iTranslated by AI
As a Solo SaaS Developer, I Am Starting to Feel That "SaaS is Dead" Is Real"
Hello. I'm Waikei.
I develop and operate SaaS as an individual.
Recently, I have been seeing the phrase "SaaS is Dead" quite often, and honestly, even from the perspective of an operator, I have to admit that it is becoming a situation where I "get it."
In this article, I will write about the reality of the "SaaS is Dead" argument from the perspective of a micro-SaaS operator, future predictions, and what I have actually started doing to survive.
What is the "SaaS is Dead" argument?
First, let's look back at what the "SaaS is Dead" argument is.
The "SaaS is Dead" argument is the debate that the very reason for the existence of SaaS is being shaken by the rapid spread of AI agents like Claude Code.
There are mainly two aspects to this.
1. The "I can just build it" problem
For disposable tools or simple in-house tools for personal use, if you rely on Claude Code, it can build them for you in a few minutes. You no longer need to master various SaaS products; it is sufficient to just vaguely command an AI you are used to using in plain language. It can be said that half-baked services have lost their fundamental reason for existence.
2. The "User reduction" problem
In user-based SaaS products, the number of human staff required decreases due to the use of AI, so the number of contract seats per customer company decreases. Billing for additional features also becomes unnecessary if AI replaces them. In other words, a structural mismatch occurs where revenue drops even if the customer does not cancel.
Why I am feeling that "SaaS is Dead" is real
There are various opinions regarding the "SaaS is Dead" argument, both affirmative and negative.
However, if you are a SaaS operator, whether corporate or individual, I think many of you have at least some anxiety that "won't I be wiped out too?"
Let me talk a bit about my own case.
I operate a micro-diary service called MydayAI.
It is a simple service where you can write a diary on the web, and about 30 to 40 diaries are written every day.
It's a service I made myself, so it's a bit awkward to say, but considering it by the standards of the AI era, to put it mildly, it is a hassle to use.
In the first place, "continuing to write" is the biggest hurdle for a diary. You have to open a browser, click around on the screen, input text, and so on...
Compared to the experience of asking an AI agent to do something, I have to say it is a hassle.
On the other hand, when I use AI agents every day, they generally handle things flexibly if I just ask vaguely without thinking about complicated operation procedures.
Compared to that experience, my own SaaS feels stale and uncool. I am getting discouraged by the service I created myself.
In the first place, for a personal diary level, each user can have an AI agent build it. You can have Claude Code or Codex create a UI for diary entry, and it is theoretically possible to accumulate the diary data itself locally as a text file.
I have recently started to wonder, do users really need to use such SaaS?
Of course, MydayAI is a relatively simple service like a diary, so it is in the category of being easily replaced by AI. However, even for more complex and large-scale SaaS—for example, CRM or project management tools—if AI automates data input or autonomously generates reports, the value of "humans clicking around on the UI" will definitely diminish. Regardless of the scale, the root of the anxiety that SaaS operators are feeling is likely the same.
Many SaaS operators are obsessively tracking detailed churn rates, but at the same time, many may not be able to face the question they feel deep in their hearts: "Isn't this service fundamentally replaceable by AI?"
What will happen as AI continues to evolve and spread?
I will write my personal prediction on what will actually happen if AI continues to evolve and spread as it is.
In conclusion, I think many SaaS products will be forced to effectively transform into "AI plugins."
Along with this, I believe there will be significant changes for both the user side and the SaaS provider side.
I will explain this from the sequence of events that led to this thought.
Recently, I had the opportunity to hear from several work partners, "In the future, we want to build SaaS where AI can autonomously proceed with work, rather than humans."
Here is an image of what that means.
For example, in a time and attendance tool, traditionally, humans needed to punch a time card every day.
In the AI era, AI autonomously detects clock-in times from PC startup logs, etc., and notifies the user via Slack: "I have clocked you in for this time. Please point it out if any corrections are needed." That is a shift to such a UX.
A structural change occurs where AI operates the SaaS and asks the human for confirmation, rather than the human actively operating the SaaS.
In the first place, it has always been troublesome for humans to master various business tools.
However, SaaS providers (often lying to themselves as well) have focused on promoting "our tool is easy to use!" and have made efforts to force their own SaaS to permeate within companies by involving those in charge at client companies.
However, now, the atmosphere has changed so that the fact that "using SaaS is a burden on humans" can be freely asserted even in discussions within SaaS providers.
I personally feel that this change is significant.
And, if the number of operators who think "Let's make SaaS where AI can autonomously proceed with work" increases, what kind of world will be realized in the future?
I predict that instead of a world where "users have AI build tools individually," a world where "users command their local AI to operate existing SaaS" will be realized.
In the first place, no matter how powerful AI becomes, it is not rational to have it build everything from scratch.
I said at the beginning that "it has become easy to build disposable tools with AI," but conversely, it is quite difficult to actually build and continue using things other than disposable tools with AI.
As a point of discussion separate from the story of AI's development capability, for example, we must consider the following points:
- It is risky for non-engineers and AI alone to build services where security is important.
- It is inconvenient to hoard data that requires persistence on a local PC (synchronization between devices, migration when buying a new PC...)
- It is better for a small number of operators to operate large-scale servers and have a large number of users share them for overall optimization in terms of cost due to economies of scale.
Considering these circumstances, the claim that "anyone can build anything with AI!" seems a bit unreasonable.
Against the background of these circumstances, I think such a model might be possible as the form of SaaS in the future:
- SaaS providers prepare a DB and publish an API that AI can operate.
- Users operate the SaaS through that API from their local AI agent or a major AI service like ChatGPT.
This structure has advantages for all stakeholders.
| Stakeholder | Merit |
|---|---|
| SaaS provider | Can significantly reduce UI development costs |
| AI provider | SaaS features are added to their own AI, and the ecosystem is strengthened |
| User | No need to memorize SaaS operation procedures; just ask the AI and it's done |
It could be said that AI is no longer the UI of SaaS, but rather a foundational existence like a browser on the web.
Until now, SaaS providers have developed services on the premise that "users have a browser." When someone decides, "I'll make a service!", no one thinks, "I'll also develop a browser for access."
In the same way, the premise is that powerful LLMs will be created by providers such as OpenAI, Google, and Anthropic, and we will create services that are easy to use from those AIs—I predict this will be one of the trends in the future.
If that happens, foundational LLM providers will hold the entrance to the majority of SaaS, and their power will be tremendous.
Services currently called SaaS may transform into mere AI plugins before we know it.
However, I do not see this as a dystopia for SaaS providers. Conversely, it can also be seen that SaaS providers are having the enormous costs of foundational model development shouldered by them. It is more of a symbiotic relationship based on a certain power gradient rather than one-sided exploitation.
This structure is very similar to the cloud. AWS, GCP, and Azure are also the foundation of the IT industry, and since the capital and technology required for construction are extraordinary, a few US companies are monopolizing the market, but usage fees are decreasing year by year. As long as multiple big techs compete, I believe it is unlikely to become a dystopian situation.
What can we do now to avoid being wiped out by the "SaaS is Dead" trend?
Based on this situation, I believe there are two major things that SaaS operators can start doing immediately.
1. Transition to an AI-friendly structure
The most direct action is to add an interface that is easy for AIs to operate.
There are several concrete ways to do this:
- Develop and publish a CLI dedicated to your service
- Develop and publish an MCP server
- Improve external-facing REST APIs, etc., to make them easier for AI agents to call
2. Rethink your billing model
As long as you use a per-user billing model, your revenue will shrink as AI-driven operational efficiency improves. You should consider shifting to a billing system that fits the AI agent era, such as per-API-call billing, data storage volume billing, or performance-based billing.
I actually added an MCP server
In my case, I wanted to start by moving my hands, so I added functionality to MydayAI as an MCP server.
This allows users to do things like the following:
- Have an AI agent write a diary for them (Freedom from tedious input work)
- Have an AI agent reference past diaries (Utilize their own records as context to improve the quality of daily AI conversations)
In fact, since I myself have been finding it a hassle to write my diary, I want to try out a way where it automatically writes my diary at midnight every day by referencing my calendar and email history.
Since it has become easier to set up periodic executions with Claude Code, this is something that can be achieved right away.
Of course, MydayAI is not a service for engineers, so there are few users who use AI agents on a daily basis; therefore, MCP integration will likely not be used explosively for the time being.
However, based on the trend that Claude Code is rapidly spreading even to non-engineers, my judgment is that I want to implement it while I have the chance.
I want to keep an eye on how many people will operate SaaS via AI agents.
Becoming an MCP server is not that difficult
Technically, publishing an MCP server is not that different from existing API development.
General-purpose frameworks and libraries for building MCP servers are already emerging, and if you add OAuth authentication for MCP servers, many services should be able to support it immediately.
Before lamenting that "SaaS is Dead," try creating one point of contact with AI agents first.
Just by doing that, I think the scenery you see will change quite a bit.
Conclusion
"SaaS is Dead" is not just inflammatory language; it reflects the reality that the significance and business models of SaaS are structurally changing due to the evolution of AI.
However, that does not necessarily mean that "SaaS will vanish."
By providing AI-friendly interfaces for services and revising billing models to fit the times, there is a possibility that they can demonstrate new value as part of the AI ecosystem.
I wrote at the beginning that I "feel like SaaS is Dead," but to be precise, I believe that while some SaaS products that cannot adapt to the changing tides of the era will be wiped out, many others will survive by changing their form. This is essentially the same as the changes in eras that have happened many times in the past, such as the wave of cloud computing and the mobile shift.
It may be a cliché conclusion, but I think the constructive approach is to perceive it not as "SaaS is Dead," but as "SaaS is Evolving."
I also started with the small step of adding an MCP server. The MCP support has only just been released, and honestly, the visible changes are yet to come.
I would like to share my findings on X and other platforms regarding how much usage via AI agents will actually increase and what kind of changes will occur in how users use it.
I usually post about web development and AI agent utilization on X (formerly Twitter).
Please follow me if you like.
Discussion