For a long time my relationship with AI was one of copy and paste. I asked for something, I got an answer in a window, and then I did the actual work: moving that answer to where it belonged. The mail, the document, the site. The AI gave advice. I was the operator.
That has quietly changed, and most people missed the shift because it arrived as a setting rather than an announcement.
This is part 2 of a series. In the first part I told how I took my sites off WordPress and now run them entirely with Claude. This part is about the principle underneath.
From window to tool
The change has a name that sounds dull: remote connections on by default. Connectors to your mail, your calendar, your documents, your publishing channel, your socials. Not something you switch on case by case, but something that is simply there.
The dullness of that name hides what actually happens. As long as AI sits in a chat window, it stays at a distance. It can tell you what you might do. But the last mile, the actually doing it, stays yours. The moment the assistant sits inside your tools, that distance disappears. Advice and execution collapse into one.
In an earlier piece I wrote about the last mile: the place where value really shows up, past the demo, where things get messy. My point was that whoever stands too far from that last mile cannot judge what works. Default-on connections flip that around. They put the assistant on the last mile, next to you, in the real work.
What it looks like for me
Concretely, because otherwise it stays a promise. My writing, images, distribution, and plumbing now run through one environment.
I write an article and it appears as a file in the right place, ready to go live. I ask for a cover and it gets generated locally. I want to stage a newsletter edition or schedule a post, and that happens through a connector, without logging in somewhere separate.
The difference from a year ago is not that the AI answers more cleverly. It is that the answer no longer gets tossed back over the fence to me. It lands where it belongs.
That feels small until you do it for a week. Then you notice that the hundred little handoffs a day, the copying, the pasting, the switching between tabs, added up to a surprisingly large share of the work. Not the thinking. The hauling.
The price, honestly
I do not want to rush this part, because it is the part that matters. On by default means access by default. An assistant that can execute is an assistant that can reach things. Your files, your contacts, your sent messages.
That is not a reason to skip it. It is a reason to stay awake while you do it. Convenience and access are the same coin. Take the convenience without seeing the access, and you do exactly what I called a kind of psychosis in the first piece of this series: a story in your head that comes loose from what is actually happening.
My rule
So I hold one line, and it is simple enough to keep.
Claude may execute where I see the result. I publish a piece with the preview right next to it. I watch a file change in front of me. But something that leaves without me having seen it, a mail, a message, a post sent quietly, that is where I draw the line. Not because the machine cannot do it. Because then I am no longer checking, and checking is the only thing that keeps distance at a distance.
That rule costs me something. It makes some things slower than they could be. But it keeps me on the last mile, which by my own reasoning is exactly where I should be standing.
The assistant has gone from advisor to operator. That is a gift, as long as I stay the one who looks before anything goes out the door.
In the third part I look at what Claude can now actually do that it could not a year ago, and how I use those capabilities in practice.
This is part 2 of a series on AI as an operator. Read part 1: how I left WordPress behind.