16 min read

End of SaaS? Useful software is never finished

Sébastien Nobour

Founder of DEVEDANOS

Alternatives

Since late 2024, one claim has kept coming back on social media and in the business press: SaaS is over. SaaS (Software as a Service) is software hosted by the company that publishes it, used online and paid for every month, often per user.

If you run a start-up or an SME, you probably use several of them. You may even sell one. This debate raises a question that concerns you: when you pay for software, what are you really paying for?

SaaS is dead, announced several times since 2024

In December 2024, Satya Nadella, chief executive of Microsoft, was the guest on BG2, the podcast hosted by investors Bill Gurley and Brad Gerstner. According to accounts of the interview, he described business applications as databases with business rules layered on top. Those rules, he predicted, would move into artificial intelligence agents able to work across several databases at once. The very idea of a business application could then collapse. The phrase “SaaS is dead” is sometimes attributed to him, but we have not found it in any account of the interview.

In March 2025, Matthew Berman, who publishes videos and a newsletter about AI under the name Forward Future, took up the phrase himself. With AI, he wrote, he could build any SaaS application in a few hours. It would be less polished than a vendor’s, but made exactly for his needs, and it could evolve with them.

SaaS is dead. […] It's hyper-customized for exactly what I need.

Matthew Berman @MatthewBerman View on X

In August 2025, Elon Musk announced Macrohard, a project of his AI company, xAI. In his view, since a software company like Microsoft does not make hardware, it should be possible to simulate such a company entirely with AI.

In early February 2026, investors joined in. Anthropic, the company that develops the AI Claude, had released plug-ins for Claude Cowork, an agent that automates legal, sales, marketing or data analysis tasks. According to Reuters, investors feared these agents would upend data providers and professional services companies.

On 3 February, shares in Thomson Reuters, which publishes legal databases, lost nearly 18%, their steepest one-day fall. Shares in other software and data companies followed. According to Bloomberg, Jeffrey Favuzza, from the trading desk of the bank Jefferies, dubbed the movement the “SaaSpocalypse”.

The debate also reached French-speaking developers. On 20 February, Thibault Houdon, a Python developer and the creator of Docstring, a platform for learning Python, described how he had rebuilt with Claude Code, in a quarter of an hour, the online forms he was paying €25 a month for.

Un exemple concret qui confirme que de nombreux SaaS vont couler.

A concrete example confirming that many SaaS products are going to sink.

thibault @ThibaultHoudon View on X

Four days later, Cloudflare, a provider of infrastructure services for the web, unveiled vinext, a rewrite of Next.js, a widely used web development framework, on top of another tool, Vite. According to Cloudflare’s article, a single engineer had built it with Claude in under a week, for about $1,100 in AI calls. Cloudflare made clear that vinext was not yet a replacement for Next.js in every application.

In September, Mike Codeur, a French developer who creates content about software development, described how he had replaced a SaaS with a tool he built himself. He rents out villas by the night, and by the hour for photo shoots, and he was managing hourly bookings with Timely, a booking SaaS. Timely’s support had taken four days to answer him. So he built VillaSlot, which brings nightly bookings, hourly slots and payments together in a single tool. He had first described the product: 20 features with their acceptance criteria, and 20 mock-ups. His AI agents then worked on their own through the night, and the first version was delivered the next day. He also tells the story in a video.

Le support technique de Timely a mis 4 jours à me répondre. J'ai donc redéveloppé leur SaaS en 48 heures.

Timely's technical support took 4 days to reply to me. So I rebuilt their SaaS in 48 hours.

Mike Codeur @MikeCodeur View on X

Not everyone drew the same conclusion. The day after Thibault Houdon’s post, Ariel, a developer who has created several SaaS products and is known on social media as Le Dev ULTIME, replied that SaaS products were not going to sink. In his view, most people have neither the time nor the inclination to build their own version of a tool they already pay for, and then maintain it. And once they have settled into their habits with that tool, they keep it. In June, he came back to the subject by sharing a post from shadcn, the creator of the shadcn/ui component library: he too could build his own tools, but he would rather someone else built them, maintained them and billed him for them.

[…] Continuez de dire "le SaaS est mort, tout le monde va coder son propre SaaS".

[…] Keep saying “SaaS is dead, everyone is going to code their own SaaS”.

Le Dev ULTIME @ledevultime View on X

The fact remains that someone still has to develop these tools. When you lack the skills to do it yourself, or the inclination, developers take it on.

These statements show what some people believe or fear. The software companies’ figures tell a different story.

The stock market fall, and the software companies’ revenue

In January 2026, the S&P North American Software index, which tracks software companies listed in the United States and Canada, lost 15%, its steepest monthly fall since October 2008, according to Bloomberg. The drop of 3 February came on top of a month that had already gone badly.

The investors selling these shares were worried about the billing model. A SaaS is most often paid for per user. If an AI agent does the work of several employees, the client company needs fewer seats, and the vendor’s revenue falls. Bloomberg spoke of seat compression, and an analyst quoted by Reuters considered that AI tools were putting per-user pricing to the test. Some commentators claimed that certain vendors had already seen their clients’ seat counts fall. We have found no vendor report that confirms it.

Share prices and the vendors’ revenue did not move in the same direction. In early February, Atlassian, the publisher of the project management software Jira and Confluence, reported quarterly revenue up 23%. Its shares nevertheless lost 6.3% that day. According to the Australian news agency AAP, they had lost 69% over the year.

In the months that followed, the vendors’ revenue kept growing. On 5 August, Thomson Reuters, the company whose shares had fallen furthest on 3 February, reported organic revenue growth of 8%. In its legal division, that growth came mainly from Westlaw, its database, and CoCounsel, its AI assistant for lawyers. The company raised its forecast for the year. On 26 August, Salesforce, one of the leading publishers of CRM, customer relationship management software, reported quarterly revenue of $11.3 billion, up 11%, a figure that includes Informatica, the software company it acquired. The next day, the investor David Sacks judged that the end-of-SaaS narrative was being shredded.

The “AI capex is a bubble” and “SaaS is dead” narratives getting shredded this morning.

David Sacks @DavidSacks View on X

More broadly, Gartner, a research firm specialising in information technology, forecast in February 2026 that worldwide software spending would grow by 14.7% over the year.

Software share prices have partly recovered. According to the financial news site 24/7 Wall St., the iShares exchange-traded fund of US software stocks hit its low on 13 April 2026, then gained 41% up to 12 October. It was then about 5% above its level at the start of the year. On 16 October, software shares fell again. As of today, the markets have not settled the question.

What the debate reveals: useful software is never finished

What Matthew Berman was after was not one more piece of software. It was software made for his needs, and that evolves with them. That requirement is nothing new.

Meir Lehman, a researcher at IBM and later a professor at Imperial College London, spent years studying how large programs in service evolve. In 1980, in an article in the journal Proceedings of the IEEE, he sorted programs into three types.

An S-program does exactly what a complete specification, written in advance, says it should, such as computing the lowest common multiple of two numbers. A P-program solves a real-world problem that can be stated precisely, but only approximately, such as weather forecasting. It is the real world that judges whether the solution is acceptable. An E-program automates a human activity, such as stock management or air traffic control. Lehman wrote that it becomes a part of the world it models: it is embedded in it. The E refers to this program embedded in the real world, not to the embedded systems of electronics. Once in service, it changes the activity it serves, and that activity, as it changes, requires the program to change in turn.

In the same article, Lehman set out his laws of software evolution. The first, which he had formulated as early as 1974, concerns programs in service that reflect an outside reality, such as running a business. Such a program must be adapted continually. Otherwise, it becomes less and less useful to the people who use it.

The reason lies in the activity it serves. Your company changes: your customers, your procedures, the regulations. Software that does not change with it ends up describing a company that no longer exists.

In 2001, seventeen practitioners of iterative methods signed the Manifesto for Agile Software Development. Its first principle reads: “Our highest priority is to satisfy the customer through early and continuous delivery of valuable software.” The second: “Welcome changing requirements, even late in development.” The Manifesto does not describe delivering a finished product. It describes work that goes on for as long as the needs change.

A living application is never finished: once you stop evolving it, all it gets are security patches and updates to the components it runs on, and it is an application that is slowly dying.

SaaS evolves too, but for all its customers at once

A SaaS is not frozen software. Its vendor keeps evolving it. But all its customers use the same version: the vendor develops a single product for everyone.

As long as the way you work resembles that of its other customers, this shared evolution serves you. When it differs, you ask for an adjustment. Sometimes the vendor does not make the adjustments you ask for.

Your teams then adapt to the tool. They keep a spreadsheet on the side, or build themselves a small tool outside the ones the company has chosen. This is called shadow IT. Our article on bespoke and off-the-shelf software describes these signs, and what they cost.

As we read it, shadow IT illustrates Lehman’s law from the users’ side: when the software does not keep up with their needs, they are the ones who adapt.

Should you abandon SaaS?

No. For an activity every company has and that does not set you apart from anyone, such as payroll, off-the-shelf software remains the right choice. In a post on his blog, Martin Fowler, one of the signatories of the Agile Manifesto and Chief Scientist at Thoughtworks, advises buying the software in that case and adapting your way of working. The same article sets out his distinction in detail.

So the useful question is not whether SaaS will disappear. For each activity, it is whether your software should adapt to you, or whether you should adapt to it.

So, is this the end of SaaS?

As of today, in 2026, the figures do not show it. The vendors mentioned here keep growing their revenue, and companies plan to spend more on software. What investors have called into question is something else: per-user billing, and the value of software that AI now makes it possible to copy in a few hours.

Coding a copy is not enough to rebuild it properly, with the quality, maintenance and security a company needs. A home-made tool also needs someone to maintain it, protect it and look after the data it holds. These are the hidden costs of shadow IT, which we described in another article.

For a simple tool that no longer changes, like Thibault Houdon’s forms, the threat is real. With AI, writing code costs little, and verification becomes the bottleneck, said Dave Farley, co-author of the book Continuous Delivery, in his February 2026 interview for the Aviator podcast. That is the experience Thibault Houdon described.

But writing the code has never been the hardest part. The hardest part is understanding what users need, delivering it to them, then starting again when that need changes. Allen Holub, a software architect and consultant, wrote as much on LinkedIn in September 2026: until an AI can hold a conversation with users to draw out their needs, this work remains human.

There's just Agentic coding. Big difference. Coding is a small part of the overall process.

Allen Holub View on LinkedIn

Developers who know their trade well did not wait for AI to put users and their needs at the centre of their work. Lehman wrote it more than forty years ago, the Agile Manifesto twenty-five years ago. The debate on the end of SaaS puts this idea back in front of a much wider audience. The value of software lies in its continuous adaptation to the needs of the people who use it.

With this approach, the question takes on a different meaning. As we read it, software sold as a product that no longer changes can be copied. Software sold as a service, which keeps adapting to its users’ needs, cannot be copied in a quarter of an hour, because what you pay for is the adaptation, and it carries on every month. AI makes this work cheaper to produce. It does not make it unnecessary.

This is exactly how we work at DEVEDANOS. We work with an approach in which users judge each feature. For custom software development, we work to a fixed budget, invoiced at the end of each month for the work done. Your application is usable within a month, with a few essential features: your users start working with it straight away, and how they use it determines which features come next.

After that, we adapt your software continuously as your business, your users and your customers evolve, for as long as you wish. You change direction whenever you need to, with full responsiveness, with no amendment or surcharge.

The budget includes corrective, adaptive, preventive and evolutive maintenance. It also includes the zero-regression guarantee: if an accepted and approved feature starts behaving differently after a later modification, we fix it at no extra cost and we lock the behaviour with an automated test so it cannot come back. The source code, infrastructure, deployment pipeline and documentation are your intellectual property once the first monthly invoice is paid. A trial period of up to 15 working days in the first month: no invoice if you do not want the deliverables.

Further reading

Our article on the predictive and the adaptive approach tells the story of what agile was meant to be. And if you are wondering what AI can build on its own, our article on building an app with AI shows where the tool stops.

If your software tools do not talk to each other, our software integration page describes what we can put in place. If the tool you are missing is at the heart of your business, our custom business application page describes what we build.

Are your teams working around software that no longer keeps up with your business? Shall we talk?

Sources

Read next