If you had asked almost anyone, at almost any point in the last fifty years, what the arrival of artificial intelligence would look like, you would have got some wonderful answers. Robots. Starships that talk. A calm voice that runs the house.

Nobody said: a text box in a browser tab.

And yet here we are. The most significant technology of our lifetimes turned up looking like a customer support widget. I find it a bit of a letdown, honestly, and I say that as someone who works with it all day. No one has ever said, “I just can’t wait to chat with a computer pretending to be a person.” Apologies to everyone who makes chatbots, but it is not the pinnacle of human existence.

So why chat?

Because it was the perfect gateway drug.

A chat window needs no manual. Everybody on earth already knows how to type a message and read a reply. It let the people who built these models put something astonishing in front of hundreds of millions of people without teaching them anything first. You typed a question, something clever came back, and you were hooked. As a way of showing off the technology, it was a stroke of genius.

As a way of getting work done, it is a poor fit, and I think most people feel that within a few weeks.

The problem is integration

Nobody wants to stop what they are doing, open a browser, and sit at a chat window.

Think about what using it for real work involves. You find the file. You copy it in. You explain the background, again, because it does not know what you were doing yesterday. You read the answer, copy it out, paste it where it belongs, and fix the formatting. Then you do it all again for the next thing. The model in the middle of that loop is brilliant. Everything around it is you, acting as a courier between your work and a text box.

The models are capable enough. That stopped being the constraint a while ago. What is missing is the connection between the intelligence and the actual work, and building that connection is the real job now.

What the chat window accidentally teaches us

Spend enough hours in one and you notice something about your own role. You are not doing much of the work any more. You are steering. You say what you want, you look at what comes back, you say “not quite, more like this”, you decide between two options, you approve.

That is the shape of the human contribution from here on: guidance and judgement, and much less of the doing.

Which raises a better question than “how do I write a good prompt”. If judgement is what people are for, how do we make giving it efficient? How do we bring a human into the loop at the moments that need one, without chaining that human to a conversation for three hours while the machine waits on every small answer?

Chat is a terrible tool for this. It is serial. It makes you sit there. It forgets. It buries the four decisions that mattered in forty screens of scrolling, and when you come back next week there is no record of what you decided, only a transcript.

What I am building instead

This is my challenge as someone who builds AI applications, and it is what I am working on now.

I am building a suite of AI-first apps that let you work naturally with your own data, held in the structured form you would expect. A to-do list. A project management suite. A social media content engine. Things that look like the software you already understand, because the information inside them has a proper shape: tasks, projects, posts, dates, owners, statuses. Not a pile of conversations.

The aim of each one is the same. Replace a three-hour chat session with a ten-minute conversation, by voice or by text, where the AI asks you the questions it needs answered, and then gets on with things while you go and do something better.

A few principles are falling out of that as I go:

  • The data lives in a structure, not a transcript. What you decided on Tuesday is a field on a record, not something on page nine of a chat.
  • The AI asks, you answer. It should arrive having done the reading, with the handful of questions only you can settle. That turns prompting on its head, and it is far quicker.
  • Decisions come in batches. Ten minutes of your full attention, rather than three hours of being interrupted.
  • Work carries on after you leave. The conversation is where the work is set off, not where the work happens.
  • Anything that goes out into the world waits for your yes. Less doing should never mean less control.

The future is here. It just needs to come to you.

I do believe the future of humanity has arrived, and I do not think that is too grand a way to put it. But it will reach most people in a quieter form than the headlines suggest. Not as a chatbot they have to go and visit. As ordinary-looking software that happens to be able to think, that meets them where they already are, and gives them their time back instead of taking them away from their work.

If you find yourself spending hours a week in a chat window doing a version of the same thing each time, that is a sign there is a small piece of software waiting to be built. Tell me what it is. I would like to hear about it.