Accidental software

At my house, software runs that you can't buy anywhere. A tool and AI assistant that helps us plan the household, built around how our family works. Among techie friends I see variations on the same theme: one has his mailbox triaged by rules he set himself, another has a dashboard showing exactly what he wants to see every morning, nothing more. On YouTube I see creators building their own editing and motion tools, because no package does precisely what they want.

And then there's a much larger group: people making software without realizing it. Ask Claude to compare a few numbers and you sometimes get back not text but a small working program: an interactive overview you can click through and filter. The question was just a question. The answer was software.

That's the pattern I'm starting to see everywhere. People are writing software. No project was started, no developer hired, no license signed. The software emerged as a by-product of a conversation.

Code that doesn't want to be a product

How is that possible? Because generating code costs next to nothing anymore, and because the tools are built around it. Claude has artifacts, a panel next to the conversation where generated software runs instantly. OpenAI has its own variant in Canvas. Behind the scenes it happens even more often: AI assistants constantly write code to calculate something, visualize something or transform data. Code you never get to see.

That code has no ambition to be a product. It doesn't want to be sold, doesn't want to scale, doesn't want to be maintained. It serves one question and may disappear afterwards.

By now it goes beyond throwaway code. Look at Claude Design: you ask a question about typography and you don't get a generic settings panel, but an interface with exactly the controls your question needs. The software doesn't adapt to the user, it comes into existence at the moment of the question. I suspect we'll find this principle in more and more software over the coming years.

The idea isn't new. Robin Sloan wrote about apps as home-cooked meals back in 2020, Maggie Appleton about barefoot developers. What stands out in their examples: they were almost always about data you don't want leaving the house. Your kids, your health, your money. That's where hyperpersonal software begins, with what's too personal for an off-the-shelf product. Back then it was a vision for developers. Today I see it happen weekly with people who can't read a line of code.

Where is this going?

Software is already becoming an everyday object more often than a product. Like a spreadsheet: nobody calls themselves a software builder for putting together an Excel. Soon, generating a tool of your own will feel just as ordinary.

Meanwhile I see my LinkedIn feed filling up with people turning their idea into a product. Understandable, building has never been this accessible. But when everyone can build, the difference is no longer in the building. Then distribution, trust, maintenance and above all the right data become the scarce things.

That splits the software market into two layers. At the edges everything becomes fleeting and hyperpersonal: interfaces, small tools, reports, generated the moment someone needs them. At the core sits what stays durable: data, data models, integrations, and the knowledge of how your organization really works. SaaS that is essentially an interface on a database will struggle against what users will soon generate themselves. SaaS that delivers real infrastructure or a network will hold.

For organizations the lesson is the same as in my previous piece on the renaissance of custom software. Invest in the foundation: one place where your data comes together, properly structured, and open it up in a smart way so AI can work with it. The tools on top of it, your people will soon generate themselves, tailored to their job, their team, their day.

There is a danger in that too. A small tool for yourself does no harm. An organization that gradually comes to lean on dozens of hyperpersonalized pieces of software without a clear framework is building a new patchwork. Generated instead of duct-taped together, but just as shaky. Someone has to keep designing the whole: shaping the foundation and setting the boundaries within which all that freedom is safe. The craft of developers, engineers and agencies doesn't disappear with this, it shifts. Less typing code yourself, more directing AI systems: orchestration, agent engineering, the architecture that keeps all that generated work reliable.

Because people and their agents writing software by accident can no longer be stopped. The only choice you have left is whether they do it on top of your foundation or next to it.

And it doesn't stop at software. Every digital profession where the problem can be explained in language has the same shift coming: architects drawing plans, engineers and carpenters making technical designs, 3D modelling, PCB design. Ask Claude today and you already get a first draft. But that's for a next post.