The Crash of 0% Usage
What running out of AI usage taught me about productivity, dependency, and why we still need to learn how to do the work ourselves
By Donald XL Robertson
I ran out of AI usage in the middle of working on a project.
And I’m not going to lie that 0% felt different.
A few minutes earlier, I felt like I could do almost anything.
That’s one of the things I love about where AI is right now. You can sit down with a task you honestly have no idea how to complete and start figuring it out.
Maybe you don’t know the software. Maybe you’ve never written code. Maybe you don’t know how to build a website or fix the technical problem stopping you.
Normally, that’s the point where you start searching YouTube, call somebody who knows what they’re doing, hire somebody, or just put the project off.
Now you can ask AI.
And when it starts working, you start working.
Things start happening.
You feel empowered.
You start feeling like you’ve finally joined that group of people who know how to get things done.
Then your usage runs out.
0%.
Now you’re sitting there looking at the same computer.
Same software.
Same project.
Same problem.
Except your AI isn’t available anymore.
And that’s when it hit me:
How much of what AI is doing, do I actually know how to do?
I Was Getting the Work Done. But Was I Learning?
That’s the part I think we need to talk about more.
I use Codex and ChatGPT because they do things I couldn’t necessarily do on my own.
I’m not ashamed of that. That’s what tools are for.
But there’s a difference between a tool helping you do something and a tool becoming the only reason you can do it.
When AI is working beside you, it’s easy to overlook that difference.
You’re clicking buttons, moving files around and watching things happen on the screen. The project is coming together.
It feels like you’re learning.
But are you?
Because when my usage hit 0%, I was still sitting in front of a powerful computer with all this software on it.
The computer hadn’t lost its abilities.
I had lost access to the thing that knew how to tell the computer what to do.
That’s a big difference.
And that’s when I realized I needed to start paying closer attention.
Maybe We Need Books Again
I know that sounds funny in an article about AI.
But I started thinking about books.
Actual books.
Manuals. Reference guides. Things you can own.
Because if I’m going to use AI to help me with software I don’t understand, maybe part of the process should be learning the software while AI is teaching me.
I don’t want to spend months getting amazing results and then realize I can’t reproduce any of them without a subscription.
I don’t want AI to just do the work.
I need it to teach me the work.
That’s a different relationship.
Instead of only saying, “Build this,” maybe I need to ask:
“Show me how you’re building this.”
“Why did you do that?”
“Where is that setting?”
“What does this code actually do?”
“Could I do this part myself?”
That’s where the real value might be.
Because These Usage Limits Change Things
Think about how strange this would sound with something else.
Imagine you’re cooking dinner and halfway through cooking, the stove shuts off:
You have reached your cooking usage limit. Come back in five days, or upgrade now.
Imagine your refrigerator doing that.
Sorry, you’ve used all of this week’s refrigeration. Your food can start getting cold again Tuesday.
It sounds ridiculous.
Now, I understand AI isn’t a stove. These systems require enormous computing resources to operate. My point isn’t that AI usage should automatically be unlimited.
My point is what happens to us when our ability to work becomes dependent on metered intelligence.
A lot of the tools we depend on aren’t really ours.
We’re renting access.
And now some of the intelligence helping us operate those tools is being rented too.
There’s nothing inherently wrong with that. Companies have to make money, and these systems cost money to operate.
But as users, we need to understand what kind of relationship we’re entering.
Because that meter matters when you’re doing real work.
Time Really Is Money Now
This also changed how I think about helping other people.
I’ve always been the type of person where somebody can call me with a problem and I’ll try to figure it out.
“Send it over.”
“I’ll look at it.”
“I’ll see what I can do.”
But now there’s another calculation.
If I’m using an AI model with limited usage, helping somebody with their project could mean using up the capacity I need for mine.
That’s strange when you really think about it.
You’re not only giving away your time anymore.
You’re potentially giving away part of your AI usage too.
And when that usage is gone, your choices might be to wait for it to reset or spend more money.
For years, we’ve understood the phrase time is money.
AI adds something new.
Now intelligence can have a meter too.
Then I Saw the Carrot
This was probably the biggest realization for me.
Let’s say I’m using AI to help build something that will eventually make me money.
My usage runs out before I finish it.
Now I need more AI usage to finish the project.
But the project hasn’t made any money yet because it isn’t finished.
So I may need to spend more money to get access to the AI that helps me finish the thing that’s supposed to help me make more money.
There it is.
The carrot dangling in front of you.
And the reason you have to keep reaching for it isn’t necessarily because you need AI for everything.
Sometimes it’s because you never learned how to do the task without it.
That’s the part we can control.
So before we get too comfortable letting AI do everything, here are five things I think we need to remember.
1. Make AI Teach You While It’s Working
Don’t just say:
“Do it.”
Ask questions.
Why are you doing that?
What does that setting control?
Where did you find that?
What would I have to do if you weren’t available?
Ask AI to explain the process like you’re going to have to do it yourself tomorrow.
Because you might.
If I’m paying for access to intelligence, I want some of that intelligence to stay with me after the session ends.
2. Own Some of Your Knowledge
When I say we may need books again, I don’t mean books are the only answer.
I mean we need to start building our own libraries of knowledge.
Get Donald XL Robertson’s stories in your inbox
Join Medium for free to get updates from this writer.
That might include:
- Books and manuals
- PDFs and official documentation
- Personal notes
- Screenshots
- Checklists
- Saved workflows
- Templates
- Keyboard shortcuts
- Code snippets you actually understand
- Instructions written in your own words
If you’re constantly using AI to work inside a particular program, learn that program.
You don’t have to become an expert overnight.
But every week, you should know a little more than you knew the week before.
Build a library that still works when the screen says 0%.
And maybe even more importantly:
Don’t just own the output AI created. Own some of the knowledge required to recreate it.
3. Stop Looking Away While AI Works
This one is important.
It’s tempting to give AI the task and go do something else. I understand why. That’s part of the attraction.
But if it’s doing something you don’t know how to do, watch it.
Pay attention to the steps. Look at the names of the tools. Look at where the files are going. Read the code. Study the formulas.
Ask questions when something doesn’t make sense.
Treat it like somebody knowledgeable is sitting beside you showing you the job.
Because that’s basically what’s happening.
4. Don’t Use AI for Everything You’ve Already Learned
This is something I think could save both money and usage.
Once AI has taught you how to do something, try doing it yourself next time.
Maybe AI had to walk you through all ten steps the first time.
The second time, maybe you only need help with four.
Eventually, maybe you don’t need help at all.
Now you’re saving your expensive AI usage for the things that actually require it.
That’s progress.
The goal shouldn’t be figuring out how to make AI do everything for you.
The goal should be figuring out where AI actually gives you an advantage.
5. Have a Plan for 0%
This is the one I’m taking seriously now.
Before you get deep into a project, ask yourself:
What happens if my AI usage runs out tonight?
Can I keep working?
Do I understand enough to continue?
Did I save the important instructions?
Do I know what the AI changed?
Do I know where my files are?
Could I at least complete the basic work myself?
Maybe it takes me three hours instead of thirty minutes.
That’s okay.
At least I’m not completely stopped.
AI Also Showed Me What I Could Learn
There’s another side to this that I don’t want to overlook.
Before AI, there were certain things I might not have attempted at all because I didn’t know where to start.
That’s important.
AI didn’t just make me faster.
It got me through the front door.
Now that I’ve been inside these programs, watched things being built, seen problems solved and learned some of the language, I’m not starting from zero anymore.
So maybe the mistake isn’t depending on AI to get us through the door.
The mistake is walking through the door and never looking around.
If AI helps me do something today that I couldn’t do yesterday, tomorrow I should understand at least a little more about how it was done.
AI can open the door. I still need to learn the room.
And that made me think about a better way to use these systems.
The AI Learning Loop
Before, my workflow was basically:
I have a problem → AI solves it → I keep moving.
Now I want it to look more like this:
ASK → WATCH → UNDERSTAND → EXECUTE → REPEAT
Ask AI for help with something you don’t understand.
Watch what it actually does.
Understand as much of the process as you can.
Execute the parts you now know how to do yourself.
Repeat the process the next time you encounter something you don’t know.
The first time, I may need AI for all ten steps.
Next time, maybe I only need it for six.
Then three.
Then one.
And eventually I might be able to handle that particular task myself.
That doesn’t mean I stop using AI.
It means I’m saving the powerful AI for the next problem I haven’t learned yet.
That seems like a much better return on the money.
I Think There’s a Better Way
I’m not backing away from AI.
I’m going deeper into it.
But I’m changing what I want out of it.
Use the powerful model when you need the powerful model.
Let it attack the part you don’t understand.
But while it’s doing that, learn.
Then execute whatever you can execute yourself.
Save that AI usage for the next thing you truly don’t know.
If I’m paying for these tools, I don’t only want the finished project.
I want the education that came with building it.
That’s a better return on my money.
That 0% Might Have Been Good for Me
I didn’t feel that way when it happened.
When I saw 0%, I felt like somebody had unplugged my computer.
I had work to do. I was moving. Things were getting accomplished.
Then suddenly I couldn’t work the same way anymore.
That was frustrating.
But it exposed something I needed to see.
My computer wasn’t useless. I just didn’t know enough yet.
That’s a very different problem.
And it’s a problem I can fix.
AI may be one of the greatest learning opportunities we’ve ever had.
Think about it.
You can have something sitting beside you that can explain software, answer questions, troubleshoot problems, demonstrate techniques and help you complete an actual project at the same time.
That’s bigger than simply having something that does your work for you.
But only if we’re paying attention.
So I’m going to keep using AI.
I’m going to keep using Codex.
I’m going to keep using the most powerful models I can get access to when the work calls for them.
But I’m also going to start watching more closely.
I’m going to ask more questions.
I’m going to save what I learn.
I’m going to buy the books.
I’m going to understand the programs I’m using.
Because the next time that meter hits 0%, I don’t want to look at my computer and feel like the work has to stop.
The goal isn’t to reach a point where I never need AI.
The goal is to make sure every time I use it, I need it a little less for the things it has already taught me.
And the next time I see 0%, I want to be able to look at the computer and say:
Alright. I learned enough. I can take it from here.
Donald XL Robertson is a music producer, entrepreneur and founder of XL Productionz. He writes about music, technology, business and the changing relationship between creativity and the tools we use to get work done.
