Since early this year, fears of a "SaaSpocalypse" have struck the software industry following the arrival of Anthropic's agentic (assistant-style) artificial intelligence "Claude Cowork" and its coding tool "Claude Code." The term combines "Software as a Service" (SaaS) with "Apocalypse," reflecting the argument that the spread of AI will sharply reduce the need for subscription-based software and leave software companies with nowhere to stand. Shares of major SaaS firms such as Salesforce and Adobe have plunged more than 20% since the start of the year. While the SaaSpocalypse fears have calmed somewhat as Anthropic clashed with the Donald Trump administration over AI use, the market is likely to be shaken again each time a new AI model is announced.
How does the software startup market view the current situation? Seoul Economic Daily heard the thoughts of Akshay Kothari, co-founder and chief operating officer (COO) of Notion, a collaboration tool service company with 100 million users worldwide. Since its founding in 2013, Notion has developed into a platform that boosts work efficiency by performing document creation, schedule management, and AI utilization on a single platform. It entered the Korean market in 2022. OpenAI and Nvidia are known to have used Notion as an internal collaboration tool from their early founding days.
Ahead of Notion's launch of its "Custom Agent" (February 24, local time), Seoul Economic Daily visited the company's San Francisco headquarters and interviewed COO Kothari. Last month, Notion launched the Custom Agent, an AI assistant that automatically executes tasks. While previous AI was limited to assistant functions, the Custom Agent has evolved into a form that automates and performs repetitive tasks without requiring instructions each time, once the user sets up a database and guidelines. In particular, when an agent is shared, all members can use it together, and it integrates with Slack, email, and calendar.
Q. Software companies are struggling against Anthropic's AI models. What are Notion's strengths?
A. Notion has supported a multi-user environment from the beginning. Our focus has always been on enabling teams to utilize it. Notion AI started in the early days of our founding. I remember our first AI product came out about two weeks before (OpenAI's) ChatGPT was launched. We have repeated this process for three years since then, and we have been able to gradually transition from legacy software to an AI world where everyone wants to interact.
I think Notion was an Anthropic partner, because we used their models and did various tasks. But the reason people choose us is that if you really think about how to improve company productivity, you need a foundation with all the permissions and connectivity to a variety of tools.
We want to help colleagues use what one person in a company has created. Most existing tools focus on individual productivity. An individual user connects (Google) Gmail or (Salesforce) Slack. But with Notion, once an administrator connects it, you can directly access everyone's email.
Q. There are many concerns about paid subscription services because of Anthropic's AI models. How do you forecast the response to the paid Custom Agent?
A. I'm really excited about the future. In some ways, I think we are in the process of transitioning from selling software to selling work. While software spending accounts for about 2% of U.S. gross domestic product (GDP), knowledge work makes up as much as 50%. That means this (knowledge work) opportunity is 25 times bigger.
We need to rethink the way people price the output of work done using collaboration tools. It feels logical to think that you would want to pay only for output where actual work has been completed. We can let people try this for various uses and choose whether it is a service they want to pay for.
When classifying email, you can choose whether to use (Claude) Opus once, or freely use an open-source model that is 100 times faster. We will give users that choice. That's why we are not charging fees for two months after launch.
People may think a data scout (data analyst) is expensive, but it actually helps you do things you couldn't do before. People will eventually recognize this. You could hire a data scientist, but it costs as much as $200,000 (300 million won). Or instead, you could invest in a work collaboration tool. It is surprising that so many people are already using these services so much. It is exceeding our expectations. We'll see how it goes.
Q. Software company stocks are plunging. What is your prediction for the future? (This question was answered by development lead Max Schoening.)
A. The future of software is really hard to predict. I think you can argue both sides. That's the case when you look at the history of computing. For 10 years, there was a high degree of specialization in the software field. Looking at the SaaS companies most affected by recent changes, I think there is now a move to break away from specialization.
There are now very general AI tools. Even if you don't know how to make a presentation, it's not a big problem, because (the agent) will learn it on its own. I think this is one axis of hyper-specialization. We'll see how it goes.
Highly specialized software will struggle, but instead, new specialized software will emerge that solves the last step of a specific problem in depth. I think this is a natural evolutionary process where bundling and unbundling repeat. I think many companies are now at a point where they can bundle very aggressively. That's true from experience as well. It has now become much easier to sell knowledge services, because most functions are bundled together within the workflow.
Q. Can the number of subscriptions grow significantly?
A. The goal may no longer be the actual number of users. Rather, it is more focused on workload. For example, seat-based pricing companies are usually interested in whether 50% of total customers are using it, but our goal will likely be total token (a unit of AI model input and output) usage. That's because our business going forward will also depend on how much people use it. And we will earn revenue from tokens. Over two months, we will look at whether users find the results useful and are willing to pay for them.
The more seats there are, the more people use it. But now we are seeing a new trend in workspaces with two or three users. For example, there are cases with 270,000 agent executions a month, with only three people using it. Now a small number of highly motivated people can do a lot of work. For a company that relies entirely on the number of seats, the only way to grow is to keep increasing seats. If a seat costs $10 a month, you need 500 seats to earn $5,000. In the early development stages, we were intrigued that a small number of seats could pay as much as $3,000.
As more and more people use data scout tools instead of data scientists, their value is becoming apparent. They think that instead of paying $250,000 (to a data scientist), they only have to pay $3,000 (for the tool cost). I believe the market will eventually understand this. There are existing software companies that grow by relying on the number of seats, while there are others that do not. We are the same. People are using AI very actively and are willing to pay for it, and this trend will continue steadily.
