Summary of Key Points
This article discusses the development of AI Agents, which are becoming increasingly capable—capable of writing code, creating videos, automating processes, and even allowing non-technical personnel to develop software on their own. However, companies are still hesitant to use them in core business scenarios. The reason is not that the models are not intelligent enough, but rather due to practical issues such as cost, stability, security compliance, and organizational division of labor. The article also analyzes the trends in Agent development, their impact on the SaaS industry, entrepreneurial opportunities, and the key risks associated with corporate adoption. It concludes that the critical challenge in implementing Agents is addressing the question of "how to trust them to perform tasks reliably."
1. Agents Becoming “Invisible Assistants” – Ordinary People Can Become “Mini-Programmers”
In the past, when using AI tools, you had to monitor every step: what the tool was thinking, which commands it was executing, and what its next action would be; you even had to correct any deviations. Now, things have changed—you simply provide an objective for the Agent (for example, “create a product launch video”), and it handles everything in the background, delivering the result directly. This shift towards invisibility indicates that Agents can handle more complex tasks, utilize a wider range of tools, and operate with greater autonomy.
Even more noteworthy is that non-technical users can now use Agents to develop software. For instance, AI-based code generation tools (like Vibe Coding) can retrieve transcribed content, process it, and connect it with other tools—something that previously required engineers or specialized integration tools, but now can be done in just minutes. However, this also brings risks: since you cannot see the entire process, you may assume the result is correct without verifying it, which could lead to errors. The more an Agent behaves like an independent employee, the more companies need to carefully consider which tasks it should handle and which processes they must oversee.
2. Will SaaS Be Displaced by AI? Only “Lightweight Tools” May Face Competition
Some claim that AI will eliminate SaaS (enterprise software), but this is probably an exaggeration. Which types of SaaS will be affected? Those with single functions, low data dependency, and easy migration—such as simple signing tools, form generators, and internal process management tools. These tools are at risk because business users can create similar functionality using natural language in just a few minutes and customize them to fit their workflows, rendering them obsolete.
However, core systems like Salesforce (customer management), Figma (design software), and ERP (enterprise resource planning) will not disappear. Why? Large companies purchase these systems not only for their functionality but also for the trust they offer from the suppliers, as well as for support services and compliance audits. Even if AI-generated tools meet the same functional requirements, companies are reluctant to replace them with core systems. For example, switching to a new CRM would require the sales team to adapt again, and there are significant risks associated with data migration. Moreover, SaaS providers will likely integrate AI features soon, making it unnecessary for companies to take such risks.
Additionally, developing custom AI tools may not be cheaper in the long run: small companies must pay for model licenses, cloud services, and maintenance, which could be more expensive than the monthly subscription fee for SaaS. Large companies that develop customized tools also face additional costs related to deployment, integration, and security, as well as the need for professional support.
3. For AI Agent Startups, Focus on Vertical Markets
Many startups fear that large model companies (like OpenAI) will compete with them, but there’s no need to worry. Large models are better suited for developing general-purpose capabilities (such as chat and coding) rather than specific industry workflows. Take private wealth management, for example: how to manage customer asset data, what information needs to be anonymized, and how regional compliance requirements affect processes? These industry-specific complexities cannot be learned from public data.
Therefore, the best entrepreneurial opportunities lie in vertical markets—focusing on a particular industry and integrating Agents into specific business processes to solve real problems. For instance, a medical Agent must understand doctors’ workflows and hospitals’ compliance requirements; a financial Agent needs to know about customer data management regulations. If a startup’s team does not have experience in that industry for more than 10 years, it is unlikely to succeed. Companies are willing to pay for such Agents not because of their intelligence but because they understand the business context. The competition in the general-purpose Agent market is fierce, and startups in vertical markets have a better chance of standing out due to their expertise.
4. Why Aren’t Companies Using Agents? Four Major Concerns Need to Be Addressed
Companies are hesitant to use Agents in core scenarios due to four main concerns:
1. Stability Issues: Industries like finance and healthcare require 99.999% availability, but current Agents typically have around 98% reliability and rely on external APIs, which can cause disruptions (for example, if two models used by a bank fail simultaneously, it can lead to operational chaos).
2. AI’s Bias: Models may cater to users’ preferences, potentially providing biased or incorrect recommendations that influence decision-making.
3. Blurred Responsibility: As Agents handle more tasks (organizing emails, responding to customers, editing documents), the responsibility for errors becomes unclear.
4. Cost Efficiency: Developing custom tools can be more expensive than using SaaS; even for large companies, the costs associated with customization, deployment, and maintenance can be substantial.
These issues must be resolved before companies can confidently adopt Agents in their core operations.
Conclusion
While AI Agents are becoming more capable, their successful integration into core business scenarios depends on more than just the intelligence of the models. It requires balancing efficiency and risk, considering costs and compliance requirements, and redefining the division of labor between humans and AI. The true value of AI Agents lies in how they can be trusted to perform tasks reliably. This is the real starting point for their adoption on a large scale in industrial applications.