Duolingo lost 68% of its market cap in six months. HubSpot, Figma, and a dozen other SaaS companies are down over 65%. The entire SaaS industry has shed more than $300 billion in market cap recently. And the common explanation you'll hear is a three-word sentence: AI killed SaaS.
I don't buy it.
I use AI every day to build software, teach executives, and run digital operations at GE Aerospace. I have not once replaced a SaaS product with an AI agent. Slack is still open. Salesforce still runs our CRM. Nobody I know has swapped HubSpot for a chatbot. So what's actually going on?
The answer is more interesting than "AI is replacing software." AI is replacing the way we interact with software. And that distinction, subtle as it sounds, is worth $300 billion.
A New Layer of Abstraction
The video that got me thinking about this (from Kale Bryce Code) walks through the economics clearly. SaaS companies have always traded at high PE ratios. Ford trades at 8 to 12 PE. Duolingo has hit 50 to 200 PE. The reason is simple: SaaS revenue is predictable because subscriptions are predictable. Investors love annual recurring revenue because the math is clean. Count the seats, multiply by the price, project the growth curve.
AI hasn't replaced the applications underneath those subscriptions. What's happened instead is a new abstraction layer forming on top of them.
Look at the timeline. MCP (Model Context Protocol) launched in November 2024. Agent-to-agent protocols followed in April 2025. Skills arrived in October 2025. Each release adds another way for AI agents to connect to existing applications and operate them without a human clicking through the interface.
The applications still exist. The databases still run. The infrastructure is intact. But the user is no longer the person sitting at the keyboard. The user is increasingly an AI agent, and that changes everything about how SaaS companies make money.
The Seat Problem
SaaS companies charge per user, per seat, per month. Salesforce's pricing ranges from free to $100 per user per month. That model works beautifully when every employee needs their own login and their own license. Investors can project growth by tracking how many seats a company sells quarter over quarter.
Now imagine one AI agent with a single subscription, serving ten people. The agent logs into Salesforce, queries the CRM, updates records, generates reports. Ten people get the value. One seat gets billed. The company that built its entire valuation model on seat growth just lost nine paying customers.
This is not theoretical. I see it forming in my own work. When I connect Claude Code to documentation, databases, and internal tools through MCP, I'm not navigating each application's interface anymore. The agent reads the docs, calls the APIs, and returns what I need. The application becomes invisible. It's still running, still necessary, but its interface (the thing that used to differentiate it from competitors) becomes irrelevant.
And that's the core of the problem. SaaS companies spent years building intuitive, beautiful interfaces as their competitive edge. That edge erodes when the primary user of your product is an AI agent that doesn't care about your UI. An agent connects through an API or MCP, follows a skills file that explains how to interact with your system, and gets the job done. No onboarding flow needed. No tutorial videos. No carefully designed dashboard.
The application layer is being commoditized. Not eliminated. Commoditized. The data and the logic still matter. The interface doesn't, at least not the way it used to.
The Pricing Model Has to Change
This commoditization is forcing SaaS companies into a pricing transition that Wall Street doesn't know how to value yet.
Per-seat pricing gave investors a clean, predictable metric. ARR (annual recurring revenue) was the gold standard for SaaS valuation. You could model it in a spreadsheet. You could compare it across companies. You could project it forward with confidence.
Usage-based pricing is messier. Instead of counting seats, you're counting API calls, tokens consumed, compute time, storage used. The revenue becomes variable. It depends on how much agents use your platform, how many tasks they process, how complex those tasks are. Good luck putting a clean growth curve on that.
This is exactly what's happening. Goldman Sachs projects the total addressable market for AI agents to grow above $50 billion, while traditional SaaS is set to decline. 2026 is where the lines cross. Venture capital is already following: investment is moving away from traditional software products and toward agent-native companies.
What used to be 50 to 100 PE ratios is compressing because investors can no longer confidently project how long these companies remain viable at their current pricing. The uncertainty isn't about whether the software works. It's about whether the business model survives.
SaaS Companies Are Fighting Back (And It's Working)
The narrative that SaaS is dead oversimplifies what's actually a more nuanced response from incumbents. According to a Deloitte survey, more than half of companies are expected to spend up to 50% of their budget on AI automation. They're not sitting still.
Salesforce built AgentForce. ServiceNow launched Now Assist. HubSpot has its own agent layer. Basically every major SaaS company now offers built-in AI agents that know how to navigate their own application out of the box.
The strategy is straightforward: if agents are going to interact with our platform anyway, we'll build the agents ourselves and keep users inside our walls. It's a defensive moat, and a smart one. If Salesforce's own agent is the best at operating Salesforce, there's less reason for you to bring an external agent that might eventually replace the need for Salesforce entirely.
I watch this play out in how companies I work with adopt AI. The first instinct is always to use the platform's built-in AI features. It's easier, it's integrated, and IT departments prefer it because the data stays within existing contracts and compliance frameworks. External agents come later, usually when the built-in features can't handle cross-platform orchestration.
The incumbents aren't dying. They're adapting. But the adaptation requires changing everything about how they price, sell, and differentiate their products. That transition is painful, and markets price in pain.
Vibe Coding and the Replication Problem
There's a second pressure point that compounds the pricing problem.
The video mentions that last June, around 10% of websites in the world were built using Lovable, one of the AI coding platforms. That number is striking, not because vibe-coded software will replace well-engineered systems (it won't), but because it demonstrates how quickly the application layer can be replicated.
I build with Claude Code every day. I've gone from simple scripts to multi-file refactors touching dozens of interconnected modules. A year ago, building a functional CRM prototype would take weeks of engineering time. Today, a solo builder with the right AI tools can have a working version in days.
This puts additional pressure on SaaS companies whose primary value proposition is the application itself. If building a good-enough alternative takes days instead of months, the switching costs that protected SaaS margins are eroding. The data lock-in still exists (migrating your customer data from Salesforce is still a nightmare), but the application lock-in is weakening.
For SaaS companies, the defensible moat is shifting from "we built a great product" to "we hold your data and your workflows, and switching would break too many things." That's a weaker position than it sounds. It's the same argument mainframe vendors made in the 1990s.
What I'm Actually Seeing
I sit at an unusual intersection for observing this. At GE Aerospace, I lead digital innovation projects where enterprise SaaS tools are deeply embedded in operations. At the university, I teach venture design and AI adoption to the next generation of builders. And in my own projects, I build with AI agents daily.
In enterprise, SaaS isn't going anywhere soon. The contracts are long, the integrations are deep, and the switching costs are real. But procurement teams are starting to ask different questions. Instead of "how many seats do we need?" the conversation is shifting to "what's the usage-based pricing for agent access?" That's a signal.
In startups, the default architecture is changing. A year ago, every new project started with "which SaaS tools do we need?" Now it starts with "what can agents handle directly?" The SaaS layer is becoming optional for early-stage companies in a way it wasn't before.
In education, students are building functional applications in hours that would have required SaaS subscriptions a year ago. They're not replacing Salesforce. They're making the argument for Salesforce harder to justify for small teams and early-stage projects.
Where This Goes
SaaS isn't dead. Calling it dead is the kind of dramatic oversimplification that gets clicks but misses what's actually happening.
What's dying is the per-seat pricing model. What's dying is the assumption that a beautiful interface is a sustainable competitive advantage. What's dying is the idea that SaaS growth curves only go up and to the right.
What's emerging is a new architecture where AI agents sit between humans and applications. The applications become infrastructure, essential but invisible, like databases and cloud services before them. And infrastructure gets priced differently than products. It gets priced on usage, on throughput, on value delivered rather than seats occupied.
For builders, this means the opportunity is at the agent layer, not the application layer. The companies that figure out how to orchestrate across multiple SaaS platforms, connecting data and workflows through AI agents, will capture the value that's leaking out of traditional SaaS.
For leaders evaluating their software stack, the question to ask your vendors is simple: what's your agent-native pricing model? If they don't have one yet, they will soon. And when they do, your cost structure for that tool is going to change.
The $300 billion in lost market cap isn't a verdict on SaaS. It's the market repricing an industry that built its valuation on a model that AI is making obsolete. The software still works. The business model is what's breaking.
What's your read? Are the SaaS companies you use adapting fast enough, or are agents already changing how you interact with them?