Don't pinch pennies on your AI sub

I've had this conversation with two smart, successful friends in the last week, so I'll assume there are more of you. Stop focusing on your usage limits. Focus on the work you can get done.

Stop being precious about a $20 or $200 a month investment that pays back ten times its cost. If you work in technology, or your work is largely digital, the model you're running is setting the ceiling on what you think is possible. Premature optimization is not prudence.

If you don't have $100 a month to spare on this, this rant isn't aimed at you. I get it. Scale to your reality. And yes, local open-source models are getting better every month. If you have the chops and the hardware, that's a legitimate path with a lot to learn. The reason I haven't gone that direction is overhead. The S/M/L of one frontier provider is enough for me to manage.

For context, I am not a big spender. Probably has to do with how I grew up. There's also a value element to it for me. I'd rather pay $10 for $15 of value than $8 for $5. And I delight in not spending the money. I could always spend it tomorrow if I need it. I can't unspend it tomorrow. Two years after moving back to NYC from a year abroad I still rotate the same two or three pairs of pants I slimmed my wardrobe down to before we moved away.

My first job out of college, I was a consultant at IBM Global Services. They issued me a work laptop with dial-up internet access, which dates me. When I moved into my first apartment I made what felt like a smart frugal call. Skip broadband, use the work dial-up at home, save the money.

I did that for six months. Then I left IBM, handed back the laptop, finally got broadband, and felt ridiculous. I was a professional technologist. I built internet services for a living. To save a few dollars a month I had taken myself out of the game for half a year. Half a year of not experimenting at home, not building intuition for what the web was becoming, not having the friction-free environment that would have made me better at my actual job. The savings were real. The opportunity cost was much larger.

Penny-wise, pound-foolish. Some line items are expenses. Some are investments. Being good with money means knowing which is which.

My $200 a month Anthropic sub right now is an investment. A $20 sub used well is an investment. The meme is "I hit my usage limit, I'll come back tomorrow." If you are in tech, what that sentence actually says is "I am willing to pause my learning on this technology for 24 hours to save twenty bucks." Read it back.

People tiptoeing around on a free model or a too-small paid one are having a markedly different experience than someone running the apex model with the thinking dialed up. They're not just spending less. They're forming a wrong picture of what these tools can do, based on a weaker version of them. On a weaker model they fail at things that work now, learn the wrong lesson, and quietly shrink the next thing they think to try.

The same impulse shows up another way. Once you've paid for a tier, the instinct is to manage usage within it. Make it last. Default to Sonnet because it stretches your Pro plan farthest. Prudent in theory, misses the moment in practice. The "tokenmaxxing" idea, where people deliberately run their AI usage hot to push the tools harder, gets one thing right. You need the room, and the push, to figure out what these tools can do and how to apply them to your life.

This is how I actually use it. I live in Opus at "xhigh" effort. Max effort when things get complicated. I am not optimizing, I am exploring. Once I've shaped a workflow and I understand what the task really is, then I dial back. Sonnet for summarization, lower effort for the parts that don't need heavy artillery. Optimization comes after I understand the shape of the thing, not before. Optimize first and you've optimized your way into a smaller version of the work.

Having spare tokens also opens the aperture on what you'll attempt. This past weekend I finally reclaimed my personal Gmail because I had Claude not just triage the inbox but rewrite the filters so I could keep it managed going forward. I've been stuck on Superhuman for years because it does this and I couldn't be bothered to replace it.

I am not telling you to start at $200. I'm telling you to use whatever you've got, treat your usage limit as an upgrade signal, and stop pre-emptively economizing on the part of your stack that determines your ceiling. If you hit a wall, upgrade.

That smart professionals are penny-pinching on this is the part that gets me. I'd cancel the family Netflix account before I'd touch my Anthropic sub. I spend $200 a month and get thousands back.