Use the browser and ditch the desktop. Desktop needs to be redone/refreshed but currently isn’t maintained.
I’m curious why you’ve given up on agent. I’m having a lot of success with it. Turing off app testing and keeping autonomy in low mode is a very similar experience to agent 2. I find myself using all the settings depending on the task at hand.
What is your experience and workflow currently?
As @realfunnyeric mentioned I would also suggest using the browser as we are working on some more immediate updates to the Desktop version. I will pass your issues on to the team here!
For what it’s worth, I never see any overriding reasons to use desktop apps, and always prefer the browser version, for one key reason: desktop apps on Windows never seem to allow ctrl-F to find text in the app window.
Instead, for all the key platforms I use on a daily basis, I ask the browser (Edge in my case) to install this site as an app. I get a nice icon in my menu bar, and I can use ctrl-F as much as I like 
I really enjoyed the Mac version while it was working, and I hope that they keep it up.
@realfunnyeric Appreciate the tip, ditched the browser.
re: persistent issues I still experience: @FranciscoCM
(1) Assistant processes my request, generates a diff, but does not provide an “Apply” button. this happens a dozen or more times per hour:
(2) Almost invariably, every Saturday, the latency makes work nearly impossible. Not sure whether this is Replit internal, or they get hit with a disproportionate amount of traffic on Saturdays, but this has been my experience over the past few weeks
Eric - re: agent experience. I’ve tried it multiple times in v2 and v2, and found it going off course, and the mistakes getting quickly compounded, rendering the entire workflow a throwaway - or requiring me to then unf*ck the litany of erroneous or undesirable changes and additions. Keeping a close watch & reviewing most diffs via Assistant has proven to be much more reliable. Happy to describe my workflow (it’s heavily spec-based), but tell me more what you mean by your question - and share a quick example, if you can.
re: desktop - having a standalone app can be nice. One benefit comes to mind - in Replit desktop, I can quickly close individual file, etc. tabs using Cmd+W on Mac - can’t do that in the browser. Got so used to it, that now in the browser I have to unlearn - but keep closing the entire Replit tab…. and as you know, bringing it back to life when you have lots of shells running various jobs is quite painful. And that’s on top of Replit’s killing my shell jobs on a regular basis, sigh.
I second the below concerns that others have mentioned.
Massive cost increases after Agent 3 launch – Many users report bills jumping 5–20x overnight for the same tasks.
Charges for failed or idle actions – Users repeatedly mention being billed when the Agent does nothing, hangs, or errors out mid-task.
Frequent errors and session crashes – “Agent encountered an error” loops halt development, waste money, and break work mid-session.
Charges for Replit’s own errors– Users resent paying when issues stem from platform bugs or failed AI attempts.
Calls for user advisory group – Several suggest a beta-testing or feedback council to prevent “closed-room” decisions.
I only use the assistant. I can’t even wrap my head around letting an agent change multiple files without me approving since the assistant gets so many things wrong or creates issues that impact other parts of the application. I like the planning vs coding tab, I always move to new tabs anytime a task is complete or after the assistant goes in the wrong direction since it’s memory includes files it provided but you didn’t accept.
That’s precisely the problem with Assistant.
Yeah, the assistant going rogue on multiple files is nerve-wracking. I’ve started using AI tools just for the initial structure, then dropping in stable components for anything production-critical. Like for features the AI struggles with (real-time stuff, file management), pre-built Web Components save me from those endless correction loops. Ever tried mixing AI generation with modular components? Keeps the assistant focused on app logic instead of reinventing complex features.
Yeah, the assistant going rogue on multiple files is nerve-wracking. I’ve started using AI tools just for the initial structure, then dropping in stable components for anything production-critical.
Like for features the AI struggles with (real-time stuff, file management), pre-built Web Components save me from those endless correction loops. Ever tried mixing AI generation with modular components? Keeps the assistant focused on app logic instead of reinventing complex features.
Yeah, those 20x cost spikes are brutal. Same thing happened to me when the agent kept erroring out and burning through credits.
I’ve been moving my production features to stable components that don’t rely on the AI runtime. Like for collab stuff, I just drop in pre-built components now instead of having the agent regenerate chat code every session. Saves a ton on credits and actually works reliably.
Which features are eating most of your credits? Sometimes there are workarounds to avoid the expensive loops.
Lately the agent has been reporting more often that “…the server has automatically restarted” when it clearly hasn’t (preview window stays as-is). This forces me to prompt the agent to restart the server when it should have just done it manually instead of saying it automatically restarted without check. This may only cost $0.20, but that adds up when making many changes daily that require restarts to pickup.
I have now cancelled my Replit subscription. It just didn’t seem sustainable to continue given all these issues and the risks of the unknown cost hikes and other performance issues in the future causing blow out costs. Firebase may not be as good, but I can try to do the same thing 10 times and it still doesn’t cost me the same as one prompt on Replit. It actually makes building something more enjoyable as I don’t have the constant fear of what pressing “Return” will cost me.
But thank you everyone for your input, and good luck with your projects.
Always sad to hear someone leaving. Especially when a big issue is cost. Because, as I have already said, all platforms will eventually increase their costs. So perhaps we will see your return when Firebase prices go 10x Best of luck though!
Here’s an example of other agent issue. I provided feedback on how a frontend field should be cleared expecting the agent in build mode would attempt to fix it. Instead, it decided on the fix and created a checkpoint. No feedback. I had to prompt it to actually implement the change. The charge might be accurate (who knows considering the vague pricing structure), but this causes confusion and is inefficient.
if I were to summarize the positives and negatives I see:
Look. I want to be able to use Replit in production enterprise environments - it’s an obvious positive path to go down but - reliability, trust, and the ability to create consistent framing/guardrails is just a necessity. Otherwise, I imagine a situation where I have to use CodeRabbit adversarially to evaluate and correct poorly executed Agent3 code - or where I just use Replit for UI/UX design and then have Cursor produce the whole back-end. That seems redundant to me. As I said - I want this to work - but my trust in the process is being blown up over and over.
Pricing is so ridiculous!. I asked the agent in the chat “does simply chatting with you cost money? The answer to this question costed 0.36 cents!!! WTH! Bye, Bye replit until you fix this nonsensical pricing.
Way too expensive now. Just asking for a simple question costs over a dollar. This was less than $0.30 a month ago. Planning migration out now.
@alex-replit @horacio-replit @ZergGirl @bc4d31197c2cc5714b98 @d5ec48107c00961da4bb @24c8a795b923fa3456be is anyone at Replit regularly active on this forum? (I don’t know if you work at/for replit, just tagged users with “replit” in their identifiers in hope this gets flagged)
Hi! Did you need help with your Replit account? Feel free to DM me as well 
