Please have a read below, and let’s all pull together to see if we can help Replit improve the latest workspace, agent 3 and pricing issues.
I am trying to get a dialogue going with a few Replit people to discuss our concerns, and also to hopefully nudge them into engaging more with this community.
To help the Replit team analyse the issues and overall sentiment, I’ve taken 26 of the recents posts that raised concerns, and asked ChatGPT to summarise them. The report is below.
Please add any additional summary bullets below for Replit to see, highlighting your key concerns, bugs, and issues that aren’t already covered. Do not write lengthy descriptions of your experience - that is best in its own post
Report: Issues raised by Replit user’s community members
13th Sep 2025
This is a detailed, structured summary of all the issues raised across 26 recent posts. They are grouped by theme, included how often each was raised, and noted sentiment strength (Minor / Moderate / Major concern) based on tone, repetition, and frustration levels. This was produced by ChatGPT, so please allow for the usual AI “interpretations”.
1. Pricing & Cost Transparency (Major Concern)
Massive cost increases after Agent 3 launch – Many users report bills jumping 5–20x overnight for the same tasks. Some cited $20–$50 charges per prompt and even $300 in 3 hours. (15+ users, Major concern)
Charges for failed or idle actions – Users repeatedly mention being billed when the Agent does nothing, hangs, or errors out mid-task. (8+ users, Major concern)
Inconsistent and opaque pricing model – Examples show wildly different costs for similar workloads, with no clear breakdown of charges. (6+ users, Major concern)
Lack of upfront pricing clarity – Marketing suggests “build anything for $25/month,” but real costs for production-ready apps run into thousands. (5+ users, Major concern)
Casual/hobby users priced out – People building for learning or personal projects feel excluded, asking for an “unlimited hobby plan” or fairer tiering. (3+ users, Moderate concern)
Frequent errors and session crashes – “Agent encountered an error” loops halt development, waste money, and break work mid-session. (10+ users, Major concern)
Lower accuracy vs Agent 2 – Users say Agent 3 ignores instructions, makes wrong assumptions (e.g., defaulting to React when told not to), or breaks existing code. (7+ users, Major concern)
Slower and less “smart” – Several felt Agent 3 performs worse than Agent 2 in speed, reasoning, and usefulness. (6+ users, Moderate concern)
Auth testing bugs – Agent 3 overrides authentication, resets passwords, or fails to handle login flows correctly. (4+ users, Major concern)
Can’t handle complex tasks well – Users say long runs often produce buggy, incomplete, or broken apps requiring expensive rework. (5+ users, Major concern)
3. Workspace & UI/UX Issues (Moderate Concern)
New workspace forgets panel layouts and settings – Panels reset every time you reopen the app; users find this regression frustrating. (5+ users, Moderate concern)
Files & panels moved around – Locations changed without warning; new users couldn’t even find files at first. (3+ users, Moderate concern)
Themes bug – Themes vanish after editing unless the Agent “reconnects” them. (1 user, Minor concern)
Users want the old UI back – Multiple complaints about unnecessary UI changes, smaller icons, and loss of familiarity. (6+ users, Moderate concern)
Desktop app neglected – Missing features (e.g., cache clearing) and no longer actively maintained per staff comments. (2+ users, Minor concern)
4. Billing Model & Trust Issues (Major Concern)
Charges for Replit’s own errors – Users resent paying when issues stem from platform bugs or failed AI attempts. (6+ users, Major concern)
Perception of “cash grab” – Many feel features like 200-min autonomy are marketed positively but hide cost traps. (5+ users, Major concern)
Lack of engagement from Replit staff – Users complain that concerns on pricing and errors get little official response. (4+ users, Moderate concern)
5. Technical Bugs & Edge Cases (Moderate Concern)
Shell input broken on Mac – Special characters like {}@ fail in the shell after updates. (1 user, Minor concern)
Environment hitting /api repeatedly – One user saw hundreds of health-check hits per minute from Replit’s own systems. (1 user, Minor concern)
App testing can’t handle auth or slow-loading pages – Reports of test agent failing login flows or misdetecting live sites as “down.” (3+ users, Moderate concern)
Missing/removed “Extended Thinking” model – Users confused about disappearance; unclear if it’s auto-applied now. (3+ users, Minor concern)
Agent changes user requirements mid-build – e.g., replacing vanilla JS with React despite explicit instructions. (1 user, Minor concern)
Need for better guardrails – Users want ways to limit what Agent can change (auth systems, core files) without permission. (3+ users, Moderate concern)
7. User Sentiment & Engagement (Moderate Concern)
Frustration over marketing vs reality – Users feel livestreams and feature launches overpromise while costs, bugs, and risks are downplayed. (5+ users, Moderate concern)
Calls for user advisory group – Several suggest a beta-testing or feedback council to prevent “closed-room” decisions. (3+ users, Moderate concern)
Mixed feedback on Agent 3 – A few users praise improvements, but majority sentiment skews negative due to cost and stability issues. (2 positive vs 20+ negative)
Overall Sentiment
Pricing & Costs: Overwhelmingly negative (by far the biggest issue)
Agent 3 Reliability: Negative, especially vs Agent 2
UI/UX & Workspace Changes: Mixed but leaning negative
Trust & Transparency: Damaged by lack of clarity on costs and failures
the agent chat window used to have a small icon at the top which you could hover over to show the running total cost for the current chat. Removal of this has critically lowered my trust in Replit’s transparency over pricing and costs
I want the workspace to return to the ability to have files AND agent tabs on the left - or wherever I want them to be
I want workspace to remember the layout of my tabs for each app, so when I return to them, it is all as I left it - not reset each time. This is a breaking change from the old workspace, as it used to remember them
allowing an AI agent to run for long periods uninterrupted is a huge mistake. I find agent works best when given smaller surgical changes that the agent and I have worked out in detail during previous Planning chats (prior to Build+Plan modes, I would always do these planning chats before asking it to make the code changes)
if there is to be a “run forever” mode, 1) please make it optional; and 2) at least allow users to specify price caps
we must never remove a detailed step by step pair-programmer mode, to ensure users can take control over detailed and surgical plan/build work as and when they need
Spent the last 5 months building in Replit and it’s been overall a good experience. The introduction of Agent 3 has been (in my experience) a total disaster. The updates and changes honestly are starting to feel like a money grab rather than a user focused improvement. Currently very disappointed.
1. The speed is shocking, borderline unusable.
2. It often stalls and fails to complete
3. It will run a simple task for 8/10 times longer than Agent 2 and then fail.
4, The cost is now ridiculous (happy to pay if it works but it rarely does). Constant billing, transparency is lacking.
5. Yesterday it would not run properly at all and roll back was broken. Was charged $4.16 for 41mins of failed code. My work flow is almost at a standstill.
6. Reported to support that tasks were taking 13 - 22 mins for a basic change, and even they agreed from my screenshots that it was “not the expected performance”.
6. The UI/UX change… Why? It has not been improved, smells like an Apple move where someone has to justify their job by shaking things up that were already working.
I’m totally in agreement with what was said above, and they really need to fix the pricing right away so we can keep building without interruption. All the other bugs can be addressed afterward, but the pricing issue matters most to me, because without that being resolved, I can’t continue building effectively and moving forward.
Respect for pricing freedom – Replit has the right to charge what it wants; the issue is not cost, but unpredictability.
Volatile pricing – My app went from an estimated $3k–$5k build to $20k+ within 3 months due to sudden price changes.
Business risk – Small/medium businesses can’t budget when costs swing so dramatically; they risk wasting months of work and thousands of dollars.
Marketing vs reality – “Anyone can build anything” isn’t true under current pricing. Realistically, production-ready apps cost $10k–$20k+.
Transparency needed – Replit should be upfront about costs. It’s still cheaper than hiring developers, but misleading messaging risks reputational damage.
Core message – Replit is an enterprise-grade product; be clear about that so customers can make informed decisions.
Another point to consider: while this isn’t specifically about the current pricing issues, I’m wondering why I can’t purchase larger amounts of credits or tokens in a single transaction. What’s the reason Replit limits purchases to a ca $60 cap and doesn’t provide an option to buy more at once?
From an accounting standpoint, my accountant bills me for each individual transaction, so having to make multiple smaller purchases increases my overall administrative costs. It would be much more efficient if I could buy a larger amount in one go.
I would say it is now borderline whether it is cheaper to hire developers or not. If you use Chat Claude etc. to help you write very detailed and comprehensive business requirements and UI, my guesstimate is that replit would be cheaper, but not by enough to make it worth trying, given:
it doesn’t produce production quality code, so you only get an MVP anyway, and will eventually have to resort to humans
prices might be increased 5X again when you are midway through your build!
EDITED: I got carried away with my initial thought.
But if they reading this, they should know people will leave and find alternatives, and a someone said before, its a shame they do not contribute on their own forum.
Thanks Steve for pulling all of this together. Honestly, I’m really surprised that none of the folks from Replit have stepped in here to acknowledge or address these complaints.
I’ve been working on a project for a while now, and just yesterday I tried to add a single image to the library. It took three minutes and cost me $1.10. There’s no way back to Agent 2 (which wasn’t great either, but at least somewhat usable). I’m getting tired of paying for AI errors while watching my working code break again and again.
Fixing even the smallest things now takes way too much time and money. If this is the new standard, especially with the current pricing scheme, it’s simply too expensive for me to keep investing in this platform. I’m already looking at alternatives like Bolt or Lovable.
I’m still pretty new to this whole vibecoding space, but I’d love to hear from the community: what exit strategies are people considering? How are you planning migrations without losing months of work?
I’m also having mixed success with new agent, below is summary of my findings.
Agent3 is more capable but could benefit from more effective workflow to save excessive costs.
Issues:
Agent doesn’t plan actions and utilize parallel tool calls (multi_edit) enough for actions but makes i.e. small incremental changes to same file, causing extra unnecessary action.
Sub-agent delegation seems ineffective, better success when agent does the work itself.
Invoking architect (Opus 4 model) on changes that the agent could reason itself is expensive.
excessive success verification at the end of workflow.
When tested adding features to existing codebase and instructing with more cost-effective workflow policy to avoid delegation, task lists and suggesting the agent to do the work itself, the action count was dramatically reduces and most file edits were done in b atch.
I thought i was going crazy. Agent 3 has been a total disaster for me. We have been developing for months on a large project that is nearly finished but i’ve made almost ZERO progress since Agent 3 was deployed and the cost has been at least 3x. But i’m happy with the costs if it actually DOES WHAT IT SAYS IT IS DOING. Most of the time now it just goes to sleep and all my work is lost. It is a total gamble if it will continue working on anything. I don’t trust it to do basically anything at all. They are going to lose all credibility soon if they don’t change course, and i say that as a die hard power user.
It feels like product direction is geared towards “how do we monetize for the enterprise” instead of “how do we support the enterprise.” It’s a cart before the horse situation.
The testing feature would be highly useful, if it worked without overrides on authentication and other blockers for Agent.
Forks need to be able to share views of the database for a way to debug. Rollbacks are helpful, but there really needs a place where you can have access to all tools in a UAT type environment.
Agent going “off the rails”, giving it very specific documents or instructions and Agent takes the liberty to make changes outside of scope or doesn’t take the precise actions you tell it to do.
Prompting feels more like gambling than productive at this price point. Pricing transparency, predictability, Agent accuracy, and success rate don’t feel aligned/justifiable with the recent update.
Pricing adjustments need a heads up to customers. Had we known that we were about to be hit with a 20x increase with 90-day notice, we can plan and adjust accordingly. This inserted a ton of risk into our plans to develop applications and now causes big concerns for continuing to invest into the platform. Positive marketing spin does not cover this up.
Code needs to be more reusable with Agent. Two examples are 1) Consistent UI in different parts of application. In a single prompt, when asked for one Create, one Update, and one View of an object, Agent created 3 entirely different UIs even when prompted to show the same fields and layout. 2) Need some kind of way to manage front end, middleware and back end mappings. If changes happen to these naming conventions, Replit could do an update wherever that field/column is referenced in code. Something like a data dictionary or data mapping tool.
The pricing is 100% aligned with ‘Enterprise’, it will box out SMB and hobbyists. I feel shareholders are saying “we want to hear this” but they aren’t communicating that change.
We need improvements that help productionize our apps, getting more people to production helps build more revenue channels for Replit - they can collect margin off of the infrastructure (compute, storage, APIs built, etc).
UI needs to support developer experience equally as Agent takes up way too much screen equity.
Positives:
Longer run times do help when building bigger product features and debugging.
This is v1 of automated testing. When safe guards are built into it, this feature is incredibly valuable.
There was an Agent interaction where Agent was “testing for optimization” - if this is actually true (overconfidence of Agent has me reluctant to believe some things it says), it gives me more hope that when I hit deploy we aren’t going to be eating cloud credits with providers due to unoptimized code.
It does seem like Agent is more accurate as the context and run sizes are bigger (as a caveat I also saw it have too much power to make damaging changes without guard rails).
In terms of pricing on mobile app the cost of the overall agent session is now missing (unless it’s been moved to an obscure place) you can only see what per prompt costs. You have to navigate to accounts page, login and look through the summary (but still cannot see each session cost) not in rollback / history either ! This is shown in the usage summary within the app as per screenshot, although the agent session breakdown is no longer in here! Bastards!!!
On the whole, I don’t have many concerns with Replit however when something in Replit breaks I need support urgently. I have two issues that have been outstanding for 2 weeks without a response from support.
The first issue is that I cannot connect my Repl to Github and just get an error from Replit. The second is that I can no longer merge forks into the main Repl, again just getting an error. Both of these are critical to my business and I just cannot spend $20,000 per year for an enterprise license to get dedicated support.
Regarding the Github issue I also experience this issue a few times this year.
To solve this issue just disconnet your Github account and reconnect again under your Account settings. It should solve the issue.
P.S I mentioned this to support a few times and they never responded
Thanks @adrianvonbausse. I’ve tried this in the past (and against just now) but just get the message “There was an unrecognized fatal error with Git. This is probably a bug in the app.“.
It’s very frustrating. Replit is a fantastic bit of kit but support is nowhere to be seen. Raises some big questions about sticking with them.
These are strong points and I tend to agree with all of them. The total price for a chat was a surprise for me. That was so convenient to see the running total in real time. Seems like a removing a big trust signal there.
I asked Michele about this recently and the reply was that the breakdown of pricing was provided after each check point, but I agree it is not as convenient as the total in the header.
I also get annoyed at how the new agent collapses chain of thought messages. Even after I click them open.
Seems like small modifications, but for people who use the app daily they are fairly big frustrations.
Aside from all the pricing changes, the performance of Agent 3 is completely horrific. I can’t get anything to update or complete now, it just hangs and i end up waiting 10-15 minutes for responses that then lose context and don’t do anything, and i’m not even sure if its charging me or not for not accomplishing even a response. It’s utterly unacceptable to force a change like this while people are in the middle of building projects. They desperately need to allow us to enable agent 2 until this gets fixed, I have spent thousands of dollars with replit and this is completely unprofessional way of rolling out changes out of no where with no way of going back. @kody-replit@matt-replit you guys need to pass on to the dev team to allow use of Agent 2 until this stuff is fixed, i’m just stuck in the mud right now with a deadline
