How Strong is the Wind of AI Agent?

By the end of 2024, Microsoft CEO Satya Nadella boldly predicted, "Software-as-a-Service (SaaS) applications will face a paradigm shift in the era of AI Agents."

Notably, Microsoft stands as a global SaaS giant, making Nadella's words a powerful testament to the potential of AI Agents, perhaps unparalleled in significance.

Yet, at this juncture, can we truly witness the AI Agent wave crashing ashore and "annihilating" SaaS?

The answer remains far from crystal clear.

I. The Cracks Beneath the $300 Billion Market

Much like today's AI Agents, SaaS was once the "darling" of venture capitalists.

SaaS, a cloud computing model, enables users to subscribe to software services via the web, eliminating the need for local software installation. Updates and maintenance are handled by the vendor.

Stars like Salesforce and Shopify have thrived on SaaS, while giants like Microsoft and Adobe have integrated it into their core strategies. According to Statista, the SaaS market has burgeoned from $31.4 billion in 2015 to an estimated $300 billion by 2025, displaying rapid growth.

Salesforce, a bellwether in this industry, peaked at a market value exceeding $340 billion.

However, beneath this prosperity lies unresolved pain points that have become shackles for SaaS.

SaaS has evolved into a fiercely competitive red ocean industry, with over 30,000 global vendors (2023 data) and 4,500 Chinese vendors (2021 data).

To stand out, vendors must heavily invest in promotion and sales. The "Research Report on the Development of China's Enterprise-level SaaS Industry (2024)" by the China Academy of Information and Communications Technology reveals that Chinese SaaS vendors' sales and R&D expenses exceed half of their revenue.

In recent years, domestic SaaS vendors like Kingdee, Beisen, and Jushuitan have seen sales expense ratios (the proportion of sales and marketing expenses to revenue) exceed 40%, with Weimob's ratio even surpassing 70%.

High sales and marketing costs devour gross profits, often leading to "loss-leader" tactics to attract customers.

Additionally, traditional SaaS has matured, with vendors offering similar products in terms of underlying technology and common functions. The lack of innovative breakthroughs means users can easily switch between vendors, resulting in low retention rates.

Pricing sensitivity also discourages vendors from customizing solutions, hindering deep alignment with user needs and weakening user stickiness.

This vicious cycle of failing to attract new users and retain old ones persists.

Furthermore, traditional SaaS products suffer from "inherent deficiencies," notably functional singularity and lack of data interoperability.

Enterprises may need to subscribe to multiple SaaS products to meet various needs, but data silos prevent seamless integration, leading to fragmented user experiences and inefficiencies.

Data security concerns also persist, as users entrust their data management to SaaS vendors via the cloud.

These pain points, despite SaaS's commercial success, present opportunities for new technologies like AI Agents to showcase their strengths and reshape the landscape.

II. AI Agent: The Transformer Arrives?

An AI Agent can be simply defined as an artificial intelligence system that autonomously perceives its environment, makes decisions, and takes actions to achieve specific goals.

While the concept is not new, the advent of large language models (LLMs) like ChatGPT has reignited discussions and research on AI Agents.

AutoGPT, based on GPT-4, launched in March 2023, showcasing a tangible path for AI Agents to land.

Equipped with multimodal capabilities and chains of thought (CoT) for problem-solving, AI Agents not only possess a powerful brain but also "perceptions" like vision and hearing, bringing AI "thinking" closer to human cognition.

AI-driven task completion in specific scenarios is now tangible.

In the context of SaaS, can AI Agents, with automation and intelligence as their key selling points, hope to transform existing pain points?

At least in some aspects, AI Agents appear promising.

Theoretically, AI Agents can significantly enhance customization. Unlike SaaS products that require users to adapt, AI Agents can automatically create workflows based on user needs, lowering the entry barrier.

While traditional SaaS products offer graphical interfaces with buttons and menus, AI Agents can simplify the front-end to a conversational interface, offering ample room for personalized visual design.

Even with developer permissions, AI Agents can "guess user preferences" and reconstruct UIs and functions accordingly.

When integrating multiple SaaS products to solve complex, multi-scenario problems, the frustration of constant switching can be alleviated with AI Agents. Users only need to state desired outcomes, with AI handling most intermediate processes.

This can be achieved through natural language instructions, akin to interacting with Siri or Xiaoai.

AI Agents' ability to "think autonomously" and analyze cross-module data potentially turns them from passive tools to proactive assistants, offering business suggestions without explicit instructions.

For instance, in traditional ERP systems, users manually enter purchase plans, inventory data, production schedules, and financial accounting. AI Agents can streamline this by understanding a simple instruction like, "Increase Product A production by 20%," and automatically adjusting plans, schedules, funding requirements, and generating financial impact reports, even warning about supply chain risks and proposing solutions.

If user experience can reach such heights, user retention will no longer be a hurdle for SaaS vendors.

III. The Future of AI Agents and SaaS: Fusion or Disruption?

AI Agents' impact on SaaS is already evident. Companies like Salesforce, Microsoft, Yongyou, and Weimob are launching AI Agent solutions.

Currently, AI Agents are more integrated into SaaS applications, handling planning and decision-making at specific nodes. The future may unfold in two directions.

First, SaaS may continue to exist but deeply integrate with AI Agents, evolving through their influence. AI Agents will be embedded into business processes to solve specific needs, with SaaS functioning as a basic tool akin to an API.

Under this scenario, while SaaS may retreat from the forefront, it won't be completely supplanted. AI Agents and SaaS will coexist symbiotically rather than in a zero-sum replacement.

Salesforce CEO Mark Benioff supports this path, believing that AI Agents cannot function optimally in isolation and need a reliable data foundation and robust applications (i.e., SaaS) for support. He sees AI Agents driving SaaS towards a more advanced form – Service-as-Software (SaSo).

The second trend involves a more radical vision – since SaaS can be simplified as "database + business logic," acting as an intermediate layer between business and data, why not let AI Agents directly "talk" to the database and handle the logic?

Nadella's vision aligns closer with this latter trend, but has its dawn already arrived?

For now, AI Agents face significant hurdles in truly disrupting SaaS.

First, AI Agents are essentially "shell" applications of large models, constrained by the capabilities of their underlying LLMs. LLMs' reasoning ability, multimodal capability, and hallucination rate affect AI Agents' usability.

In the OSWorld benchmark test, the best AI Agent score is 42.5% (from ByteDance's UI-TARS-1.5), while the average human score exceeds 70%. While sufficient for personal tasks, enterprise business demands higher reliability and lower fault tolerance, posing stringent demands on AI Agents.

From a cost perspective, while large model vendors reduce pricing for reasoning APIs, the overall cost is still high for large-scale deployment. Complex tasks requiring multi-step reasoning incur significant invocation costs, potentially leading to rapid resource consumption.

Model fine-tuning, deployment, and iterative optimization in vertical industries also involve substantial technical and financial investments. Being "cool" isn't enough for enterprises to fully embrace AI Agents.

Moreover, AI Agents don't fundamentally resolve SaaS's data security and compliance issues, introducing new uncontrollable factors. Tracing and accountability become complex once AI Agents encounter issues.

In all aspects, while AI Agents show promise, we haven't yet witnessed a "killer application" with a real chance to replace SaaS at the enterprise level.

In summary, AI Agents may reshape the SaaS industry, but this transformation is still in its nascent stages. SaaS vendors must seriously consider how to transform or integrate now, as procrastination could be detrimental.