe

Produced by | Industrialist

"'Is the SaaS model destined for obsolescence?' This question has haunted me since the start of the year and has become a hot topic within the industry." A founder of a domestic SaaS enterprise confided in Industrialist.

This anxiety is widespread. While discussions about the SaaS model's irrelevance in the AI era were once limited to forward-looking investors and Silicon Valley ventures, two recent events have intensified the debate.

In May of this year, the general-purpose Agent Manus was unveiled. Despite mixed reviews for its C-end performance, its impact on the B-end was profound. "As soon as Manus was released, we began exploring it, even redesigning parts of our software's delivery processes, capabilities, and result interfaces for inspiration," shared the founder of another CRM enterprise.

He's not alone. Within a SaaS enterprise focused on data insights, Manus is viewed as a pivotal AI TO B move. "This signifies that AI is now genuinely capable of problem-solving and delivering tangible results through strategic planning." This enterprise has undergone structural adjustments over the past year or two, and Manus's arrival provided a much-needed "vote of confidence" for these changes.

Apart from Manus, another catalyst for this discussion was a "consensus report" from Sequoia China's internal sharing session, which included participants from Microsoft and OpenAI. The session underscored "outcome delivery" as the new benchmark for product delivery standards in the AI era.

In contrast, domestic SaaS providers have struggled to achieve full-fledged outcome delivery, balancing standardization with "customized delivery" across sectors like CRM, ERP, and industrial software.

How should SaaS enterprises adapt? Is the SaaS delivery model still viable in the AI era? What is the optimal integration model for Agents and SaaS?

"SaaS is returning to its core concept: it's not just software but a mode of delivery," an investor emphasized.

Challenges and opportunities coexist as China's SaaS embarks on the AI journey.

I. Two Paths in the AI Wave for SaaS

China's SaaS enterprises have already taken action.

Broadly, there are two paths. The first involves building Agent PaaS based on PaaS logic, integrating AI into existing products while offering Agent capabilities as standalone features and deeply embedding them into the software interface. This approach ensures that SaaS providers maintain their business models and meet current AI demands.

Some SaaS vendors prioritize high-value, demand-driven scenarios such as marketing, sales, data, and customer service. This path offers quick results and targeted resource allocation, avoiding the pitfalls of comprehensive transformation.

From an AI perspective, this path aligns with the varying maturity and data quality across sales scenarios. Implementing AI in scenarios with abundant, high-quality data not only meets existing customer needs but also provides a technological foundation for penetrating larger clients. This strategy balances short-term commercial returns with long-term technical barriers.

The future prospects of this path are promising. Many SaaS enterprises have informed Industrialist that Agents will also require personalization, prompting the preparation of "Agent PaaS" with underlying capabilities evolving from general to industry-specific modules.

This approach is currently mainstream among domestic SaaS enterprises, including Salesforce, Zoho, NetEase Wisdom, among others.

The second path involves creating end-to-end Agent products that operate independently of existing software. These new products target new scenarios with distinct commercial pricing, exemplified by Beisen and Kingdee's related offerings.

Take the "AI Interviewer" and similar products launched by various vendors. Unlike integrating AI with existing products, the AI Interviewer functions as a standalone AI Agent, billed and operated separately.

This path necessitates organizational restructuring. A new delivery model, with different personnel, cycles, processes, and standards, is required for this AI commercial product.

However, similar to the first path, the core lies in scenarios. Beyond inherent SaaS directions, enterprises must validate the commercial value and market demand of these scenarios. Many enterprises judge such products based on "completing PMF validation within a set timeframe."

"This essentially places us on the same battlefield as AI startups, but we still hold a significant advantage," a founder revealed.

II. Three Major Challenges: Cost, Organization, and Business Model

Beyond the paths, a more fundamental question arises: Does the AI transformation of SaaS offer tangible value?

According to Gartner's "2024 AI Agent Market Traps Report," over 80% of Agent promotional videos depict idealized scenarios, significantly overshadowing actual usage effects.

This transformation is more challenging than anticipated. Data shows that over 60% of global SaaS enterprises are still unprofitable, with tight cash flows.

In this context, even with low token prices, AI transformations remain costly due to intelligent agent reasoning training. For instance, building a talent profiling system for an HR SaaS enterprise cost 2.7 million yuan in data cleaning alone.

The issue extends beyond cost to establishing a new service model. The two transformation paths—integrating AI into existing products and creating new end-to-end Agent products—both face business model or development bottlenecks.

In the former case, AI is treated as a value-added service, posing two challenges: quantifying AI's impact within the enterprise and pricing. Many vendors have noted that customers are reluctant to pay separately for value-added services, and even when they do, the fees are minimal.

This has prompted many SaaS vendors to transform "value-added services" into standalone Agent products this year. While still embedded in original products and processes, the pricing mechanism is clearer and more resolute.

Delivery presents another hurdle. "Currently, we handle deliveries ourselves. There are almost no external channel partners for AI Agent-based deliveries, and we're still figuring it out," a vendor leader admitted.

The latter path's challenges are evident. Starting from scratch with end-to-end products requires validating new business scenario needs, necessitating the PMF validation process again. Considering the current AI product startup landscape, with AI technology iterating, the difficulty of starting a business from the AI application layer is greater than before. It's crucial to avoid both large factory scales and AI technology development ranges.

Behind these challenges lies the certainty of SaaS vendors seeking change amidst directional confusion.

With the emergence of general Agent products like Manus, outcome delivery and agent forms are increasingly becoming established TO B service models. SaaS enterprises must disrupt their inherent product and delivery models. However, large models' capability boundaries haven't been fully explored, and the AI application form of Agents is still maturing.

Moreover, the Agent-based TO B service model differs from the traditional SaaS model, necessitating changes in service processes and delivery logic. SaaS providers must undertake painful adjustments to build a new delivery model. For instance, a new CRM enterprise's service approach currently aligns project indicators, ultimately deriving the product service form from these indicators.

III. Navigating the Path for SaaS+AI

"Actually, Salesforce doesn't know the best way either. They're taking it one step at a time," a corporate leader shared.

Salesforce, currently the world's most valuable SaaS enterprise, has seen its stock price soar with advancements like Agentforce, reflecting market optimism.

Salesforce's approach to large AI models involves strengthening its CRM product model by embedding AI capabilities into original modules for product evolution and launching the end-to-end Agentforce product, helping enterprises build various Agent products leveraging Agentforce's robust PaaS capabilities.

Additionally, to ensure delivery, Salesforce has recruited new AI sales staff and established a dedicated AI delivery system.

The fact is that SaaS's AI transformation, or AI's native transformation, is a gradual process. Any attempt at a "Great Leap Forward" is doomed to fail. Like autonomous driving's early stages, after market education, most manufacturers shifted from the "one-step" L4 approach to the "gradual" L2 path, gradually achieving commercialization through assisted driving. No surviving autonomous driving manufacturer failed to make this timely pivot.

Similarly, SaaS vendors need "gradual" AI transformations, passing through a "semi-automatic" transitional stage.

For example, while Salesforce's Agentforce emphasizes autonomous task execution, it still relies on the original CRM data model and API interfaces, treating AI agents as "intelligent plugins" for the existing system. Domestic vendors like DingTalk have launched AgentStack, allowing enterprises to combine AI capabilities with existing functional modules through a low-code platform to create "toolbox-style" solutions.

This intermediate state's rationale lies in reducing the risk of technical reconstruction while enhancing user stickiness through incremental AI functions.

Moreover, international vendors' actions offer new insights. For instance, Microsoft's "Agent Stack" vision aims to build an intelligent hub across enterprise applications, dispatching multiple Agents through a unified framework to complete complex tasks, even planning to replace traditional databases as the enterprise's core operating system. Large model vendors like OpenAI are aggregating developers through the API ecosystem, promoting AI Agents as standardized service interfaces. With the introduction of the MCP/A2A protocol, this interaction mode is becoming increasingly feasible.

Regardless of the approach, the market consensus is clear: SaaS's inherent form is evolving.

"The future still belongs to Agents, and multi-Agent interaction will undoubtedly shape the future. Just imagine an enterprise procurement process facilitated by multiple Agents—a demand analysis Agent interfacing with the financial system, a supplier price comparison Agent integrating e-commerce platforms, and a contract generation Agent linking with legal modules, ultimately achieving a fully automated process," the enterprise founder from the beginning of this article envisioned.

"But before that, 'SaaS+AI' will serve as the bridge, as any TO B service necessitates industry knowledge, processes, permissions, and security—pillars that remain indispensable."