Trending Topics
Trending
Vibe coding won’t kill SaaS… but it will kill weak SaaS
This article is part of our Opinions section, where we invite industry professionals to share their views on the most pressing technology questions of our time.
A decade ago, I watched the blockchain frenzy unfold with a mix of awe and real discomfort. The pitch was irresistible: decentralisation would solve everything from vehicle registries to water management. But as many of us learned the hard way that technology doesn’t fail because the code is weak. It fails because replacing what people already rely on is hard, and because trust, once lost, doesn’t come back easily.
Today, I see the same pattern with AI. On one side, rampant optimism. On the other hand, we’re making the same mistake: underestimating how much adoption, regulation, and human accountability actually matter. The current flashpoint is “vibe coding” – named Collins Dictionary’s Word of the Year for 2025, and now at the centre of what some investors are calling the “SaaSpocalypse.” The argument goes: if anyone can prompt software into existence, the era of paying for Software as a Service is over.
My read: SaaS isn’t ending. It’s just ceasing to be what it has been for the last ten years. AI isn’t killing SaaS. It is exposing which products were never truly essential in the first place.
The illusion of “done”
The excitement around tools like Cursor, Claude Code, and Lovable is real. By early 2026, 92% of US developers report using AI coding tools daily. The barrier to building software has dropped lower than at any point in the last two decades. When a manager who has never touched a code editor can prompt their way to a working dashboard, the traditional software vendor can start to look redundant.
Companies are already building internal tools, data scripts, and simple workflows in a fraction of the time it used to take. But a prototype and a production-ready system are not the same thing. Vibe-coded output skips the parts that actually make software trustworthy: peer review, edge case testing, and any plan for what happens six months later when something breaks.
It looks finished, but it isn’t. A CodeRabbit analysis of 470 pull requests found that AI-generated code contains 1.7 times more major issues and 2.74 times higher security vulnerability rates than human-written code. A real production breach in early 2026 exposed 1.5 million API keys and 35,000 user email addresses from a vibe-coded app whose owner had not written a single line manually.
This may shift: Anthropic’s Claude Mythos Preview, launched in April 2026, demonstrated autonomous zero-day vulnerability discovery across every major OS and browser – not by design, but as a byproduct of general improvements in reasoning and code. AI is creating the security problem and building the tools to fix it. The transitional period, however, is now.
There’s also a subtler problem. A July 2025 METR randomised controlled trial found that experienced open-source developers were actually 19% slower when using AI coding tools, despite predicting beforehand they would be 24% faster (according to Wikipedia).
The productivity gains from AI are not evenly distributed, and for seasoned engineers, the overhead of directing and reviewing AI output can outweigh the speed. The developers who benefit most are those who can hold a clear picture of the system while AI handles the implementation. The difference between a good engineer and a great one hasn’t gone away; it’s only harder to see.
What’s actually at risk
If your SaaS is primarily a polished UI on top of a database, the clock is ticking. Simple workflow tools, generic dashboards, lightweight automation, basic CRM clones: if a well-briefed AI agent can replicate your core value in an afternoon, companies will stop paying for the licence. Enterprise SaaS platforms like Salesforce, ServiceNow and Workday – which support mission-critical operations, enterprise security, compliance requirements, and integrations across dozens of systems – are far harder to replace even if AI can replicate some functionality. The long tail of thin products between those two poles is where consolidation will be brutal.
The customer mindset is shifting too, and faster than most vendors want to admit. For years, SaaS was sold as a tool to help people work. That understanding is coming apart. Customers no longer want a tool; they want a result. If a system doesn’t demonstrably shorten a process, cut costs, or move a metric, the conversation about renewing the contract becomes very short.
The products that will survive are not the ones that bolt AI onto an existing interface. They are the ones that use AI to actually execute work, not just organise it. There is a meaningful difference between a platform that helps a sales team manage their pipeline and one that moves the pipeline forward on their behalf. The first is a dashboard. The second is an executor. The market is now actively sorting them.
Why SaaS still matters
AI can generate a solution. It cannot carry responsibility for one. In any serious enterprise environment, “who is accountable” is still a human question, and it will remain one for a long time. AI has no memory of the political decision made three years ago, no feel for the ethical sensitivities of a specific industry, and no skin in the game when something goes wrong. Someone still has to sign off, and that person needs a system they can stand behind.
Regulated industries make this even more concrete. You cannot prompt your way to a SOC2 certification, a clean compliance audit, or a working integration into a legacy system built before most of today’s developers started their careers. Highly regulated industries face particularly acute governance questions around vibe-coded tools, like data leakage, ransomware exposure, and the question of who owns an application when its creator leaves the company.
Companies pay for enterprise SaaS because they need audit trails, data sovereignty, and the kind of robustness that only comes from running in production for years. Vibe coding is fast at the start and expensive later.
The new moat
As the cost of writing code approaches zero, what actually becomes scarce is context. In Y Combinator’s Winter 2025 batch, 25% of startups had codebases that were 95% or more AI-generated, but those startups then brought in engineers to rebuild the critical components properly. The AI got them to market. It didn’t give them the moat. A vendor that has spent ten years accumulating and structuring data on a specific business process has something no prompt can replicate. The value isn’t in having AI. It’s in what that AI is operating on, and how deeply embedded you are in the workflows that actually run the business.
The other thing that becomes genuinely hard is integration. Modern enterprises run on hundreds of tools, and the problem is rarely a missing feature. It’s that nothing talks to anything else cleanly. The SaaS products that will thrive are the ones that act as connective tissue across a fragmented stack, turning data into coordinated action. Not a function. An operating layer.
The pricing reckoning
Per-seat pricing makes sense when humans are doing the work. When AI agents are doing it, the model breaks, and this is now playing out in real time. Seat-based pricing has dropped from 21% to 15% of SaaS companies in just twelve months. Gartner projects that 40% of enterprise SaaS contracts will include outcome-based components by 2026. Intercom already charges $0.99 per AI-resolved ticket. Zendesk charges per automated resolution. Salesforce prices its AI agents on completed actions, not human seats.
2025 saw more than 1,800 pricing changes across the top 500 SaaS and AI companies, which is an average of 3.6 changes per company. The dominant emerging structure is hybrid: a base platform fee for predictability, with variable consumption or outcome components layered on top. That reframes the vendor relationship entirely. Either the software produces a measurable result, or it doesn’t survive the next budget review. For vendors who built their businesses on inertia and switching costs, this is the more uncomfortable part of the AI story.
Related articles
About The Author
