Back to blog
Marco Polo
Marco Polo recording
The Marco Series

The cost of keeping several AI coding projects moving

What happened when I hit AI coding limits while keeping several projects moving.

Levi Lais

Levi Lais

Founder, VirtualExperts

December 27, 2025

3 min read

I was trying to get my finances done, doing a massive refactor of the VirtualExperts backend, and downloading and running AI models for a client. With all four projects humming at the same time, some with multiple agents for the same project, my AI budget had to go up.

Two weeks ago, I hit the Claude limit about two days early. I basically had no AI, and I'm like, I can't not have AI. So I paid 20 bucks for Cursor thinking I'll make this work. I got the last couple things I needed done using Cursor, but switching environments was a pain in the neck. I was using VS Code and ended up having to close all that down and open Cursor.

I mostly switched back to Claude once my usage was back. Then, the second week in a row, I hit Claude's max. It was Saturday and I wouldn't get more usage until 4 p.m. on Monday. So I turned on overflow and spent like $60 in 45 minutes. The pay-as-you-go was insanely expensive for what I was doing.

I went back to Cursor, and within about 10 minutes I hit the month's budget for the $20 plan. I upgraded to the $60 plan, and within about an hour to an hour and a half I hit that max. It was like, you've got to be kidding me right now. I needed to see what their Auto worked like again as a backup option, because I still had all these projects going.

I did get my reports in last night. I made it. Still tons to do, but I was able to get all my year's data from all the necessary accounts for the two businesses. I've just got more cleaning and working and scrubbing to do on that.

The $400 a month for AI is insane. But Cursor's new agent stuff is so nice. With Claude, you kind of drill into a spot and you're just sort of there. With Cursor, I feel like I can hop over to another agent, hop over to another agent, see that this one's done, and check back. It's like getting fresh, clean context for all the ideas that I have.

Right now I have five different agent windows from the last 32 minutes on VirtualExperts. Every idea that pops into my head, I just ask it to do it. That is a much more useful tool for the way I'm working than the terminal. I have it on Auto right now, and it's working great. It's doing what I'm asking. I haven't had to step in and ask it to use something smarter hardly at all. I don't know what they did in their Auto, but it's working.

In the last two months, I've tested Codex, Claude and Cursor. I'd say Cursor and Claude are neck and neck right now for me. Codex was days of beating my head against a wall and then being like, I need to just go back to Claude. What am I doing? But I'm pretty excited about where all this is headed.

Join the waiting list

Your AI team is coming

Join the waiting list. No payment required.

Join waiting list