My usage cost has gone up 5x and I have to stop using the service.
Been using the agent for building apps for the last 5 months. I compared a full day of using the agent months ago compared to after agent 3. Even with testing and level down to minimum my spend is now 5X what it was before. The service is now no longer economically viable. The agent has actually broken features that were working before to add a new feature then spent more money helping it fix what it already implemented and was already working before. This is extremely frustrating.
I have to use my PRO OPEN AI chatgpt account with 03 w/ research mode to actually create code examples and guides to attach to my Agent requests so it can actually have a better chance of getting the job done. I don’t mind some cost optimization and paying money for value. But right now there is no way i can continue to use this service under this pricing model.
Completely Agree 1000%. I can no longer use this service under the current pricing model. I can deal with agent mistakes but not if it costing this much more.
Here is what is starting to piss me off. If I am in a situation where I have to repeatedly ask replit the same question over and over and over again to try to fix an error or get something to work, shouldn’t there be some type of discount for never getting it right after 20 to 50 tries but charging each time ? I am asking replit to get the autocomplete feature working on a form from google places api. it was working before, now it is not. And it is trying again and again and again and charging me every time !
I figured it had to be pretty complex for that much code. I would love to look at what you have built when you’re done! I’m trying to get back at mine, it’s hit or miss with agent 3, i’m going back and forth with support right now, hoping to get some help to get going again. I do see its very good when it does work I think i need to make some tweaks i’m sure to get it to work with my older codebase and get refactoring started.
Let’s goooooo! A few questions for you (I’m trying to prepare for testing in the next 1-2 months):
Are you doing unit testing, and if so, how are you doing this?
Are you taking advantage of Agent for user testing? How are you finding the results? (As long as Agent and I have been on the same page for expectations, the results for me have been quite good. Wondering how it’s working at scale for you.)
Have you done any prompts that have triggered a large-scale test of the entire app? If so, would love to hear what you’ve done.
Full disclosure: I used gpt-5 via CLINE to complete them, as it was the cheaper/more accurate option from my experience.
I just instructed gpt-5 in plan mode to create an .md file with a phased plan to implement complete coverage of >=90% for the entire codebase. I then went phase by phase until it was complete.
In my experience, the LLMs are pretty good at inferring unit test. I am happy with the results.
