Will AI replace software?

Recently, the notion that "AI will replace software" has been rampant, causing an unprecedented plunge in software stocks and, coupled with concerns about the profitability of AI, plunging the entire market into deep anxiety. While concerns about AI profitability are not entirely unfounded—given the persistently high costs of AI and the reliance on traditional revenue models like advertising and referrals—the conclusion that AI development will fundamentally disrupt the software industry is somewhat of an overreaction.

This discussion comes against the backdrop of a recent surge in landmark AI applications. First, Clawbot's sudden popularity allowed users to build personalized AI systems on their local devices, automating operations and processes, essentially functioning as a "personal assistant." Then, Anthropic released Cowork, further enhancing the agent's ability to perform complex tasks and demonstrating AI's potential in cross-process collaboration. These developments signal that AI is gradually shifting from an auxiliary tool to a system application with execution capabilities. In the near future, it's entirely foreseeable that AI will integrate with personal devices and data resources, allowing users to command AI to invoke various processes to complete tasks, transforming it from a supportive "secretary" into a capable and efficient "doer."

A more critical issue is that some software may no longer constitute a rigid requirement. In AI-driven workflows, systems rely more on callable, segmented functional modules. For simple scenarios, AI can even dynamically generate or temporarily construct the necessary capabilities. From this perspective, this will indeed be a significant blow to the software industry.

What impact will AI have on software?

Before drawing conclusions, let's take a look at what substantial impacts the development of AI will have on the software industry.

Software development has become easier

I think no one will question this anymore: Previously, developing software required a team spending a long time. With AI, a simple software can be completed with just a few senior engineers handling architecture design, code integration, and final debugging. Clawbot, for example, was developed by a single person (although this analogy isn't entirely accurate, it shows how much AI helps programmers improve their development capabilities). While AI can't write extremely complex programs (and is unlikely to do so in the next few years), it can certainly break down programs into smaller parts, which experienced engineers can then debug and integrate. This significantly shortens the software development cycle and greatly reduces software development costs.

What impact will this have?

First, there are plug-in and tool software programs. Most of these programs are charged monthly and may disappear completely in the future because it is very easy for AI to write a plug-in. Moreover, AI can customize it according to specific situations, making it more in line with its business logic than ready-made plug-ins on the market.

Secondly, there will be more and more alternatives for large-scale software. While writing a completely new large-scale software is impractical, some functions can be implemented by AI, so achieving the goal does not have to rely entirely on a specific program. This will lead some lightweight users to consider other alternatives.

Disruption of the pricing model

Currently, almost all SaaS software is charged on a monthly billing plus per user model, but this model may not be sustainable. As mentioned above, simple utility software will lose user dependence, and the number of users using lightweight applications will also decrease. Therefore, the monthly billing model will no longer be cost-effective for users. This model was already very unpopular, and now that they finally have the opportunity to avoid it, wouldn't they spread the word?

Secondly, charging per user will become increasingly difficult because AI improves work efficiency. What used to require dozens of people can now be done by just a few, or even entirely by a single AI "person," with everyone else using the same AI machine/agent to access the software. In this situation, software companies need to change their business models and find alternative approaches, such as charging based on functionality and usage (similar to tokens).

Software Restructuring

Many large software programs today are feature-rich and have sophisticated interactive features. However, as AI matures, users may only need a portion of these programs' functionality. This is because AI breaks down tasks and plans and implements workflows in the most economical way. For AI, the resource consumption of large software programs differs from that of smaller functional modules; it will inevitably choose the low-resource-consumption process. This will force large software companies to modularize their software, thereby retaining heavy AI users.

This actually aligns with user logic. Although many software programs are feature-rich, most users only utilize a portion of their features, and existing pricing strategies are mostly designed around these popular features.

Change of interaction mode

Currently, software interacts with users through the software's front end, but AI only needs to call APIs, which is completely different. Many complex software programs often require a lot of time and effort to design and develop their front ends, and different customer types may even have different designs. This is one reason why software is difficult to replicate easily. Now, all of this is worthless. For AI, it only needs to understand the API to work. Users also don't see the software's front end, and those complex buttons and menus no longer exist.

App Store?

In the foreseeable future, AI may become a crucial entry point for user interaction, much like Apple's iPhone ushered in the app era. Will a model similar to the App Store emerge? I think it's highly likely. If AI becomes an entry point, all software accessible through this AI will require platform permission, and users will need to purchase the software/functions they need through the platform to create a customized AI work platform. In that case, major AI platforms will have significant bargaining power and may even charge a "tax" on listed software (similar to Apple's tax), further squeezing software companies' profits.

Advantages of existing software

For complex and ever-changing large-scale software (whether engineering or industry-specific), AI is almost impossible to imitate or replicate. AI excels at providing users with a very clear goal, allowing them to break it down and complete it step by step. However, if it's a complex process or environment, where users struggle to accurately describe all the situations, AI will find it difficult to be of assistance.

This is also the biggest advantage of most SaaS companies. Of course, users can break down the entire process into individual tasks and let AI execute them step by step, which I believe isn't difficult to achieve. However, this increases the user's time cost, and after weighing the pros and cons, users may still consider using professional software. To illustrate with a practical example, consider annual tax filing. The forms are publicly available, and the instructions are clearly written. However, people prefer to pay for software or hire an accountant rather than do it themselves. For AI, tax filing is a simple, single task, and I believe AI can replace it in the near future. But compiling and submitting all your tax forms to AI is a complex task, and this is where software companies have an advantage. If they break down the functionality for AI or integrate AI themselves, it will greatly enhance the software's competitiveness. Would you rather pay for tax filing software + AI, or would you rather compile the materials yourself and teach the AI step by step? And if you let AI file your taxes independently, who is responsible for the result? Wouldn't it be even more troublesome if something goes wrong?

Therefore, software companies may actually benefit from more powerful AI, making it easier for users to complete tasks. I might feel confident building my own ship, but building an aircraft carrier, no matter how similar it looks, I'd still worry about whether the internals are fully functional. It's better to spend the money to solve the internal issues than to spend the time and effort on the external aspects. Essentially, software companies are like everyone sharing the cost of using an aircraft carrier. This is also the thinking of many SaaS companies today: I know you can solve the problem yourself, but I can make you more efficient, freeing up your time to focus on your core business.

Will software be replaced by AI?

The answer is clearly no. Careful consideration of the above analysis reveals that AI's impact on software companies is enormous, but the core issue remains: as long as AI cannot independently and quickly develop complex software, it will always need software to function. However, AI developing complex software is almost unrealistic because its understanding of complex workflows and the human world is still far from sufficient, and these gaps are difficult to bridge through self-learning. If AI were to one day be able to handle these tasks, I believe the problems we face would be far greater than software replacement.

While AI won't replace software, it doesn't mean all software will survive. What AI will disrupt is the software industry's profit model, use cases, and business logic. The competitive advantage of SaaS isn't the ease of software development; its true value lies in handling complex processes, something AI cannot do in the short term. Smaller software, especially utility software, will be more significantly impacted.

To put it another way, we're talking about readily available software. Software companies themselves have strong technical teams and development capabilities. Wouldn't they embrace AI and make their software an AI entry point? Wouldn't this transform the interaction from searching for and clicking buttons into communicating with AI? Wouldn't this lower the learning curve and increase user reliance on the software itself? For example, Photoshop has an extremely high learning curve, but users can clearly describe their goals. So, shouldn't we consider designing an AI entry point—whether on desktop software, a web page, a mobile app, or even in the cloud—where users simply give instructions and it can complete the image modification? For heavy users, a semi-automatic mode might be considered—handling clearly defined tasks to AI and manually handling less clearly defined tasks. Wouldn't this be more attractive than a fully automated mode that relies entirely on third-party AI?

In summary, the idea that AI will completely replace software is a pipe dream, but it will certainly impact the software industry. It will reshape the entire industry and, in the near future, disrupt how users use software. Companies that can adapt and change will not be greatly affected, and this will present a significant opportunity for small software businesses.