Near the end of the conversation one of the makers puts on his business hat and brings up a favourite case. It is an article from the Harvard Business Review in the 90s, about computers. The question back then: the computers are here, why are we not seeing the productivity gains?
The answer from that time is exactly why this story matters again now.
The filing cabinet had to go
What companies did at first was put the computer on the side. You kept your paper filing cabinet, your pen and your existing process, and next to it stood a machine that did something. That did not unlock the gains. What you had to do was more radical: throw out the filing cabinet. Throw out the paper and the pens. Put the computer at the center, and let every process run through it.
Only when the computer became the heart of the work, instead of a helper on the side, did the gains come loose. Back then that took ten to fifteen years.
The same is happening now with AI
The parallel is uncomfortably familiar. Most companies set AI up today in exactly that way: on the side. An extra tool, a subscription, a chat window next to the existing process. And then they are surprised the big leap does not come.
The companies that do get it, the maker says, do it the other way around. They put Claude at the center of everything. At Anthropic itself you can see it clearly. People who join do not ask their questions to colleagues, but to Claude. A question? Claude. A code review? Claude. A security review? Claude. It does not sit on the side of the work, it sits in the middle.
And this time it goes faster. With computers the shift took ten to fifteen years. Now much of the work is already digital, and the AI can operate a computer itself, write code and run that code. That compresses the transition.
The question I ask
This is where my work and this story meet. The question organisations ask themselves is usually: which AI tool do we buy? That is the wrong question. The better question is: do we dare to throw out our filing cabinet? Do we dare not to supplement the process, but rebuild it with AI at its heart?
I do not want to make it prettier than it is. Throwing out the filing cabinet is easily said. It is a renovation, not a button, and not every company should do it at this pace. A lot is at stake: trust, safety, people who have to come along. But the direction is clear, and the companies that dare will not be a step ahead. They will be an era ahead.
That closes this series. It began small, with a button in my editor that I no longer switch on, and ends here, with the question of how you build a whole organisation around it. The thread is always the same: AI may do the work, as long as you decide where it gets to stand at the center.
Part 7 and the close of a series on how the makers of Claude Code work with it themselves. Start at part 1: I dropped Plan mode.