Vibe coding versus SaaS: the other side of the conflict

By David Hurtado · June 23, 2026.

The article starts from an image circulating on LinkedIn that argues that AI-generated software (what has become popularized as 'vibe coding') can never reach real production, because to serve hundreds or thousands of users you need a complete technical stack—with layers of security, data management, legal compliance, scalability and team deployment—that someone who programs by prompting an AI simply does not consider. David Hurtado accepts that premise as true, but points out that it is only one of the two sides of the problem.

The second side is the one the author develops at length: there is an enormous market for single-user software—applications for a freelancer, a three-person team, or the creator themselves—where that complete stack is entirely unnecessary. The stack layers serve to solve three problems that a solo user does not have: scale (many people at once), shared data (separating one user's data from another's) and obligations to third parties (clients, contracts, regulators). Without those three problems, almost none of the additional layers are necessary.

The author goes layer by layer through what stays and what disappears in the single-user scenario. What remains is the application itself—screen, logic, functionality—and a place to store the data with backup, although in much simpler versions. What almost disappears includes: role-based permissions (they exist to govern many users with different access levels; with a single owner-user, a simple login or none at all suffices), data isolation between users (without a second user there is nothing to separate), request limits (no one collapses themselves), caching and content distribution networks (they solve slowness in the face of global traffic; a single user does not generate that traffic), legal compliance and service agreements (they are promises to third parties; without third parties, there is no one to promise them to) and all the team-deployment machinery (it exists so that several people can touch the same code without stepping on each other; if you program alone and in spurts, it is superfluous).

As for the economic aspect, the article notes that this is the part that hurts the most. For years, the individual user has been paying monthly subscriptions that include dozens of features designed so that the product can serve any type of client. Someone who only needs the tool for themselves pays for an enormous capacity they never use. In the author's words: 'single-user SaaS was subsidizing the rest.'

At this point evidence of real impact is introduced. In December 2024, a remark by Satya Nadella on a podcast about the collapse of SaaS went viral. In February 2026, the financial press spoke of the 'SaaSpocalypse': the software sector lost approximately $285 billion in market value. What is striking, according to the author, is that the cause was not that AI failed, but that it worked well: clients began to remove user licenses instead of adding them, directly attacking the per-user charging model that sustains traditional SaaS.

Another data point pointing in the same direction: 51% of the software licenses that companies buy are not used, an all-time record. That is, half of what is paid for is touched by no one. At the concrete-case level, the company Blinkist replaced about $60,000 a year in SaaS with applications built with vibe coding tools. From there emerges what the author calls the '90/10 rule': if a tool covers 90% of what you need and charges you no license fee, the 10% that is missing does not justify a recurring payment.

The article underscores a structural advantage of those who program their own tool: they do not compete on features, they win by removing. SaaS has to serve all the use cases of all its clients. The vibe coder only has to serve their own. That is why 'the 90% I need' beats 'the 200% they sell me.'

As for the profile of those who practice vibe coding, 63% declare themselves non-developers: product, marketing, founders, designers. The author illustrates this with a video by Dot CSV where, testing the Fable 5 tool, the creator builds his own version of Notion with, in his own words, 'the minimum features he needs.' He also mentions that in the latest batch of Y Combinator startups, one in four had 95% of the code generated by AI, and that Gartner calculates that by 2028, 40% of new enterprise software will be created with vibe coding techniques.

However, the article qualifies the most optimistic version: many people do not know what they need until they see it. And there, packaged SaaS, where someone already did that discovery and design work for you, still has value.

On when the complete stack still wins, the author is clear: the boundary is not enterprise versus individual, nor in size, nor in money. It appears exactly when more than one person needs to see or touch the same data. At that moment, roles, permissions, data isolation and almost the entire stack are necessary. And that cost does not rise gradually: it makes an abrupt jump when going from one to two users.

Finally, the author notes that there is one element that never disappears entirely even in the single-user scenario: storing the data well, making backups and maintaining a minimum of security hygiene (not leaving passwords exposed, not opening the application to the internet unnecessarily). Although he points out that this could be easily resolved outside the app itself, following concepts such as 'file-over-app' and 'local-first software.' The article ends inconclusively on the point of the security of AI-generated code, a topic the author appears to address in other installments.