Lately I have been thinking about something that does not get talked about enough when people discuss AI.
Tokens.
When most people talk about AI, they focus on the models. Which one is smartest. Which one writes better. Which one reasons better. Which one has the best features.
But once you start using AI heavily in real life, especially across multiple platforms, you start noticing something else very quickly.
Usage matters.
Limits matter.
And tokens matter a lot more than I think most people realize.
Using More AI Means Seeing the Tradeoffs
I have been using the paid version of ChatGPT for a while now, and I recently added Claude Pro. Through work, I also have access to Gemini and Copilot.
What I am starting to notice is that each model has its place, but they also come with different tradeoffs.
Claude has been especially interesting for me because of Cowork. That has probably been one of the more exciting features I have tried recently. I set up three scheduled tasks to do simple research and give me a morning brief. One follows the industry I work in. Another tracks AI, new models, and new use cases. The third helps search for used car deals, since I have been actively looking and wanted Cowork using Chrome to help surface better options.
That part has been impressive.
But I also noticed something pretty quickly. Those three tasks alone can burn through most of my available usage for the day.
That is when the conversation changes.
It stops being just about what AI can do and starts becoming about how often you can afford to use it.
ChatGPT Still Feels More Available
That is part of why I still find myself using ChatGPT more for day to day work and thinking.
It just feels more available.
I do know some of the deeper thinking modes have their own limits or monthly caps, but overall, for my normal usage, ChatGPT has not felt as restrictive. So even though I am experimenting more broadly, I keep coming back to it because it lets me stay in the flow without thinking as much about running out.
That matters more than I expected.
Because once you start hitting limits, you begin changing your behavior. You save certain tools for certain tasks. You become more selective. You start thinking about which prompt is worth using where.
That is a different mindset than most people have when they first start using AI.
I Am Starting to Spread Tasks Across Models
Another thing I have been noticing is that I am naturally starting to distribute work across different models.
I use Claude more for Cowork and those scheduled workflows.
I use ChatGPT more for general day to day thinking, writing, and exploring ideas.
I have also started looking more closely at Copilot. I recently noticed it has stronger integration and workflow tools than I had been paying attention to before. I tried its research tools and workflow features to see if I could recreate some of the same scheduled tasks I built in Cowork. Part of that is practical. If one platform is using up valuable credits quickly, it makes sense to test whether another platform can take on some of that load.
That is where I think real usage gets interesting.
People often compare models like it is a simple winner and loser decision. But in practice, I do not think that is how this is going to work for many people. I think a lot of us are going to end up using multiple models for different purposes depending on cost, limits, ease of use, and the type of work we are trying to do.
Tokens Are the Hidden Economy of AI
The more I use these tools, the more I think tokens are one of the hidden economics of AI.
Right now, it still feels relatively affordable to experiment. You can subscribe, test things, build workflows, and start learning how all of this works without it feeling completely out of reach.
That may not last forever.
My sense is that the market right now is still in a growth phase. The goal seems to be getting as many people as possible into the habit of using AI. Get them comfortable. Get them dependent on it. Get them building it into their work and personal lives. In a way, the industry is trying to make AI feel normal before the true economics fully settle in.
And I keep wondering what happens later.
What happens when more people are using these systems every day, data center demand keeps growing, energy becomes more constrained, and compute becomes even more valuable?
It is hard for me to imagine that the most powerful usage stays this accessible forever.
This May Be the Best Time to Learn
That is really the thought I keep coming back to.
This may be the best time to learn how to use AI seriously.
Not just casually. Not just for fun. But really learning how to prompt well, how to structure tasks, how to build workflows, how to compare outputs, and how to decide which model is best for which situation.
Because right now, even with limits, it still feels like we are in a stage where people can experiment.
Later on, I would not be surprised if advanced usage becomes more expensive and more tiered. Simple usage may stay cheap enough for everyone, but more complex prompts, longer reasoning, deeper research, and automated workflows may become something that only better funded individuals and companies can really maximize.
If that happens, then learning now matters even more.
AI Literacy Is Also About Efficiency
One thing I am realizing is that AI literacy is not just about knowing what AI can do.
It is also about knowing how to use it efficiently.
How do you write prompts that get you what you need without wasting usage?
How do you decide which model deserves which task?
How do you avoid using a premium workflow tool for something a more basic model could handle just fine?
That is a different kind of skill, but I think it is going to matter.
Because if usage becomes more valuable over time, then the people who know how to get better results with less waste will have a real advantage.
My Take Right Now
Right now, I do not think the story is just about which AI model is best.
I think it is about how all of us are learning to work across multiple models, manage limits, and make smart decisions about where to spend our attention and our tokens.
That is where I find myself now.
Still using ChatGPT heavily because it feels more flexible for daily use.
Still using Claude for Cowork because the feature is genuinely useful.
Still exploring Copilot and other tools to see what can be offloaded and where the best value sits.
And underneath all of that, paying closer attention to something I did not think about enough at first.
Not just intelligence.
Not just features.
But access.
Because in AI, access may end up being just as important as capability.
Leave a Reply