Fed up of Replit’s costs and my lack of trust in them, I have moved to Firebase + gemini +firestore etc. etc… The summary is that Replit has much better functionality for vibe coding. Everything, from UI development, to coding, and seeing and fixing its own issues, is WAY ahead of firebase. I can build the same thing in Replit maybe 5X quicker.
Firebase / gemini requires a lot more hand holding. It makes a lot more errors which it does not catch. And you actually need a decent amount of technical nouse - though I am not a coder. You just need to know how to think like one. Someone with no understanding of technology, architecture etc. would have no hope (and I mean zero), in building a functional app.
However….
If you persevere, you can navigate a way through for a fraction of the cost or Replit - in fact I am yet to be charged anything. Also because it is so painfully slow, you actually do read what it is doing, so actually feel more in control, rather than just blindly trusting it. It is also forcing me to actually learn more about how the codebase is structured. And at the end of it all, I will have something that is totally independent of Replit and their scam pricing, without the fear that one day anything I build would be held to ransom.
My journey feels very similar to that I went on with Replit, but I am learning how to fly a DC3, rather than a jet.
Firebase is also exceedingly / excruciatingly polite when it screws up
If you were to go a different direction, why these? I’d go any IDE / Claude before that. Gemini (and maybe because my exposure to it is strictly through Google search) has been the absolute worst AI that I’ve been exposed to.
From all the reading I’ve done, there are quite a few other routes I’d bet on being a long term solution before going those 3.
Because I am not a full stack developer. The only other route I could think of is VScode+Kilocode+openrouter → claude code / gemini / chatGPT etc. etc.. I tried that but couldn’t get simple things set up like Drizzle on my local - due to my lack of skills / knowledge. I was wrestling too much with the basics which Studio seems to take care of. Is that the sort of approach you were thinking of? - I might give that another go.
I didn’t do much tinkering, a little with Cursor, Claude, and VS, but I’m in the same category (somewhat technical but not a developer, my background is middleware and ERP sales). I tried Lovable but by the time I ramped up on it, Agent 3 started working for me again and the cost dropped 75% on my prompts after Claude 4.5 (running on High w/testing turned on).
Overall, my personal take is pricing has good value at the moment. I’m basically paying $30-50/hr for a developer, EA, and QA. The results have been pretty good, and now it handles all of my CI/CD pipeline to get the app into production. In my opinion, yes, costs are a lot higher than historically, but prices on all these tools are going to get significantly higher. I’d rather race with the best solution to get something into production and monetize it as fast as possible while rates are where they are than to hit the brakes and start from square one
I think the major thing is having a clear path to monetization in a category that could sell at a decent price. For hobbyists, I could see it’d make more sense to tinker with different tools if you know you’re not able to recoup $5k to develop an app on Replit. If I were in that boat, I’d probably be learning on Cursor, VS, and Claude Code.
@daniel1389 I agree with you, I have been down the same path. Looked a few other IDE’s and then Tried out Firebase last thursday and produced a fully functioning app for my printing business by Monday. I am also not a coder, but understand enough to understand when things are not right and need changes, but every wall we came up against, it fixed in no time.
The ONLY thing it really battled with was Google Auth, which I got Grok to fix. It nailed every page layout, page function, DB setup, API setup and every instruction I gave it. Working piece by piece was the win.
Also, you can switch from Prototype agent, to the code ask agent where you can ask questions before committing anything with the build agent, same as Replit.
Plus its free, and you get to use every other google tool in the process.
Out of all the apps I have started on Replit and hit the wall with, the first app I try on Firebase, works first time. Maybe I was lucky, but thought I would add my 2c
And if you have not done page layout design in Google Stitch, before you get to Replit, you are wasting your time getting Replit to design. Get all your pages laid out in Stitch, then either feed the HTML code or images of those designs and off you go!
Interesting, I haven’t used stitch yet. Replit has been all web app which I felt didn’t need a super intuitive UI (eventually an Agent will just sit on top of it), then I’m using Lovable for the web design for the website.
What I found best with using stitch is once you do a decent first layout, you can hit option layout and 5 options are generated. Choose the one you like, delete the rest, or keep hitting option or ask the agent to fix. Best part is when the agent suggests things to add, either to that page or new pages relating to that page, that you normally would have forgotten, you just keep adding them, and deleting what you dont like or would not need.
Basically you build your app with layout in mind, with an app that suggests functions, you can zoom out, check all your screens, run through the flow, and determine if everything is correct. What I then do is plug the code into Groke Code Mode and view each pages live, and if necessary ask Grok to fix and add stuff before I plug updated code back into Stich page concerned.
Maybe they will actual listen to the costing issues and now people are trying to hack or find alternative tools. Took a break and came back. Still way too expense for simple simple things. Please give us Agent 2 back or fix the dreadful ridiculous pricing. BTW what is the point of the monthly plan when you hit the credit limits within an hour of using Agent 3 on low settings. lol
To be fair to kclubb, I think he is making the point that it is misleading. The only reason to charge $40 per month, when you consume that in minutes, is to mislead people into thinking that is going to be the order of magnitude of the charges they will face over a month.
Sorry but I disagree. The $40 is a fee for access to the service. Sure, they give you $40’s worth of credits as part of that, but nowhere does it claim you’ll only need to spend that many credits. This is a usage based service.
I would also suggest that lower prices are attainable for those who have a strong workflow and prompting strategy. I appreciate that might come across as “it’s your fault for not prompting carefully enough”. But I genuinely accept that for non-tech vibe coders who don’t have the experience on how to plan a dev project and manage a developer, it is difficult to know how to prompt in a way that keeps costs under control.
If we want to encourage non-techs to build software, then all we can do is ensure better support is available throughout the industry to guide people with how to prompt in smarter ways that doesn’t burn cash. Please believe me, it is possible! Right now, as an experienced dev, I am struggling to spend anywhere near the amounts of per-prompt or per-hour money I am seeing talked about on here.
I suppose that is a valid perspective - a classic two part pricing model.
I do believe you. I am sorely tempted to dip my toe back in given how hard it is to get Firebase working, especially as one of my workspaces somehow became corrupted and won’t start. But I do resent spending $0.5-1 per non coding prompt, and fear getting stuck in uncertain doom loops of repeatedly being told everything is working when it isn’t, but with increasing levels of “I am sorry I screwed up again” empathy statements.
Plus I have just discovered Chat GPT codex and how that works…
Do you have a suggested workflow between Codex and Replit? Or do you only use replit Agents?
Ah, I haven’t dived into this who Codex side yet - I am a pure Replit person right now, and just not enough time to play with all the fun AI toys out there.
I totally hear you about paying for things that don’t work - especially when it tries to convince you it does. But I guess that’s the price we pay for being at the forefront of bleeding edge technology.
By the way, I had to remind myself what the definition of “bleeding edge technology” is. I wonder if everyone currently signing up to platforms like Replit (but others too) need to read this and agree to be bound by it before proceeding - I am only half joking
Keep Firebase studio in Prototyping mode seems to be the answer. It doesn’t seem to get stuck on loops fixing its own mistakes. A refreshing change.
I have now created a full auth process, DB set up, sign up with google, multi language in next js - basically the foundation for the app for the sum total of $0. It has taken a while, but am feeling pretty confident about where this might go.
Because costs are low I have two app, one for playing / experimenting, and then the real dev environment. I deploy often and commit / push only when something isn’t broken. Also using ChatGPT do to security and tech debt code reviews linked directly to the repo via codex Getting there….
I started a big project in replit almost a year ago and used firebase + firestore for authentication and database. Recently tried to move to Bolt but it doesn’t allow to do it, it forces to use bolt database. I asked chatGPT if I could deploy everything in firebase but it told that I had to use a different hosting for the backend (doing some container stuff and put it in Supabase or something like that).
Just came back to Replit but planning to find out how to do the deployment outside if this idea gets some traction. I have little knowledge of DevOps…
