AI keeps getting better at doing things with each new model release, and lately it has got even more proactive. It does things in one shot, even when I didn’t ask it to. Most of the time what it does is pretty close to what I wanted.
So I’ve been handing over more and more. Whatever it can do, I let it do, and I superficially check the work, and if it looks ok I move on.
Am I getting lazier?
That’s not a rhetorical question. I’ve really found myself asking it. Every time I move that line a little further I have a reason, and it feels like I’m justifying the behavior.
So here are the reasons I use.
It’s rote work, and I would have done it the same way.
I know how to do this already. I could sit down and do it myself if I needed to.
I understand what is happening conceptually. The details don’t matter as long as the high level objectives are met.
Each one of those is true. Each one gives away a little more than the one before it.
There have been moments when the line has moved back the other way.
The AI gives me something that isn’t what I wanted, or is naive in how it’s built, or misses the point entirely. Either way I move the line back and do it again.
And sometimes moving it out is right. The models genuinely got better, so work that sat with me last year belongs on the other side of the line now. Following that isn’t rationalizing, it’s just keeping up.
So the line moves both ways.
Recently I was working on a public github project (gdocs-mcp) that gives an AI the ability to make edits to Google Docs. I asked my Claude Code agent to add the ability to list folders, create them, move things, rename things. It built the functionality, five tools to cover it, and they worked.
There lay the problem. It worked. Nothing was broken, nothing failed, all the tests passed, there was nothing telling me to look deeper. I could have stopped there and called it a day.
But something bothered me. I had just added five tools to an already large toolset.
I thought about this problem and had a hunch that the five could be one. These are just filesystem operations, and it has been trained with plenty of information and commands to use filesystems. It knows ls, it knows find, it knows mkdir and cp and mv, cold, from millions of examples.
So I pushed down that path. The five tools became one that takes shell commands.
The AI knew how to build the five operations so it did, and it isn’t smart enough to reason if there’s a better way (yet). It might be picking from the most common pattern it’s been trained on. It might have all the knowledge, but in some cases we still have to make those intellectual leaps for it.
What I came away with was the knowledge that a model’s own vocabulary is a design surface. I use that when I build things that an AI has to use. This was something I wouldn’t have had if I had just accepted the original solution, if I had ignored that itch because it worked.
Now go back and read those three reasons again.
It’s rote. I know how to do it. I understand it conceptually.
Every one of those questions is about the output. Is the work as good as if I had done it. None of that is about me. So yes, those reasons were helping me rationalize. That is why the line kept moving. The question I was asking myself was the wrong one.
If I had taken the five tools, I would have something that solved the problem and nothing else. Something I can use. Nothing inside me would have changed.
Now I ask a different question before setting that line.
What is it that I want to get out of this?
If it’s just the output I’m after, I hand over more. If it’s something I want to learn from, something where the intricacies interest me or have something to teach, I keep more.
Also I follow those itches, those slight annoyances to see where it leads. Usually that’s my intuition telling me that something’s off, or there’s something to learn here.
So I guess in some ways things are still the same as before we had AI helpers. Time is the currency we use to learn and grow. AI just makes it cheaper, but we still get to choose where to spend it.
The other side of this is Brains, But No Hands, on handing the rote work over.