There was a necktie I wore every morning. One day, standing in front of the mirror, I asked myself: Why am I wearing this?
Atlassian has evaporated 76% from its 52-week high. Salesforce, 56%. Figma has been setting new 52-week lows since shortly after its IPO. Asana, HubSpot, too. There are no exceptions. Naval Ravikant's "Software was eaten by AI" is an exact reversal of Marc Andreessen's 2011 declaration, "software is eating the world."
SaaS was a necktie that went beyond being a tool. It was about formality, not functionality. If you didn't wear one, you became someone who didn't understand work. Jira made the whole team wear a necktie every Monday. It made them declare their work, assign difficulty scores, and attach progress markers. HubSpot created a process for putting customers into columns and moving them from one to the next. Figma brought designers and planners together on one screen and made them leave endless comments. People repeated actions unrelated to productivity, and no one questioned it. The moment you asked why you were wearing it, you became a heretic.
"Our team uses Jira." It wasn't a tool choice but a declaration of identity. Saying you used Notion was a signal that 'we are that kind of team.' SaaS's only moat was not technical superiority but this habit.
For this dress code to remain in place, one premise was necessary: someone had to put on the necktie themselves. People registered tasks, people entered customer information, and people drew on screens. A necktie exists only if there is someone to wear it.
Then, in an instant, that premise collapsed.
AI produces results without formality. It solves problems without tasks being registered, completes code without weekly planning, and creates designs without opening a screen. When something that doesn't need to wear a necktie gets the work done, the necktie is useless.
Figma has been falling since shortly after its IPO. This is not a crisis for design tools. It is a crisis for the premise that "people have to gather and work together." What Figma sold was not a design tool but friction in the name of collaboration, and the more friction there was, the longer people subscribed. The moment Figma automates design with AI, the reason for people to open Figma disappears along with it. A feature introduced to save the tool from itself erases the reason the tool exists.
The revenue model called "per-seat pricing" collapses along with it. Three agents replace a tool that 1,000 people used to log into. Agents don't occupy seats. If there is no one to sit in the seats, the formula breaks.
Why did this happen? SaaS was, by nature, a business designed to make "human inefficiency" its source of revenue. Because people forget, it created reminders. Because people fail to communicate, it created collaboration tools. Because people make mistakes, it created workflows. It put a price tag on each of the limits of human cognition. The problem is that a more fundamental way to compensate for those limits has appeared. SaaS was a building erected on human flaws, and AI is technology that reduces those flaws themselves. It isn't the building that changed. The ground did. Remodeling won't solve it.
The second shock is broader. The entire ecosystem built on top of the necktie is shaking. Salesforce implementation consultants, Jira training companies, HubSpot-certified partners. When necktie culture disappears, the stylists who chose the ties and the schools that taught people how to tie them disappear too. The real size of the SaaS ecosystem is not subscription revenue but the entire service economy built on top of it.
Korea is taking this shock from a different angle. Korea's SaaS ecosystem is small compared to the global market, and most of it runs on outsourced development centered on large companies. There are a few exceptions, but the market itself has been less dependent on subscription billing models. The age of the closet ended before it was even filled with clothes.
Soon, AI agents will cancel SaaS subscriptions on their own. Agents have no habits. They have no loyalty. They move in 0.1 seconds toward whatever produces better results. A necktie with no one to wear it. That will be SaaS's final landscape.
The next scene is even stranger. What remains where SaaS disappears is only APIs. Software without screens. Atoms of functionality that agents call and respond to, without dashboards for people to look at or buttons for people to press. Until now, the value of software was measured by "how long people stared at the screen." That metric will be overturned entirely. From now on, it will be "how often agents call it." In a software market without UIs, brands lose their meaning, prices are broken down by API call, and competition is decided solely by response speed and accuracy measured in milliseconds. An era when invisible software beats visible software. Then, a "good product" will not be one people like but one agents choose.
Now oxygen tanks have appeared. ChatGPT, Claude, Gemini, Grok. We could live without wearing a necktie. It was a habit. Oxygen tanks are different. Without one, you can't breathe. AI has become survival infrastructure, not a choice.
The necktie called SaaS held human habits as collateral. When a more comfortable necktie appeared, we switched. The oxygen tank called AI holds physical resources as collateral. Semiconductors are needed to run thought, and electricity is needed to run semiconductors. Mines do not appear overnight, and power plants are not built by declaration. You could untie a necktie, but you can't take off an oxygen tank. The class of the collateral is different.
When neckties disappear, dress codes disappear too. Two-week sprints, pipelines that put customers into columns and move them along. These were not universal truths. They were simply the formalities of a particular era, enforced by the tools of that era. When the tools disappear, the very concept of "how we work" becomes a blank page. On that blank page, the standards change. Results, not process. Performance, not formality. The industrial-era contract that said a good worker was someone who faithfully followed the process is broken. In a world where only results remain, process is a cost.
A deeper crack is coming. When SaaS disappears, the outlines of the "organization" blur too. Who is on which project, what the goals are for this quarter. The things that told us this were not managers but Jira, Slack, and Notion. People wearing the same necktie became a team. When the neckties disappear, the ranks scatter too.
The irony is that SaaS companies themselves are the last to realize this. Atlassian talks about "smarter project management" as it adds AI to Jira. The possibility that project management itself is disappearing does not exist in that language. As long as you interpret the world through the language of your own product, you cannot see the future outside that language.
Without realizing it, I too have started giving greater weight, when reviewing investments, to teams that build better tools for AI than to teams that build better tools for people. The question follows. What will the ultimate moat be for tools built for AI?
UX is meaningless to agents. Simple speed or accuracy is not enough. Isn't the real moat creating a structure that agents voluntarily gather around? Agents use data while leaving data behind at the same time, and that data improves the performance of other agents. A vein of data in which the act of using it generates the raw material.
No matter how far open source advances, not all data can be free. Scarce data remains scarce. I think about this every day: what structures they will form into a value chain, and how incentives between agents should be designed.
YC's Garry Tan defined the companies that will survive this crash like this. Companies led by CEOs who write code themselves and stand at the forefront of open source, like Jack Dorsey and Tobi Lutke. CEOs who are still in manager mode? I'm not so sure.
The questions converge. That SaaS subscription you're paying for—is it money for a tool, or the price of a necktie? What is your oxygen tank connected to? And the data you are creating right now—is it becoming someone else's vein of ore?
