From HAL to Jarvis, part 7 of 7. A history of artificial intelligence, told as a story, for people who will use it. It began as a talk.

Try to be one of the people on whom nothing is lost.

Henry James, The Art of Fiction, 1884

James’s advice to the young novelist is, word for word, the specification for Jarvis.

Jarvis, honestly

Take the three things Jarvis does and grade them.

He talks. Done. Any language, any register, any subject, at the speed of reading. Turing’s test, passed and forgotten.

He acts. Half done. The assistant that runs your computer exists and works for hours. The suit does not, yet. The body is a few years behind the voice.

He knows Tony. That is the gap. Memory that persists. Permission to act. Years of context. And the trust to act on all of it without being asked twice.

The model is here. What is missing is the part where it is allowed to know you.

Not a technical problem

And that part is not a technical problem.

“Allowed to know you” takes three things, none of which is a model.

A memory. One that persists from one day to the next, rather than a chat window that evaporates when you close it.

A permission. To act. A gate, with a human behind it, that says which decisions the machine may take and which it must bring back.

And a description. Of how you work, in writing. Your words. Your rules. Your exceptions. The ten terms you would never let a new colleague get wrong. The three sentences that must not break.

Tony never writes a prompt, because Jarvis has read everything Tony ever wrote. Context is a specification you did not have to type. The rest of us will have to type it, at least once. That typing is the work that era three handed back to us, and it is where Asimov left it: saying what you want, precisely, in writing, so that a literal mind can do it and a human can check it.

What it does to the work

The question nobody asks me in the room and everybody asks me in the corridor: what happens to us?

Here is what I can say honestly. A job is a bundle of tasks. The machine takes tasks. The reading goes. The first draft goes. The reconciliation of two files at eleven at night goes. What stays is the two ends: deciding what is wanted, and judging whether what came back is right.

Those two ends used to be a small part of the day. They are becoming the day.

And here is what I do not know, and neither does anyone who tells you otherwise. We learned judgment by doing the tasks. The junior read the hundred files, and that is how she became the senior who knows which one is wrong. If the machine reads the files, we will have to teach judgment on purpose. No company has finished working that out. It is the most interesting management problem of the next ten years, and it is yours.

The habits have a name

One last thing, and it is an invitation.

In early 2025 a researcher called Andrej Karpathy gave a name to something programmers had started to do: vibe coding. You describe what you want in plain language, the machine writes the software, and you do not read the code. You go with the vibe.

It works. For a weekend. Then the project grows, the machine forgets on Thursday what it decided on Tuesday, and you are back on Monday morning, with every variant you ever had.

What fixed it was not a better prompt. It was the oldest habits of engineering. Write the rules down once, in one place. Agree on the words. Let the machine check what a machine can check. Keep a human at the gates that matter. And every time something breaks, turn the failure into a rule, so that it cannot break the same way twice.

Not one of those habits is about code.

I am writing a book this way. Machines check the style, the structure and the citations, and they attack every chapter before I see it. Two gates stay mine: I approve the plan, and I read the draft. A book is not a test suite. It needed the same discipline anyway. So does a report. So does any piece of work that can be described.

I call it vibe engineering. This series was where the machine comes from. That one is how to work with it.

Vibe coding is the talent. Engineering is the institution. The second one works on everything.

From HAL to Jarvis

Instruction to understanding. Seventy years for the machine to catch up with the stories. The next few years are the stories catching up with us.

You are all Jules Verne now. Write it down.

The machine is here.

The remaining work is ours: saying what we want, in writing, and letting the machine know us.

The series