This past March, at a demo day for a San Francisco startup accelerator, one team put up a slide. The title was short: "We don't use SaaS." The audience laughed—and before the laughter had faded, twelve investors raised their hands. That team walked away with more business cards than anyone else in the room that day. "SaaSpocalypse"—the end-times of software—became Silicon Valley's biggest buzzword of 2026, and the question it raised crossed the Pacific to land on the monthly invoice of every Korean founder.
The Claim That Software Has Started Eating Itself
The logic of SaaSpocalypse is simple. As AI agents began automating complex workflows, it became possible for a single agent to take over tasks that had once been split across separate subscription tools—CRM, project management, customer support, design drafts. From a company's point of view, if one AI agent can organize sales data and fire off follow-up emails instead of maintaining an annual Salesforce contract, the logic that justifies the existing subscription starts to wobble. The entire premise of the subscription-software market was that it "handles the tasks too tedious for humans to do"—and now agents have begun handling those tasks more directly.
Several observations converged to sharpen the debate. The agent features from OpenAI, Anthropic, and Google began handling complex, multi-step work in the second half of 2025, and in Y Combinator's latest batch a noticeably larger number of startups positioned themselves around "replacing a specific SaaS category with an agent." The stock prices of flagship SaaS companies like Salesforce, ServiceNow, and Zendesk wobbled every time an AI-agent announcement landed—a pattern that kept repeating. It was a signal that investors had begun taking the structural shift seriously.
When The Economist reported on the trend head-on, it was in the context of the debate spilling out of Silicon Valley's insider community into the broader conversation about management strategy. The forecast that "30% of the SaaS market will be replaced by agents by 2030" began circulating not as a radical scenario but as a conservative estimate—and applied to a global SaaS market estimated at $4 trillion, it is not a number you can wave away. This isn't a passing tech fad; it means the very revenue structure of the software industry is shaking.
The shift feels close in the Korean context too. A large share of domestic startups and solo operators run on some combination of Notion, Slack, Figma, HubSpot, and Zapier. A SaaS subscription stack running 150,000 to 500,000 won a month is hardly unusual. Add team collaboration tools, accounting software, and an email-marketing platform on top of that, and some end up spending several million won a year. The claim that this stack could be replaced by a single unified agent within six months to a year lands as pressure to re-examine your fixed-cost structure.
But SaaS Won't Disappear as Fast as You Think
There is a substantial body of skepticism toward this outlook. The argument is that the claim—that SaaS will be replaced by AI agents anytime soon—rests on several hidden premises, and that not all of them hold.
The most frequently raised objection is the structural cost of switching. SaaS tools don't just provide features. Notion holds years of accumulated document structure and team collaboration habits; Slack holds the channel history with customers and partners; HubSpot holds deal history and sales-pipeline data. Migrating all of that to an AI-agent-based system is not simply swapping out software. It pulls in data migration, team training, the redesign of work practices, even an adjustment of how you collaborate with customers. That cost often runs well past several months' worth of subscription fees. This is why the simple math of "cancel and save" doesn't add up.
The incumbent SaaS companies are pushing back hard, too. Salesforce launched its own AI-agent platform, "Agentforce"; Notion integrated an AI assistant as a core feature. HubSpot has likewise begun bundling AI-driven automation into existing subscriptions. They are reshaping themselves to fold agents into the platform from the inside. The speed at which AI-native startups grab SaaS's seat and the speed at which incumbent SaaS companies absorb AI features are competing right now. Which one moves faster is still an open question.
The question of how mature agents actually are can't be left out either. Today's AI agents handle simple, repetitive tasks well, but limits remain when it comes to reliably understanding complex business logic or context specific to a given organization. The declaration that you can "manage sales with agents alone, without Salesforce" draws attention—but how many teams actually run that reliably is a question of a different order. Most of the cases made public are small-scale pilots, or applications to simplified workflows. It is worth keeping in mind that the SaaSpocalypse discourse spread fast on a mix of hope and fear, more than on actual operating experience. Replace your SaaS stack in a hurry without seeing that gap, and bigger operational problems—feature holes and data discontinuity—may be waiting.
What to Do Right Now Is Write a Task List, Not Swap Tools
So what action does this debate demand from Korea's solo operators and small-scale founders?
Research on the future of work has confirmed a pattern over and over for the past several years: automation tends to replace specific "tasks" first, rather than eliminating an entire job. Researchers in the field say the real impact of automation becomes visible when you break a job down not as a bundle of roles but as a set of tasks. AI agents don't stray far from this pattern. They don't replace all of Notion at once; instead, specific tasks inside Notion—recurring meeting summaries, fixed-format document drafts, tag classification—move to agents first. This lens fixes a clear starting point for re-examining your SaaS stack. Before asking whether you need to swap a tool wholesale, ask first which tasks inside that tool can move to an agent.
Here's the order for a practical audit. Write down the list of SaaS tools you use now, and next to each one, note the tasks you actually perform on repeat. If the tasks that "an AI agent could handle with little difference in the outcome" make up more than half of your time using that tool, then that tool is a candidate for replacement over the medium term. For a tool where team collaboration history, customer data, and long work records have piled up deeply, waiting for AI features to be built in is more realistic than swapping it out right away. Before rolling out a new agent tool across the board, apply it to a single pilot project for six weeks first, confirm the switching cost and the feasibility of data migration firsthand, and then decide—that is the right order.
You also have to account for the particulars of the Korean market. When you collaborate with domestic companies, local tools like KakaoWork and Naver Works are often the de facto standard. Even if you replace your global SaaS stack with agents, domestic collaboration tools are likely to stay for the time being, for ecosystem reasons. The SaaSpocalypse debate is aimed mainly at the U.S. enterprise market, and its starting point differs from the tool-selection criteria of Korean small businesses. Before importing a debate spreading in the U.S. and applying it as-is, it's right to check the ecosystem of the market you actually belong to first.
Here's how I would put it. What Korean founders need in this debate is not the resolve to swap their stack in a hurry. It is an accurate grasp of which tasks your current tools are used for, and how much. Changing tools without a task list is no different from buying moving boxes before you know what you need to pack.
Teams that merely watched SaaSpocalypse will, at some point, pay the cost of a switch made a step too late. Teams that rushed to swap their stack, driven by fear, pay an even steeper tuition. Somewhere between the two lies the thing to do right now.
