So I know the screenshot is kind of not showing much but its actually showing a lot.
I got this response like at least 4 times, consuming at least 5usd worth of tokens (1.5 more less every time) and in the end it got stuck every single time.
It does not give you a button to carry on, it does not give you back the amount spent, it does not even leave the work done so far.
With previous models in which I could choose from, working with tasks that I used to work with will get challenging to say the least.
I am thankful though since I got some credits to try it, so its actually not that much of a complaint but i would like to keep working with v0 and Vercel in the way it used to be before, or at least to remodel the agent in a way that it actually adds the code it was working on!
To clarify, i asked to create 10 replicas of a form and success page. 20 components might seem excessive, but they are all replicas from each other and it was something that the previous models could handle easily arriving at about 10-15 components created at the same time, without spending as much tokens, credits and without making mistakes (or not even delivering the code created so far!)
If there is any other way to create code with this model in a way that it wont stop working and implement it? I liked some of the changes, but it could be problematic on a potential bigger scale (in my case, that i handle around 10 corporate apps/accounts). This might drive me (and possible much many others) to pass on some of these new features if there is no possibility to effectively consume the product that self deletes or stops the code that was requested (and paid for)
So, big fan of v0, I have been using it extensively since last year. It is great, especially for someone that comes from design. It always seemed to me that v0 was the superior ādesignerā out there.
Now adding the Agent mode is a welcomed addition, it has great potential and clearly this is the way. But. As it stands something is terribly wrong, either agents configuration, model, whatever. The main issue is that the agent mode is much more costly, which would be fine if the output was high quality. But itās not, I have completely burned through the gifted credits yesterday and today. Today especially was the worst experience, burned the $20 and barley got what I asked for, in fact is about 50% of what I asked extensively even providing documentation and explicit requests. Design is the worst it has ever been, but also it lacks any kind of project awareness, always defaulting to basic shadcn components instead of using the custom global components already present in the project. I was used to stop explaining basic UI or Development good practices as it would at least cover that, but now it just seems like Iām working with the older models, where every steps was a chore but we didnāt have anything better.
Bottom line:
Agent mode = cost more provides less. Not a great combination. While I believe it is a necessary step in the right direction, it needs a bit more time in the oven.
This update has negatively impacted a project that was in progress. The agent mode is far inferior to what it was before; now I have to spend $10 in credits to get a result that I would have gotten for $0.10 with the previous version. This update needs to be reverted urgently.
Please add the option to switch between modes. The current Agent mode feels unfinished and inefficient. Even if it were fully functional, I would still want the flexibility to control and use other models. The ability to edit specific elements in real time was an excellent feature and should not have been removed. I think this community enjoys autonomy and this clearly takes it away. I hope this doesnt turn into squarespacish where stay at home moms will be āvibe codingā their next MLM offering
Performs deep code review and finds issues that would otherwise go unnoticed
Less overhead on our part to decide what would be suitable for which task
Ability to reproduce design from another website sounds amazing, havenāt tried it though
Good suggestions on how to move forward with the project
Cons:
Works on changes that break the system (instead it could simply suggest and then proceed if user confirms)
Creates duplicate files that already exist and on top of it, copies the same codeā¦
Initially creates a separate project referencing the existing project, probably a version difference issue (creates app at root instead of src).
Burns through credit a lot more due to lack of choice in cases where we actually know which model would have been better to restrict output size
Extensive research material in the response, everything in it is important but would be better if it was shortened
Suggestions:
Respect usersā opinion to have a model selector and an agent toggler that would allow user to control the output and whether the process should be totally automated or semi-manual.
Context window seems to be smaller which possible ignores certain files and duplicates them
Improve the agentās code review knowledge by having it regressively check if the file changes and code it has planned align with the userās request while also ensuring that it hasnāt already been implemented in some other file with same or different name across the project
Add buttons in the response window that allows reverting file or cancelling changes
Iāve been using v0 for a while now and have really valued how it makes developing my projects faster, easier, and just generally more productive. When the option to select GPT-5 became available, the jump in capability was remarkable and it completely transformed what I could accomplish in a short time and with minimal effort.
My experience with agent mode has been the complete opposite - this update is breaking my workflow and making it impossible to continue my project with v0. It does not produce what I ask for, even with explicit, clear instructions. It frequently claims to have implemented changes it hasnāt, makes significantly more errors, ignores best practices that GPT-5 followed without much guidance, takes far longer to produce results, delivers much poorer overall quality in the UI and in the code itself, and costs more in both time and money.
At this point, v0 is simply not producing the quality or accuracy I was getting with GPT-5, or even earlier versions, and I cannot move forward with my work under these conditions. It is wasting my time and is putting me at risk of missing deadlines.
I request an immediate option to disable agent mode and select GPT-5 directly again. If this is not resolved, I am a day or two away from having to migrate to another platform and request a refund.
Guys, PLEASE REVERT BACK!!! This Agent mode is god awful!
Your previous version of being able to select small, medium, large ect was soo good! It literally shaved about 3 weeks of dev time for a recent UI implementation needed on a project.
Agent mode on the other hand has been the complete opposite, its like going back to 2022 again where the it takes so much time to prompt and nothing changes. Now I am spending more time trying to get the agent to do things when it would be far easier for me to code it. I am finding that I am having to step in and make far more manual changes myself too.
So far, really poor experience, you guys had it nailed
I thought the problem was with me. The previous version understood everything better and did everything perfectly. But this update canāt do anything right. Itās a complete disaster!
tl:dr; Agent mode does not work. after about 40 iterations trying to do simple things I could previously do easily I am forced to cancel the subscription and move to a more stable tool.
I would like to contribute and back other comments in this thread.
In the last two days. since the release of Agent mode I was not able to make a single prompt to do my I intend it to do.
Thatās after about 700 prompts to the same project which worked decently and was providing me great value.
At this moment, I am attempting to migrate to another tool as it simply does not work.
In addtion, it now burns through credits like crazy, In about 1h I was able to spend what usually I spend in 2-3 days. I used to spend about 3-5 dollars a day (5-6h)
while now I was able to spend 8$ in less than an hour
Unfortunately I just canceled the subscription as I donāt see a way for me to continue developing using the tool. I cant stress it enough, It does not work, at all!
The idea is good, but it was much better before this update, I have been wasting a huge amount of time on a project for 2 days, the model takes too much initiative on code changes, it makes incomplete code, I have a response that lasted more than 5 minutes and took 2ā¬ from my wallet, for a code reduced by 80% (which does not work). Horrible for professional use, I request a return to v0.dev, where I would be forced to change the AI model. You are killing your own AI and itās very very sad
I miss GPT-5, it solved bugs that all 3 other llm modes werenāt able to solve for 48 hours. For a brief period, I was able to make real progress. Now with agent mode, itās back to being dumb and making the same mistakes over and over again.
The idea of Agent Mode is great, and I appreciate that you are working on making v0 better and better. However, this update needed more work before being released.
First of all, the quality of the responses has suffered a major downgrade. I used to use v0 to develop Django templates (HTML), and it was perfect. I would send a file as a sample and specify that I needed the same thing done on another file but with different data. The response was great, and with a few small iterations, I had the page ready for production. Now, the output is a disaster ā it breaks the format, invents new blocks that donāt exist, and fails to follow the design pattern of the sample (both in code and UI design).
Also, Agent Mode tries to āreviewā the work. With a new prompt, it reviews it once and shows me the code, but if I ask for a fix, it reviews the same thing five times and writes the same code five times. This also multiplies the billing by five. Luckily, Iām using gifted credits, but for any prompt that would normally cost 15ā20 cents, it now costs 60ā80 cents (though this issue doesnāt happen with new chats ā new chat charges are correct).
Please roll back this change. I canāt use v0 anymore with this new version.
I agree with pretty much everyone here. This needs to be rolled back or give us the option.
In 2 days Iāve exhausted my monthly credits with the new āmodelā making tons of changes I never requested or hinted at. Iāve had to force it to revert meaning itās costing me more. Itās a relatively poor role out as the intended features like checking itās own workā¦clearly not functioning properly.
This is the worst update ever made. All my projects cannot be continued from now! this is getting worse!!! Agent mode is destroying all projects and dont get any ideas how to fix my problems!!! I already cancelled my subscription
