Keep in mind much of the guide is about how to move from 30s chats to doing concurrent 20min+ runs
----
Spending
Claude Code $$$ - Max Plan FTW
TL;DR: Start with Claude Max Pro at $100/mo.
I was about $70/day starting day 2 via the pay-as-you-go plan. I bought in $25 increments to help pace. The Max Plan ($100/mo) became attractive around day 2-3, and on week 2 I shifted to $200/mo.
Annoyingly, you have to make the plan decision during your first login to Claude Code, which is confusing as I wanted to trial on pay-as-you-go. (That was a mistake: do Max Pro.) The upgrade flow is pretty broken from this perspective.
The Max Plan at the $100/mo level has a cooldown of 22 question / 5 hour: That does go by fast when your questions are small and get interrupted, or you get good at multitasking. By the time you are serious, the $200/mo is fine.
Other vibe IDEs & LLM providers $$$
I did anywhere from about 50K to 200K tokens a day on Claude 3.7 Sonnet during week 1 on pay-as-you-go, with about a ratio of 300:1 of tokens in:out. Max Plan does not report usage, but for periods I am using it, I expect my token counts to now be higher as I have gotten much better at doing long runs.
The equivalent in OpenAI of using gp4-4o and o3 would be $5-40/day on pay-as-you-go, which seems cheaper for using frontier models… until Max Pro gets factored in.
Capping costs
Not worrying about overages is typically liberating. Max Pro helps a lot here. One of my next experiments is seeing about self-hosting of reasoning models for other AI IDEs. Max Pro goes far, but to do automation and autonomy, and bigger jobs, you need more power.
That’s exactly what a paid comment would say
user: christophilus
created: October 16, 2014
karma: 11683
I do wonder how much astroturfing these companies do though...
Accusations of shilling are explicitly not allowed, but they make the brain of the average hacker news feel large so nobody downvoting that one.