Will AI replace software?

This article was originally published on my blog and translated into English. For the latest version and future updates, please visit the original post: https://jaketao.com/language/en/will-ai-replace-software/

The concept that "AI will replace software" has recently gained significant traction. This has led to a significant decline in software stocks, and when combined with concerns about the profitability of AI, has caused widespread anxiety in the market. While there are legitimate concerns about the profitability of AI, it is important to note that the cost of developing AI remains high, and revenue models continue to rely on traditional methods such as advertising and recommendations. The assertion that AI will disrupt the software industry may be somewhat exaggerated.

This discussion takes place against the backdrop of a recent surge in groundbreaking AI applications. First, the sudden rise in popularity of Clawbot allowed users to build personalized AI systems on their local devices, enabling automated operations and process execution --- essentially functioning as a "personal assistant." Subsequently, Anthropic released Cowork, further enhancing agents' ability to perform complex tasks and demonstrating AI's potential for cross-process collaboration. These developments send a clear signal: AI is gradually transitioning from an auxiliary tool to a system application with executive capabilities. In the near future, it is entirely foreseeable that AI will integrate with personal devices and data resources, allowing users to simply issue commands for AI to invoke various processes to complete the required tasks. This evolution will see AI evolve from its current role as an auxiliary "secretary" into a "hands-on" and capable assistant.

A more critical issue lies in the fact that certain software may no longer be considered an essential necessity. In AI-driven workflows, systems will rely more heavily on callable, specialized functional modules. In simple scenarios, AI can dynamically generate or temporarily build the required capabilities. From this perspective, this will certainly have a substantial negative impact on the software industry.

What impact will AI have on software?

Before drawing conclusions, it is essential to examine the substantive impacts that AI's development will have on the software industry.

Software Development Becomes EasierIt is generally accepted that software development has become easier. Historically, the development of software has been a time-consuming process for teams. With the use of AI, a simple program can be completed by just a few senior engineers handling architecture design, code integration, and final debugging. For example, Clawbot was built by a single person, though this analogy isn't entirely accurate. However, it clearly demonstrates how AI can greatly enhance a programmer's development capabilities. While AI is not yet capable of writing highly complex programs, it is possible to break a program down into smaller components. These components can then be debugged and integrated by experienced engineers. This approach has been shown to significantly reduce the software development cycle and lower development costs.

What are the potential implications of this decision?

First, it should be noted that plug-in and utility-type software programs, which are typically billed monthly, may become obsolete in the future. This is due to the fact that creating a plug-in with AI is a straightforward process, and AI can also be customized to specific situations. As a result, these solutions often align more closely with a company's business logic than off-the-shelf plug-ins currently on the market.

Secondly, there will be an increase in the availability of affordable alternatives for large-scale software. While it may not be feasible to develop a comprehensive, large-scale software application from the beginning, the integration of AI can implement specific functionalities, thereby empowering users to achieve their objectives without the need to rely exclusively on a particular program. This may prompt some light users to explore alternative options.

Disruption of Pricing Models

At present, almost all SaaS pricing models rely on a combination of monthly and per-user fees. However, this model may not be sustainable. As previously indicated, users will become less dependent on basic utility software, and the number of light users will also decrease. In such cases, the monthly billing model becomes financially disadvantageous for users. Given the model's previous unpopularity, users now have the opportunity to avoid it. It is likely that they will disseminate information about this fact.

Furthermore, per-user billing will become increasingly challenging as AI enhances work efficiency, allowing tasks that previously required significant manpower to be handled by just a few, or even a single AI "user," with the rest accessing the entire software suite through that AI machine or agent.

In this scenario, software companies will need to adapt their business models and explore new approaches, such as charging based on features and usage frequency (similar to a token system).

Redesigning Software Architecture

Many large-scale software applications nowadays provide comprehensive features and sophisticated interactions. However, as AI technology continues to evolve, users may find that only a portion of these features is necessary. This is due to the fact that AI is capable of breaking down tasks and planning and executing workflows in the most efficient manner. For AI, large software applications consume resources differently than smaller functional modules; it will inevitably choose the workflow that consumes fewer resources. This will compel prominent software companies to modularize their software in order to retain substantial AI users.

This aligns with user logic as well. While many software applications are very comprehensive, most users only utilize a portion of their features. Existing pricing strategies are mostly designed around these popular features.

Shifting Paradigms in User Interface Design

Presently, software interfaces with users through its front end, whereas AI merely needs to call APIs --- adopting a wholly distinct approach. The design and development of these front ends can be a time-consuming and labor-intensive process, especially given the need for distinct designs to meet the diverse needs of different customer segments. This complexity makes software replication a challenging endeavor. However, this will inevitably lead to a decline in its value. In the context of AI, the primary requirement is for the system to comprehend the API in order to operate effectively. The software's front end will no longer be visible to users, and the complex buttons and menus will be removed.

App Store?

In the near future, AI may emerge as a pivotal interface for user interaction, similar to the transformative impact of the iPhone on the app era. It is not yet clear whether a model similar to the App Store will emerge. I believe there is a strong possibility that this will happen. If AI becomes a gateway, all software accessible through that AI will require platform authorization, and users will need to purchase the software or features they need through the platform to build a personalized AI workspace. Consequently, major AI platforms will possess substantial bargaining power, potentially imposing a "tax" on software listings (akin to the "Apple tax"), further reducing software companies' profits.

Advantages of existing software

For complex, dynamic, and large-scale software --- whether engineering or industry-specific --- it is almost impossible for AI to mimic or replicate it. AI is adept at understanding a user's clear objective, breaking it down, and executing each step in a sequential manner. However, when dealing with complex workflows and environments, users may face challenges in accurately articulating all the circumstances, thereby hindering the effectiveness of AI.

This is also the greatest advantage of most SaaS companies. Users could break down the entire process into individual tasks and have AI execute them step by step. While this is not difficult to achieve, it would increase the user's time cost. After carefully evaluating the advantages and disadvantages, users may still decide to utilize professional software. To illustrate with a practical example, consider the annual tax filing process. The forms are available to the public, and the instructions for filing are clearly outlined. However, individuals often prefer to allocate their financial resources towards acquiring software or engaging the services of an accountant to manage their finances, rather than undertaking these tasks themselves. In the context of AI, filing taxes is a straightforward, single task, and I believe AI will be capable of handling it in the near future. However, organizing all your tax forms and submitting them to AI is a complex process. In this regard, software companies have a significant advantage. If they decide to develop distinct AI functions or incorporate AI capabilities into their own software, it will significantly boost its competitiveness in the market. Would you prefer to invest 50 yuan in tax filing software equipped with AI capabilities, or to personally organize your materials and educate the AI to perform the task systematically? Furthermore, if you are responsible for preparing your own taxes using an AI file, who will be held accountable for the results? If there is a mistake, it could lead to further complications.

Consequently, software companies may enhance their competitive edge by leveraging AI, thereby streamlining user tasks. If I were to build a boat, I might feel confident, but when it comes to building an aircraft carrier, I'd still worry about whether the internal systems are fully developed, despite how similar it looks on the outside. Instead of expending valuable resources on this endeavor, it would be more prudent to invest in a solution that can deliver tangible results. In essence, software companies enable collaboration and cost sharing for the utilization of their "aircraft carrier" model. This is also the approach many SaaS companies take today: I understand that you have the capacity to solve the problem independently, but I can assist you in achieving a more efficient solution. This will free up your time, allowing you to focus on your core business operations.

Will software be replaced by AI?

The answer is unequivocally "no." If you reflect on the aforementioned analysis, you will see that while AI has a significant impact on software companies, the fundamental principle remains unchanged: as long as AI cannot independently and efficiently develop complex software, it will always require software to perform the work for it. It is not practical for AI to develop complex software because it still lacks a comprehensive understanding of complex workflows and the human world. Bridging these gaps through self-learning is a significant challenge. If AI becomes capable of handling these tasks, I believe the problems we'll face will be far greater than simply software being replaced.

The fact that software won't be replaced by AI doesn't mean all software will survive; what AI is disrupting are the software industry's profit models, use cases, and business logic. The true value of the SaaS model lies in its ability to manage complex processes, a capability that AI will not be able to provide in the near future. Smaller software applications, particularly utility tools, will be significantly impacted.

In summary, we are currently discussing off-the-shelf software, and software companies already possess strong technical teams and development capabilities. It is conceivable that they would adopt AI to transform their software into an AI gateway. It is my understanding that the interaction model, which currently requires users to find and click buttons, would be transformed into a dialogue with AI. This would reduce the cost of using and learning the software, thereby increasing user reliance on the software itself. Take Photoshop, for example. The learning curve is quite steep, but users can articulate their goals with clarity. It would be worthwhile to consider the development of an AI interface, whether in desktop software, web browsers, mobile applications, or the cloud. This interface would allow users to simply issue commands, and the AI would complete the image edits. For power users, a semi-automatic mode might be ideal. They could assign clearly defined tasks to the AI and handle the rest themselves. This approach may be more appealing than a fully automated model that relies entirely on third-party AI.