In the global software industry, the "moat" once meant complexity. Good software is hard to write and even harder to maintain. Companies pay tens of thousands of dollars in annual subscription fees not because they love a product, but because they can't build one themselves. This logic sustained the SaaS industry for two decades. From Salesforce to HubSpot, from Slack to Notion, countless firms built billion-dollar ARR empires on the premise "you can't code, so you rent."
But starting in 2025, that logic is breaking apart — not because of a better SaaS competitor, but because of a tech revolution that lets everyone write code.
SaaS Winter: Customers Start Building Their Own
Numbers don't lie. Since the start of 2026, a Morgan Stanley basket of SaaS stocks has fallen 15%, deepening a 11% slide in 2025, marking the worst start since 2022. HubSpot, Klaviyo, and other former stars saw sharp share price drops. Wall Street analysts euphemistically called it "renewal pressure." In plain language: customers don't want to pay anymore.
It's not that products got worse. It's that customers suddenly realized they could do it themselves. The catalyst is "Vibe Coding": the explosive maturity of AI-assisted development tools. GitHub Copilot, Cursor, Replit Agent enable non-technical teams to build functional applications in days. Not perfect, but good enough. And "good enough" is deadly for a $3,000 per month SaaS subscription.
A Series E tech company ran an experiment: its engineering team spent less than a week using AI tools to connect GitHub API and Notion API, rebuilding an internal project management system that covered 80% of a legacy enterprise software's core features. Result: they cancelled a subscription costing over $30,000 per year.
This isn't an isolated case. A SaaS customer success manager revealed that churn in Q1 2025 was nearly double expectations, with a new churn category: "customer-built alternative." A decade ago, building a CRM in-house required dozens of engineers, millions of dollars, and at least a year. Today, a product manager plus an AI assistant can produce a prototype in three days.
The Gap Between Works and Works Well
Software has an old rule: building something accounts for 20% of the effort; making it stable accounts for the other 80%. AI can handle the first 20%: functional code, API integration, UI generation. But the remaining 80% — error handling, edge cases, security, scalability, maintainability — requires engineering judgment rooted in deep understanding of real-world business logic.
Companies that cancel subscriptions and switch to in-house builds may soon discover an awkward reality: when things break, nobody fixes them; when requirements change, nobody updates; when security issues arise, nobody is accountable. SaaS companies don't sell code; they sell "someone is accountable when problems happen." This argument is not yet convincing to companies still in their honeymoon period with self-built apps.
Three Survival Paths: From Selling Software to Selling What AI Can't Copy
Facing this crisis, SaaS firms are not without options. But the essence of each path points to the same direction: pivot from "selling software" to "selling what AI cannot replicate."
Path 1: Become the system of record. Salesforce remains hard to replace not because of a great UI — many users complain about it — but because it has become the customer data hub for countless enterprises. A decade of customer data, workflows, and organizational knowledge is embedded inside. You can use AI to write a better CRM front-end, but you cannot move the data or the organizational inertia built around it. When your product becomes the customer's memory, they can't leave.
Path 2: Sell security and compliance. AI-generated code knows nothing about SOC 2 certification, data encryption standards, or audit logs. For highly regulated industries like banking, healthcare, and government, "works" is far from enough; compliance is a hard requirement. A self-built system that fails a compliance audit could trigger a fine of $3 million — far more than the $30,000 saved on subscription fees.
Path 3: Transform from product to platform. Instead of resisting customers' urge to build, embrace it: turn your product from a fixed-function software into a platform that customers can freely extend. Let them build what they want on your foundation using AI. One telling data point: when technical staff only had access to modules relevant to their work, usage jumped from 35% to over 70%. Not because the software improved, but because it finally became "theirs." In this sense, AI is not SaaS's gravedigger; it's the force pushing SaaS to evolve.
After Software Ate the World, AI Eats Software
In 2011, Marc Andreessen wrote in the Wall Street Journal: "Software is eating the world." Fourteen years later, the prophecy came true: from ride-hailing to food delivery, from office work to social media, from finance to healthcare, almost every industry has been reshaped by software. But Andreessen didn't foresee that after software ate the world, AI would start eating software itself.
SaaS's moat was built on the premise that "building software is expensive." AI is shattering that premise. When development costs approach zero, software itself is no longer scarce. What becomes scarce is data, trust, compliance, and organizational knowledge that takes a decade to accumulate. When the moat is flattened by AI, what's left?
The answer varies by company. Some will die because their entire value was just the code layer. Some will transform because their real value lies below the code: in data, in workflows, in the organizational memory of customers. SaaS is not dying; it's undergoing a brutal value reassessment. Survivors won't be the best coders — they'll be the ones who best understand what lies "outside the software." When everyone can write code, the code itself has no value. What has value is everything behind the code.