Owned, not rented
Why the software a company runs on may once again be something it owns
Most companies run on rented software. A monthly fee buys a finished product, someone else keeps it running, and for everything that works the same way in every company, that is a good deal. It starts to pinch in two places: when a company needs something the product was never built to do, and when it wants to do more with its own data than the product allows.
Both have long been accepted as part of doing business. The alternative – software built around the company itself – has been out of reach for most of them.
Building good software has been expensive in every direction. Good developers exist, but finding and keeping them is hard. One good developer is rarely enough; something above the ordinary takes a team, and a team costs a lot of money every year. Writing the code itself takes time. Technology companies are set up for exactly this – software is what they make. For most other companies, small, mid-sized and many large ones, it has never been part of the business.
And money is only one side of the gap. Even where a development team exists, the people who understand the work and the people who build the software often do not understand each other. A finance department knows what it needs and struggles to put it in terms a developer can build from. The developer hears a request and builds what he understands. Much of what matters falls into the gulf between the two.
This is where AI changes more than it first appears to. It makes building cheaper, so a task that is still done by hand because it is too specific for any product becomes worth automating. Software built for one company also carries far less weight than software built for a whole market of customers: no endless configuration, no edge cases of other businesses, only what this one company does.
It also changes the conversation. A model does not only stay patient. It stays with what is being described – it does not drift to its own next point, as people in a conversation often do. People describe things in very different ways, and a model adapts to the way each of them talks. It can take what someone in finance says and turn it into something a developer can build from, whether the obstacle is a language barrier or simply two different ways of talking about the same thing. The earlier note Two sides of one coin looks at the layer of context that makes this work.
None of this is an argument against software as a service. Where a use case benefits from standardisation, it is one of the best ideas in business software: functionality that would otherwise take a custom build, for the price of a subscription. Workflow tools push the model further and let companies shape parts of it to their own way of working.
Open source pushes it further still. More and more companies publish the software they sell as a service as open source, often under permissive licences such as MIT or Apache 2.0. That code can be changed at its core – the business logic, not only the interface. It becomes a building block. This ecosystem of building blocks is a large part of why AI coding works as well as it does. AI rarely builds anything from the ground up; it recombines parts that already exist and have proven themselves. Seen soberly, it is a very fast bricklayer.
Configuring, though, is not building. Sooner or later a company adapts its work to the software rather than the other way around. And there is a quieter cost. In a hosted product the data lives with the vendor, and getting all of it out tends to be expensive, not on offer, or simply painful. So companies stay – with systems that are slow, outdated and unintuitive to use – because leaving would mean leaving behind what is arguably their most valuable asset today. Software running on a company's own servers at least keeps that data within reach, however old the application around it is.
Cheap building changes the calculation. Software a company owns keeps the data in a form the company controls, and it moves in whatever direction the business wants to move. Software a company could buy and own has always existed. What is new is that building one – shaped to this company and nobody else – comes within reach of companies that could never have afforded it. Such an asset adds to what a company is worth: through the software itself, and through the way the company runs because of it.
It also compounds. Modern software the company owns becomes a building block for the next thing, and that for the thing after. Vendor software rarely works that way. Using it as a foundation depends on it having been built the right way, on the vendor opening the right doors, and on incentives that point in the same direction – too many moving parts.
Most of that owned software does not have to be AI, and much of it should not be. What runs a business is largely ordinary, deterministic logic: an invoice that goes to the right customer on the right date, records that stay consistent. Even that is often not automated reliably today. AI is the accelerator that gets it built, not the thing that runs it.
A useful picture is the human body. The deterministic software is the bones: fixed, structured, doing one job well. AI is the joints – the part that bends where rigid logic cannot. A document or a screenshot goes in, and its content comes out structured, something that until recently needed a large pipeline to do reliably. A complex system gets wrapped into one component that the rest can talk to. A step in a workflow writes the small piece of code it needs. A body made only of joints could not stand upright, and software made only of AI would be too unreliable to run a business on. With the right gates around it, it can be trusted like any other program.
That still leaves questions without full answers. Independence has a price: software a company owns needs someone to look after it once it is built. For mid-sized and large companies with their own IT departments, that is manageable. For smaller ones, two answers are emerging. Third parties now maintain software at low cost. And a company no longer needs a permanent development team: it can bring in very good developers for a specific project – something that used to be too expensive and is becoming affordable. What happens when nobody takes that role is the subject of an earlier note, Unattended.
Security is the other open question. Building has become so easy that a lot of software is now written by people building software for the first time, and much of it ships without anyone reviewing it. The tools that catch what goes wrong are catching up, but they are not there yet.
Neither question makes the shift less worth pursuing. Cheaper building reaches the work nobody could ever justify automating – the tedious, manual steps that fill a working day. That is often where a company gains the most: people get that time back for the work that actually needs them.