Where the value goes when building is free.

Over recent months of focusing, building and scaling AI products, we kept seeing the same UX pattern emerge: experiences converging towards a chat, a memory, and the views, artefacts and canvases they produce. So we started playing with a thought experiment that drives the pattern to its limit.

What if every digital experience converges to a chat, a memory and a view?

Every app, website and business tool gets built the same way. You describe what you want to an AI that remembers everything: every file, every database, every decision made along the way. It takes anything as input (text, voice, documents, connections to your other tools) and produces anything as output. The simple things ChatGPT, Gemini and Claude already manage: reports, presentations, spreadsheets, images, videos. And the things they don't, yet: fully working apps, long-lived databases, edge functions, servers, the kind of thing a business runs on for years.

If that is how everything gets built, building software stops being the expensive part. Software was expensive because of people: developers, designers, project managers, and years of maintenance after launch. In the new world that work becomes a conversation and a few dollars of compute. Anything you build, a competitor can rebuild in days.

So where does the value go?

Nowhere new.

The things that hold value in that world are the things that have held value in business for a hundred years. What is new is that the build is no longer one of them. For thirty years, “we can build software others can't easily copy” was itself a durable advantage. That one is gone. And when one advantage disappears, the remaining ones carry all the weight.

We count five.

Memory. What you know that others don't. Not general knowledge: accumulated private history. An insurance company's value is not its forms, it is decades of claims data telling it how to price risk. The same principle applies now. Imagine a product that has handled a company's payment disputes for two years and holds the record of which arguments succeed with which retailer. A competitor can copy the software in a week. They cannot copy two years of recorded outcomes. When machines can produce unlimited output, knowing what correct looks like, because you recorded the results, becomes the scarce thing.

Objects. The physical things AI can't make: inventory, warehouses, machines, licensed infrastructure. AI writes software; it does not deliver a package or build a data centre (yet). Whatever cannot be reduced to information keeps its old economics.

Audience. When anyone can make anything, the question stops being who can build it and becomes who people go to. A brand people recognise, a person they trust, a name they choose over the dozen alternatives that now exist. The product can be copied. The reason people pick you cannot.

Tribe. Different from audience: an audience chooses you, a tribe can't leave you. Every member is there because everyone else is, and each new one makes it more valuable for the rest. You can't leave WhatsApp, not because it's the best messenger, but because everyone you know is on it. Copying the software doesn't move the crowd.

Signature. The AI platforms are explicit that they are not responsible when the output is wrong. So putting your name on the result becomes something people pay for: contracts, licences, guarantees, a professional who signs the work and answers the phone when it breaks. Auditors, insurers and banks have always earned their living this way. They still will.