So I started out vibe coding a accessibility program about 5 months ago, its been quite a journey, slowly learning about the Ai coding tools out there and as well just seeing the new models come in, each one more copiable than the last. I found that Cursor AI was a great tool for my use, the 20$ subscription with slow requests after I used up the fast ones was great for me, as I could keep going with my project through the whole month.
So now with the new pricing, 20$ gets me about 6 days of usage before I hit my limit and am stuck with Auto which really is only good for the simplest things. I have tried buying credit at open router and using Kimi K2 through Cline, that was promising, I was able to make some progress, but I used that $20 up in about a day, which end up being too costly for me.
So how does a person go about vibe coding these days where $20 a month doesn’t get you far? I suppose I could spend $60 on the upgraded plan, but I think that’s just a straight 3x more usage, so 18 days and then I’m waiting for the next cycle to roll around?
Its one thing if the AI capabilities I was paying for was a hit every time, but the reality is that I spend a lot of time just trying to get the AI to accomplish what I want it to. Its like every prompt is a gamble weather I’ll make meaningful progress, or create a mess that I have to clean up. Yes I am way further a head now with my program than when I started, and am super impressed. But these new prices are really throwing up a wall for me. If I spend 1/3 of the money on meaningful AI programming and help, then 2/3 of it cleaning up messes or just going down dead ends, I need an affordable subscription that gets me through the whole month if I am going to get anywhere. I don’t see that being offered right now.
So how do you folks manage with your projects? Do you have strategies where the subscription gets you through the month? Do you find yourselves cutting back? Spending more? I do feel like if I knew more about programming, or perhaps just good vibe coding strategies, I could go further. I feel like I know a fair bit, and am conscious about learning as I go.
Sadly I just don’t think current AI tech is advanced enough for vibe coding just yet. Yes, it can be done, but as you noted it takes a lot of rework and involves lots of dead end rabbit holes.
The old Cursor billing plan made it work, because we were getting a great deal paying only for requests regardless of token consumption and getting 500 a month. That gave us a sweet window of time where we could vibe code and not worry about the inefficiency… however, Cursor was the one having to pay the final bill with the AI providers, and I’m sure they were losing a ton of money. Now they’re passing the cost onto us, which is a bummer but makes sense from a business perspective.
I can’t truly answer your question because I’m not a vibe coder, I’ve been full-time software dev for a long time now. However these are the strategies I use which may be helpful.
Take extra time planning and crafting prompts. I use a sketchpad to organize my thoughts, outline exactly what I understand and/or want to happen. If I’m trying to fix a bug: gather as much details about it as I can. Trying to add a new feature: map it out as thoroughly as possible. When prompting the model, take the time to write a long and detailed prompt, with proper grammar (yes that makes a difference).
Break the work into as small of tasks as possible and open a new chat for each task. Long-running chats will exponentially increase in token usage and should be avoided at all costs.
Create a good set of rules in Cursor Settings. I use my rules to define the app and its intended uses, specify business rules, and guidelines for the AI such as what software versions are in use, instructions to refer only to official documentation for those versions, etc. This will save you time in your prompt writing and make it easier to constantly open new chats.
For each new task, use Auto mode first. If that fails, then I will first see if I can revise my prompt or approach to make it work better. If that doesn’t work, then switch to more expensive model.
Yeah Cursor is done when it comes to vibing. The most promising next thing for vibe coders is Kiro, with even better value for money than Cursor used to have.
Thanks, I appriciate the replys. Kiro looks promising, though I wonder if when its up and running if it’ll charge as much as Cursor. And I do think if I were more focoused and planned I could probably be at least 2x as efficient in the tolkens I use. I do find the sessions I accomplish the most are the ones where I sit down free of distractions and carefully interact with the AI. Got hopes for Kimi if Cursor gets it implemented well!
It will. Virtually all new products like this operate at a loss to start with, using low prices to draw in a big user base. We get a cheaper service, but also they get a bunch of beta testers who are providing critical feedback to help them develop the service (and paying something at least).
Then once the user base is big enough and the product is polished enough to justify it, prices go up.
If you’ve heard anything about AI you’ve probably heard that the cost of making and running these LLMs is astronomical. There’s no way we will be able to keep using them for cheap.