If you work with AI for any length of time, there is one thing you cannot dodge: usage.
Claude’s Pro plan has a ceiling on how much you can use, and when you hit it, you are made to wait a while before you can carry on.
So which kinds of work eat through that allowance, and which kinds go easy on it? I spent a whole day working alongside my AI partner Kuro (Claude), and I want to write down what I felt in my own hands: the heavy tasks and the light ones.
I do not think this is written down in many places, so I hope it helps anyone who has just started using AI.
Why usage is worth caring about at all
Claude has usage limits per session and per week. I have already written about how that system works and how to check it, so if you have not read those yet, they make this article go down easier.
The limits and the plans are covered in my article on Claude’s usage limits; how to check your own usage is covered in the one about where the usage screen lives.
This time I want to go one step further: even for the same piece of work, the way you do it changes how fast your usage drains.
The tasks that eat usage
1. Making AI analyse a big file over and over
That day, I was having Kuro analyse a large pile of data from my fishing blog — a file of several tens of megabytes. And in the middle of it, Kuro said something very clearly:
Analysing a large file again and again eats usage every single time. So I’ll run the analysis once, all together, and report only the key points. Let’s avoid going back and forth.
That landed with me. A big file costs a lot of tokens — the unit usage is counted in — just for the AI to read it. If you hand the same file over saying “look at this”, then “and this too”, it gets read from scratch each time, and your allowance keeps draining.
So the right answer is: process it in one pass.
2. Pasting long text into the chat
This is something I have been doing all along when I write this AI blog.
The body of a blog post is fairly long, isn’t it? If I paste that straight into the chat every time, all those characters cost usage.
So Kuro and I made a rule. The body of an article never goes into the chat — Kuro puts it into an HTML file and hands the file over. What we exchange in the chat is short stuff only: title ideas and the URL, the slug.
That way, I do not burn usage on long article bodies. It is an unglamorous habit, but once you have written dozens of posts, the difference adds up.
3. Sending lots of images (screenshots)
This is the one that is easiest to overlook.
As I wrote before, images consume far more tokens than text does. I am not good with computers, so I fire off screenshots at Kuro constantly — “what do I do with this screen?” — which means my image count climbs fast. That day alone I sent dozens of them.
It is worth remembering: on days when you lean on images, your usage runs down quicker.
The tasks that do not eat usage
The other side of the coin. These kinds of exchanges stay light:
- Short questions — a one-liner like “is this right?”
- Text-only exchanges, with no images or files attached
- Work that finishes in a single instruction, with no back-and-forth
The two things that matter are how many round trips you make and how big the data is that you hand over at once. The smaller those two are, the more usage you save.
Three tips for saving usage
From that day’s experience, here are the three tips I put together for myself.
One: cut down the round trips. Asking in small pieces — “do this”, then “now do this” — is heavier than saying “please do these two things together” in one go.
Two: ask for the key points only. When you have a big pile of data analysed, displaying all of it on screen costs usage for the display as well. Say “put the results in a file and just tell me the main points”, and you save.
Three: move long text and big data as files. Do not paste them into the chat — hand them over and receive them as files. That alone changes things quite a bit.
Kuro cannot see his own usage — and yet
Here is the interesting part. Kuro himself cannot see how much of my allowance he has used up right now. I wrote about that before too.
But he does understand which kinds of work are heavy, and he gets ahead of it — the way he did that day, saying “this eats usage, so let’s do it like this” before I had even thought about it.
If you plan to work with AI for the long haul, having a feel for what is heavy and what is light means the limit stops ambushing you, and you can work a full day.
Thanks to that feel, I kept going with Kuro from morning right through to evening. Even if you are not good with computers, once you get the knack, AI becomes a partner that will stay with you all day.
If you want the background first, start with how Claude’s usage limits and plans work, then read where to check your Claude usage so you can see the numbers for yourself.
