I want to be clear about something before I start, because the premise of this article could easily be misread: I am not someone who stumbled across AI tools recently and was surprised they existed. I have been using them since well before ChatGPT made the whole category famous. The names of the tools most people haven’t heard of — the early versions, the ones that felt like prototypes, that required patience and a willingness to treat output as raw material rather than finished product — those were part of my daily life for years before the rest of the world caught up.
At this point, I talk to AI more often than I talk to most people I know. That is not an exaggeration and not something I say with embarrassment. It is simply accurate. I show it pictures of things I’m uncertain about — a plant that doesn’t look right, a mark on a wall, a label in a language I don’t read. I use it to experiment with my rye starter, to troubleshoot my lotus flowers through each stage of growth, to work out exactly how to descale a specific model of coffee machine without voiding whatever warranty I pretend I still have.
It is woven into how I move through the day in a way that feels, at this point, completely ordinary.
My primary tool for all of this has been ChatGPT. It knows my history. It knows my context. When you use one tool consistently over a long period, you notice how a shared vocabulary, a sense that you do not have to re-explain yourself every time builds up.
That matters more than most people realize.
So when I decided to spend a month learning Claude properly, I was not doing it as a beginner. I was doing it as someone already deep in the water, deciding to swim in a different direction.
What I was actually giving up
The Netflix part of this deserves more honesty than it usually gets in articles with a similar premise.
When you are studying, working, and trying to maintain something resembling a social life, the windows for anything like passive leisure become very small.
The honest truth is that my Netflix watching had been concentrated almost entirely into a narrow slot: late at night, often past two in the morning, the final twenty minutes before sleep when my brain had run out of productive capacity but hadn’t quite accepted that it was time to stop entirely.
This is not a particularly good way to watch anything. The shows barely registered. I would start an episode with the vague intention of paying attention and end it having absorbed almost nothing. The experience was less enjoyment and more the sensation of having done something — a placeholder for rest that wasn’t really restful, borrowed from the next day’s energy and mood in a trade that rarely felt worth it.
Cancelling it, in other words, was less sacrifice than it sounds. What I mostly felt when I did it was relief. The decision cleared a small but genuine ambiguity out of the end of each day: the question of whether to open the laptop one more time, or to start an episode, or to just read — a question that had been resolved badly too many times — simply went away.
The slot became quieter. I used it to think, or to read, or sometimes to continue a conversation I’d started with Claude earlier that day.
What a different tool actually reveals
The assumption going in was that moving from one AI to another would feel fairly minor. The underlying capability, I thought, would be similar enough that the adjustment would be mostly cosmetic — a different interface, different defaults, a slightly different personality, but nothing that would fundamentally change what I was doing.
That assumption was wrong, and being wrong about it was the interesting part.
The most immediate difference was in how Claude handles extended, iterative work. ChatGPT, in my experience, is very good at discrete tasks — at answering a specific question, generating a specific output, completing a specific request. Claude behaves differently when the task is less defined: when you are working through something that does not yet have a clear shape, when the problem itself is still being formed, when what you need is not a solution but a thinking partner who will push back on your framing and help you arrive at a better question before you try to answer it.
In practice, what this looks like is a tool that asks clarifying questions you hadn’t thought to ask yourself, that surfaces assumptions in your thinking before they become problems in your output, that seems more comfortable saying “I’m not sure this is the right approach” than simply proceeding. That behaviour is easy to find irritating if you are in efficient mode. It is very useful if you are not.
I noticed this most clearly in the context of writing.
I gave Claude a half-formed article brief — a rough idea with three or four plausible directions, none of them quite resolved. With ChatGPT, in this situation, I would have added enough specificity to the prompt to get something usable, and then shaped it from there. With Claude, I deliberately gave it less.
What came back was not a draft. It was a question: who is the reader, and what do you want them to do differently after reading this?
Two sub-questions followed. I had answers to none of them. Working through them took about fifteen minutes. The structure that emerged from that conversation was different from anything I would have arrived at alone, and considerably better than what I had been planning to write.
A second example was smaller but in some ways more telling.
My rye starter had been behaving inconsistently — good activity some days, sluggish on others, no obvious reason. I had already asked ChatGPT about this and received a competent list of variables to check: hydration, temperature, feeding ratio, flour freshness. I had worked through most of them without a clear result. I described the same problem to Claude, but this time I did not ask for a solution. I described what I had tried, what I had ruled out, and what was still confusing me, and asked it to help me think through what I might be missing.
The conversation went in a different direction — into the relationship between the specific flour I was using and fermentation speed, which led to a question about my water source and whether I had tried filtered versus tap, which led to identifying something I had not thought to consider.
That is not a dramatic example. But it illustrates the difference between asking a question and having a conversation.
The gap I hadn’t expected to find
The part that surprised me — and this is coming from someone who had assumed they were already close to the ceiling of what AI use could look like — was how much difference the quality of engagement makes, even at the level of individual conversations.
I had been using AI extensively, but I had been using it efficiently. There is a difference. Efficient use treats the tool as a means to an end: you input something, you extract something, you move on. The goal is to minimise friction and maximise throughput. This is reasonable. It works. But it leaves something on the table.
What the month with Claude did, partly because the tool invited it and partly because I had freed up the mental space that the 2 am habit had been occupying, was push me toward slower, more exploratory conversations. Not every task, not every interaction — but some of them, the ones where I was genuinely uncertain about something or working through an idea I didn’t yet understand fully. In those conversations, the quality of what I got out was dramatically better than anything I had previously produced in the same amount of time.
OpenAI’s own research, published in late 2025 and based on analysis of more than 1.5 million ChatGPT conversations, found that roughly three-quarters of interactions fell into three basic categories: practical guidance, seeking information, and writing help. That figure covers everyone from occasional users to people who open the tool every day. The distribution suggests that most engagement, regardless of frequency, stays close to the surface — getting answers to defined questions rather than working through undefined ones.
I had always thought of myself as being on the other side of that divide. What the month clarified is that the divide has more than two sides. There is casual use, there is regular use, and then there is something that might be called deliberate use — where you are paying conscious attention to what the tool does well, what it does poorly, and how the quality of your inputs shapes the quality of your outputs in real time. That third mode is where the gap is largest, and it is not a gap that experience alone closes. You have to be looking for it.
What this means for people who think they already know
If I am writing this for anyone in particular, it is probably for the people who feel, as I felt, that they have already figured out how to use these tools — that their existing practice is close enough to the ceiling that there is not much more to explore.
The honest message is that the ceiling is higher than it looks from where most of us are standing, including those of us who have been at this for a while.
That is not a comfortable thing to conclude, because it implies ongoing effort rather than a skill you acquire and then have. But it is, I think, accurate. The tools are developing quickly enough that a month of deliberate attention to one of them will reveal things that were not visible before — not because the tool is new to you, but because the attention is.
The Netflix subscription is back now. I am watching it at two in the morning again, somewhat sheepishly, in twenty-minute increments that I barely register. Some habits are genuinely hard to break, and I am not going to pretend otherwise. But the month clarified something that I suspect will be harder to unlearn: that there is a version of this that is better than the version I had, and the difference is mostly a matter of how much of myself I bring to the conversation.
That seems worth knowing.
