The Linuxification of the economy
11 September 2026
There is an idea that software as a service (SaaS) companies are in trouble because in the future everyone will build their own custom tools exactly the way they want and nobody will need these subscriptions any more.
This is unlikely because companies want to focus on building their own products, not managing fifty different internal tools to replace all manner of SaaS subscriptions they currently use. Writing software is only one part of the equation. An important part, no doubt, but there are other time consuming aspects of developing and maintaining an internal tool which people do not want to do. Time is the critical factor and giving a lot of it away to save some monthly costs is usually a very poor exchange.
It does not mean that SaaS companies in their current form are necessarily safe. There will be more transformational changes driven by artificial intelligence in this area. It does smooth away certain frictions with writing code, reducing the time it takes to build useful tools. And it is more plausible to build alternatives when the underlying activities are roughly the same across different organisations so the maintenance burden for the software can be shared.
Coalescing effort around one core piece of open source software per activity makes more sense than everyone building slightly different versions of the same basic system. Once the foundational version of the software is established, variations can be built on top of that according to people’s preferences. Similar to how the Linux kernel and surrounding ecosystem works.
The example of Git
Linux itself has an example of users rising up and building their own open source version of a tool they needed – Git. Before it was created, Linux developers used BitKeeper – not a name people are familiar with any more because the product is functionally extinct. BitKeeper was a proprietary version control system, allowing software developers working collaboratively on a single codebase to make updates in an orderly way. It was widely used by Linux developers although there was significant tension within the project about using a proprietary tool, albeit one provided to open source projects for free. Eventually, the simmering disagreement erupted over restrictions in the free version of BitKeeper, leading Linus Torvalds to invent Git.
Git rapidly took over as the version control system developers used far beyond the Linux kernel and remains so until this day. It is open source, anyone can install it, run it and modify it for free with no restrictions on functionality. Successful businesses like GitHub and GitLab have grown up around it. Although BitKeeper was not SaaS, the transition from BitKeeper to Git is a good conceptual model for how to think about how SaaS will evolve in the age of artificial intelligence. And Git has demonstrated that an open source core is a viable foundation for a sound business.
Many SaaS businesses are built around routine, regular tasks such as payroll, accounting, payments, customer support, task management as well as a variety of other activities that are more closely associated with specific types of organisation, for example churches organising serving rotas, giving or small groups. All of these activities need predictability and reliability, a stable core of software that doesn’t change much once it is settled.
It is the predictable nature of these activities that allows SaaS companies to build products around them, confident of durable demand. It also means using nondeterministic artificial intelligence extensively within these products is not a desirable trait. At the moment, such products are built entirely by the SaaS company’s internal team. The code is closed source because the users generally want a service that just works, they do not care whether the code is open or closed – unlike the Linux developers for whom restrictions on a proprietary tool caused a practical problem as well as a philosophical one.
Widening the pool
The emergence of artificial intelligence changes two relevant factors. One is that it reduces the time it would take to build an open source alternative. Most users of SaaS products do not have the unique combination of skills that Linus Torvalds has, enabling him to understand both conceptually and practically what the new product required as well as the software engineering skills to execute the initial version of Git. It is not that artificial intelligence tools can turn everyone into Torvalds – far from it – but that it vastly widens the pool of products susceptible to this type of change because it makes it much easier for users with a particular need and skilled software engineers to work together and create a robust production system. The other factor is that the combination of artificial intelligence and open source offers something that users do care about – the opportunity to tailor the software to their exact preferences rather than being stuck with a rigid, closed source option.
SaaS will continue to be a viable model but instead of offering a single version of a product that everyone has to use, providers will be able to offer multiple versions tailored to the exact needs of individual organisations built around the same basic kernel of open source software – the Linuxification of the economy. This is the conceptual equivalent of factory managers in the early twentieth century seeing that electrification meant that they could use a variety of miniaturised motors each suited to the specific task in the factory, rather than make everyone dependent on a single source of power.
Instead of these big, closed source monoliths we would have a set of open source kernels for different activities that multiple SaaS companies could build their businesses around. This would offer a level of responsiveness to users previously impossible. Not only could different companies offer their own bespoke versions of the software but each individual user could have a tailored version that exactly fit their own needs.
In the past, the amount of development time required would have made this approach impossible but with artificial intelligence writing the code, this shift is now conceivable. However, it does not mean every individual nontechnical user will be chatting away with an LLM and pushing changes to the software. A controlled process would be needed to interpret the requests and ensure the code being written is good quality, in line with the principles and standards set by the open source maintainers but this is also within reach. We are working on a new, open source artificial intelligence system that offers such control and we intend to release it in the coming months.
A humane vision for the future
When we discuss the impact of artificial intelligence on people, this is a more humane vision for the future than the prospect of everyone engaging directly with LLMs on a regular basis which seems to be the vision of some in the technology industry. Instead of pushing further into daily life, this approach of using artificial intelligence to create better tools for people would allow it to recede into the background – its proper place.
People would no longer be wasting their time wrestling with systems that don’t quite fit. Software would align more closely with their needs, reducing the amount of time in front of a screen to accomplish the same task. Machines would operate quietly in the background of life, always dependent on people for instruction, never galloping uncontrollably away. Ultimately giving us greater agency over the tools we use, not taking it away.