In scarcity, you think. In abundance, you defer to the tool. Today we unpack this principle through a week with Claude credits running on empty.
Here's a moment that hits anyone who gets serious about AI tools. You're paying your monthly subscription happily, and one day a message pops up — "Credits low." It feels a little panicked. Work stops, you debate paying more, you start hunting for free alternatives.
This essay explains why that panic is actually a gift. Even if you've never used an AI tool before, you can follow along — we'll go slowly. Today's example is a week with Claude Code credits on empty, but the principle applies to any resource — money, time, materials. Three years from now when Claude is renamed, the spine of this essay still works.
Start with the principle. When resources are plentiful, people don't use their heads. There's no need. The moment deliberation crosses ten minutes, we hand it to the good tool. "Just throw this at Opus." Another $5 and the problem is gone.
But when resources shrink, the brain turns on. You can't afford the $5 solution, so you have to invent the $0 solution. That invention is the point. When we deliberate, we re-examine the problem. "Does this really need AI?" "Can't I do this in ten minutes myself?" "Isn't half of this just three lines of Python?" The type of question changes.
This isn't a new idea. Businesses work the same way. Well-funded companies buy growth through ads. Cash-strapped startups have to figure out word-of-mouth. Cut the ad budget and the first company stalls. The second company keeps going — because of the muscle they built in want. Strangely, the knowledge disappears in front of AI. A subscription tricks us into acting like the resource is infinite.
My own story. I burned through a month of Claude Code credits in twenty days. The work stopped. For about two seconds I thought "how do I get through ten more days?" — then I decided to try free models. I pulled free models through a service called OpenRouter.
Day one was rough. What Sonnet did in three seconds, free Llama took eight. The quality was worse. "Paid really is better," I muttered. But from day two something strange happened. I started writing sharper prompts. Free models don't handle vague instructions well. So I broke tasks down, added examples, specified formats.
By day three the realization hit. The prompts I was writing now, if fed to a paid model, would produce far better results than before. Sonnet had been quietly absorbing my lazy phrasing. The free model exposed it. What I learned about prompting that week beat what I'd learned in a month of paid use.
Picture a kitchen. A full fridge lets a cook repeat familiar recipes. Beef? Steak. Chicken? Fried chicken. But shrink the fridge to one onion and two potatoes and suddenly the cook gets creative. They think about what to do with the onion, what the potatoes can become, what to substitute for what isn't there. Constraint teaches the cook.
AI tools work the same way.
Throwing every small task at Opus is the comfort of a full fridge. Comfortable, but growth stops.
Let's check with numbers. Over that week I saved about $15. Not a big amount. But I gained something far bigger.
| Metric | Abundant month | Scarce week |
|---|---|---|
| Problems solved | ~80 | ~30 |
| Avg prompt length | 2-3 lines | 8-10 lines |
| "Is there another way?" moments | almost none | every time |
| New workarounds learned | 0-1 | 6 |
The month wins on volume. But six "is there another way?" moments stack into something that lasts a month later, a year later. The abundant month had output, but nothing left inside me.
Here's the first aha moment.
A week with credits on empty teaches more than an unlimited month.
Money isn't the issue. The count of deliberations is. Abundance erases deliberation. Scarcity forces it. People learn as much as they deliberate. That's the core principle today.
How to apply this? Simple. Carry one question in your pocket.
"What resource will I deliberately shrink this week?"
The answer varies by person.
One week each month, work without paid models. Use only free ones, or spend ten minutes thinking before handing anything to AI. Fastest skill growth you can get. Awkward at first. Not awkward by day three. After a week, returning to paid feels like the same tool got twice as strong.
"I'll finish this in one hour" — self-imposed limit. Give yourself plenty of time and perfectionism eats four hours on one task. A one-hour limit forces the important first. The rest wasn't important anyway.
Cap it at three. If you want a fourth, one of the three has to go. Unlimited tools means no mastery of any. Constraint breeds skill.
One line to remember all three: Want. Think. Stay. Three words.
Today's principle in one line.
In abundance we defer to tools. In scarcity we use our heads. Skill lands only when the head is used. We used Claude Code credits as the example, but money, time, tool count — the same shape applies. Service names will change. The effect of want doesn't.
Engineer scarce seasons on purpose. One week a month without paid tools. One hour a day without AI. The muscle built in that time is your real skill. It survives when abundance returns.
Muscle built in want survives plenty. Technology changes. The principle doesn't.