Brutally Honest Replit Review (Not Sponsored)

So this is going to be a very honest review of  Replit, especially their agent version three,   which they released about two days ago. And this  is their brand new vibe coding agent that a lot of   people have been using and talking about recently.   They also raised $250 million a few days ago,   which a lot of people are talking about. Which  also means that they're going to be reaching out   more to influencers and like YouTubers and so  forth and trying to bias their videos by doing   sponsorships with them. They have actually  reached out to me before over here and then   I just said thank you for the email, but I don't  accept sponsors. So this video is not sponsored   by them. Basically I'm going to pay for it myself  and see how good it actually is and making useful   web applications. So the thing that I'm going  to try and make with Replit in this video is

a tool that I really want myself because I've  been doing a lot of manual work to like achieve   the same result. And basically the motivation for  this tool is that AI SEO has gotten quite popular   and basically AI gets most of its information from  Reddit, Wikipedia, YouTube and so forth when it's   like giving answers in its search results. And  what that means for me personally, because I make   YouTube videos, is I want my YouTube videos to  appear in like AI search results. Whenever someone   like asks AI question, hopefully one of my videos  shows up. And the way they do this with videos   is that they look through the transcript of the  video to make sure like whatever is being talked   about is the thing that the user searched for. But  the YouTube auto transcription is often wrong and   it like misspells company names and other things  which is bad for wanting to appear in the search   results. So what I do myself is I use AssemblyAI  and then I go over here, press transcribe new   file. I then choose the audio of the video that I  want transcribed. So I choose like this over here   and then it transcribes the video as shown over  here. But the thing is, it's still like wrong in   the sense like it says CodeX instead of CodeX.   There's a lot of typos still in the video. And   the way to resolve these typos is actually to then  take this transcript, which is more accurate than   the YouTube transcript, and then give it to Gemini  2. 5 Pro. So I give the video and the transcript. I   say list out all the typos. So it lists the typos  out here based on the like visual elements of the   video and then it gives a corrected transcript  over here. And then I upload this to YouTube and   my videos appear in the search results better and  like ChatGPT results better because the transcript   is actually really accurate. Now this is like a  manual process that I've been doing so far. This   is what I want to make with Replit. So it's going  to be more useful than the random to do apps that   you see other people making. So firstly, I'm going  to upgrade to Replit Core over here. It is $25 a

month, so we're going to press continue here. So  I just paid for Replit. It is $30 if I include

tax. And I'm going to describe the application  that I actually want. So instead of typing,   I'm going to use my tool HyperWhisper. There's a  coupon code down below if you also want to use it.    And I'm basically going to say, hey, so I want you  to make an application where I can upload a video   file and then from the video file it extracts  the audio, it sends the audio to AssemblyAI to   transcribe everything that's being said and then  it passes the transcript alongside the video file   to Gemini 2. 5 Pro and it tells Gemini 2. 5 Pro to  basically correct any of the typos that are in   the transcript and then allow me to download the  transcript again as a. txt file that I can then   upload onto YouTube. The user interface should  be pretty simple, just a basic like upload video   file and then it should show me the progress of  like roughly what stage it's currently on. And   also check with the AssemblyAI docs and also check  with the Gemini 2. 5 Pro docs to understand how it   should work and fit together. So you can see it's  transcribing everything that I just said and it   pasted into here. So just to be sure, I'm gonna  paste in the URL for AssemblyAI copy and paste   it over here and then press Start chat. And now  it says Creating application. So we'll see what

comes up with. And it seems that it's searching  through Replit integrations and now you can see   has integrations for Gemini AI and app storage.   And it basically gave me this plan over here. So   like I've made applications before. Vibe coding  was a big thing and this plan looks pretty good   to me and it's pretty much the same thing that I  would do as well. So I'm going to press build the   entire application instead of doing start with a  design and then I'll just fast forward through the   building part so I'll press start building here.   Hey so as a short aside, if you're interested in   learning more about vibe coding and vibe marketing  as well and all the best tools that are available   and being part of community where people are  actively discussing these ideas such as like   what tech stack you should be using, how you  should get customers and paying customers for   your applications, and you also want to learn from  me and get some mentorship from me as well, then   I offer all of that in my AI startup school. There  will be a link down below if you are interested in   joining it. And many people have already found it  pretty helpful because if I go to this testimonial   page over here you can see that many people have  already managed to get their first users, landing   deals with clients and also make money from their  mobile applications and web applications all over   here. There will be a link in the description down  below if you are interested. So do check it out.    And now it should be coming to life. Okay,  so now after about four minutes of work it stopped

and asked me for a Gemini API key. So I'm going to  go and create one for myself. And now I can press   continue here, but it seems that there's another  error happening here where I need to have an   Assembly API key of course, but it didn't prompt  me to add one. So I actually don't know where to   add this API key. So I'm gonna press Debug with  AI and it seems what happens is when you press   that then it basically makes a new chat and I  can go to a previous chat which was doing the   like main work over here. Now unfortunately it's  a bit slow because it's going for the entire code   base just to ask me like hey, do you want to add  the AssemblyAI API key? So actually I'm just going   to press stop and then go back to the previous  chat and it seems I can't actually go back to   a previous chat which was going to make the entire  project. So that's annoying. Anyways, I'm going to   manually add this Assembly API key to my account  secrets over here and then copy the key over,   paste it into here and save and I can say continue  with the project plan. And I guess the nice thing   is that like when I click on a piece of work over  here thing is that it shows me how much like it   actually used when I was doing that piece of work.   But the fact that I wasted 80 cents in credits for   this Assembly API environment variable thing that  I was trying to fix with it is pretty bad when I   could've just gone here and added it to my account  secrets. And hopefully it continues with the plan   properly, even though this is a brand new chat.   So whilst we're waiting, they also seem to have a

High Power model mode here which is five times the  normal cost. If I go to learn more, it basically   upgrades the agent from using Claude Sonnet to  Claude Opus, which is of course, I think about   five times more expensive as well. Reading through  the page it does seem to be information about cost   transparency, which is quite good because like, a  lot of people can end up wasting a lot of money on   vibe coding tools if they don't exactly know what  they're doing. So now, after about five minutes

of work, it's done, taking about $1. 09 and I can  upload a video. But it says up to 500 megabytes.    Unfortunately, the video I had planned to upload  is actually 1. 72 gigabytes. So let me see if that   actually still works. Maybe press open and it  says it has to be smaller than 500 megabytes. So   I'll say make the video size limit 5 gigabytes.   And now that task is also done, taking about 25   cents. But anyway, let's test it out. So upload  this video. And it still says it has to be smaller   than 500 megabytes. The bottom right corner still  says, please upload a video file smaller than 500   megabytes. Okay, now that task took 32 cents.   Let's try one more time to upload here. And I'm   clicking the file. There's no actual indication  that it's being uploaded. Let me try with a   different file here. Interesting. So it suddenly  changed, meaning that it was being uploaded in   the background somehow, but didn't give me like  an indication it was being uploaded. But anyway,   let's press stop processing. Let me refresh a page  and try this one more time. Okay, select this file   here. Stop processing. No video file uploaded. So  it's not uploading the file. Hey, so I get error   no video file uploaded for some reason. And then  press Enter and see what's happening now. Okay,   so now it made another change and it claims that  it should be working now. So that took 48 cents.    And let me refresh the page here. All right,  let me check my activity monitor to make sure   it's actually uploading. So I can go to activity  monitor here. And you can see I am sending data at   like 14, 17 megabytes per second. And now I get a  500 error file too large over here. So even though   I told it to make the file thing smaller before,  it didn't do that. So I'm just checking online   if there's some kind of like file upload limit,  but it seems that there should not be because it   says over here that they allow for 256 gigabytes  of storage space. So I can say I get a 500 error   saying that the file is too large. The file that  I'm uploading is only about 1. 6-1. 7 gigabytes.    So now let's hope that it fixes this error now.   But I'm actually curious by clicking around and   seeing like what are the dev tools here? So it has  a console network. This is a network request that   failed over here. So this is kind of like Chrome.   It's running a version of Chrome in the cloud or   it has all the same properties as Chrome. I can  turn off the dev tools and it basically made this   fix after using 61 cents. So let me refresh a page  and try uploading it one more time and then let's   see our network to make sure it's uploading.   And you can see it says file too large again   over here after it's finished uploading. So like  maybe I can turn on this High Power model mode   here to have it fix the error and say I'm getting  the same error again and press Enter. And I think   this is like maybe a hard built in limit that it's  actually not aware of because the previous coding   agent that I was using when not having High Power  mode on was just not aware of this limit. Since   this is running on Replit. This may be a platform  specific issue. Let me check the documentation. So   it's searching online for the Replit documentation  itself because for some reason it does not have   it built in perhaps. So what it decided to do is  just restart the server because the documentation   doesn't specify it. So let me try one more time.   Upload here, start processing. Okay, sweet. So it   says upload successful file has been uploaded and  now it says spawn FFMPEG end of whatever. Failed,   try again. And it like doesn't keep the file on  the cloud. So I have to upload it again after   it fixes this particular error. So I'll just paste  the error into here. And bear in mind that this is   now using the High Power model. Okay, now it fixed  the FFMPEG issue for about $3, which is insane.    So I'll turn off the max mode over here, the High  Power model, and then try that one more time and   then do to upload select this, start processing.   Okay, now you can see it's been uploaded again and   it says ffmpeg spawn failed. So it's weird because  it does say that it's installed FFMPEG over here.    So maybe let me like exit out of everything, close  all these tabs and go back to the application.

Actually let me press publish and see if this  will work. So it's still bundling everything   together. But I guess it's nice because I can see  logs over here. I guess once it's done, I can see   analytics of like who's coming to website and from  what countries and browsers and then I can buy my   own domain as well maybe. So I'm going to go to  a convenience store and then come back and see   how it's doing. Okay, so I just came back and it  seems that after building for about six minutes,   it says the deployment failed to initialize or I  should contact Replit support. And basically if   I try to look for the logs, if it will load, then  it just says something went wrong, nothing in the   logs. And I guess the only thing that I can do is  press debug with agent. But I kind of had an idea

whilst I was at a convenience store, whether  on the preview version if I upload a smaller   file. So for example, File that's  like 50 megabytes, so will that work instead, 65   megabytes here, press open, start processing and  I still get the same error of spawn ffmpeg like   not working. So I'm just gonna look for the rest  of the application whilst I'm waiting for this

to like be fixed. And it seems that there are a  bunch of integrations over here, like for Spotify,   Dropbox, OneDrive and so forth in usage. Over here  I can see how much I've used. So you can see so   far that I've used about $12. 32 in agent usage  without actually getting a working application,   unfortunately. And then you're charged for all  these other things as well, like app storage   operations, app storage, data transfer. And now  the strange thing is over here it says that I'm   on the starter plan and that I should upgrade to  Replit Core when I press this, it says that I'm on   the Replit Core plan for your current plan here.   So that needs fixing. But yeah, anyway, let's

see if this fix that it's coming up with actually  works. And it says the page is unresponsive now,   so I should have to exit the page or wait and  let's try again. And despite refreshing the page,   it still seems to be working in the background,  which is good at least because it means that,   like, it didn't stop working when I had to refresh  a page. Okay, now after about eight minutes of   working, it applied these fixes over here and  it took about $1. So let's try republishing   the application by pressing republish over here.   Okay, so the publishing actually worked. So we'll   open up over here, choose the smaller file, press  start processing, and hopefully this should work.    And it says request entity too large over here.   So it seems that in production I'm getting this   error, but if I do in the preview mode instead,  then I think I get the FFMPEG error and I get   the FFMPEG error. So in development or in preview  it actually uploads, but in production it doesn't   upload because it's like too big. Okay, so let's  give it one more shot before calling it quits.

So I'm gonna turn on the maximum autonomy mode and  the High Power mode and have it fix this FFMPEG   error because, like, that's more important.   I want it to work in preview mode at least so   I can press Enter. Okay, so now it claims to be  done again after using $3. 68 for that job. We'll   refresh it over here, and then after refreshing,  we can select the file again, start processing,   and it gives the exact same error. So basically,  like, the publishing does work, except the file   uploading doesn't work. Like at this endpoint  over here, it tried to fix this twice. It

failed both times. Maybe because it thinks that  FFMPEG should be able to work like with Replit,   but it just doesn't work with Replit. And  if I go to my overall cost by going to home,   going to usage over here, then basically it  seems that I used $13. 34 in credits out of 25,   and I basically don't have a working solution.   Or for example, I could have used Claude code

for $20 a month or ChatGPT and used CodeX CLI with  ChatGPT for $20 a month. And I would have achieved   like actual working solution. And it would be  much cheaper as well in terms of getting tasks   done with Claude code and ChatGPT's CodeX because  they actually own the product. So this is kind of   what's happening. You have Anthropic who are  selling to Replit via their API, so Claude   Sonnet and Claude Opus, and they're charging a  markup because they want to make money on the API   and then you have Replit who want to make money on  you. So they're charging another markup and that's   because you're going through like another product.   Whereas if you go directly to the provider,   Claude Code is made by Anthropic, which is  why they're in the same color and Anthropic   is only charging a markup once and they're not  charging a markup themselves over here there's no   markup to Claude Code via Claude, there's just a  smaller markup over here. And actually I'm pretty   sure Anthropic are also making a loss on Claude  Code because they raised $13 billion recently,   which is much more than Replit is even valued  at. So clearly they're subsidising their own   tokens the most for their own products over here.   So this is probably subsidized Claude Code. And I   know this is the case because if I go to my own  Claude Code and I type in npx ccusage here, then   I can see that over the last month I've actually  used $3,000 of Claude Code usage in the last month   all over here. Whereas I basically only pay  about $200 a month for that plan. So I'm getting   $3,000 of usage for $200 a month, which is like  a ridiculous deal. And Replit are like physically   unable to offer that because they don't have their  own models and they're being charged a markup now,

even though this application did not work,  despite me giving the errors back to Replit,   I think that basically if you're trying to make  something completely new that hasn't really been   done before, for example, what I'm doing over  here, there's not a service that does exactly   this. If you're trying to do something completely  new, then it's probably better that you use Claude   Code or CodeX CLI instead because you will have an  easier time in making something new. Because for   Replit at least they're making a really generic  environment for you to be able to make a pretty   generic app. And that app probably already exists  somewhere else. So you see people making to do   apps or like project manager apps or whatever.   You can try and use Replit to make one and maybe   you'll get there, but it would be better if you  just use an existing like to do manager app,   unless you can't find that specific application  out there already. But chances are you won't be   able to make that with Replit like I can't do over  here. So in conclusion, I think that if you're   trying to build something completely new that  hasn't been done before, chances are you won't   be able to do with Replit, so you should not waste  your money. But if you're trying to make something   that has been done and you know it's been done  on Replit, but you want like the added bit of   customization or something and you have like more  money to burn because you don't mind the markup   and you want the easier process of deploying  things and so forth, then you should probably use   Replit. But it is worth bearing in mind that you  will probably end up paying way more to get the   same result because you're not being subsidized as  heavily via Replit because they don't have their   own foundation models. But if you actually do want  to learn how to make something completely new and   use the right vibe coding tools for this, then  I cover a lot to do with that in my AI startup   school over here I basically cover a lot to do  with vibe coding what the best models are which   are completely unbiased because I'm trying out  every single model and not being sponsored by   them. And I also cover like various vibe marketing  strategies so that once you've built your products   you can actually get in front of users and earn  paying customers and so forth. Anyways, do check   it out if you're interested. There will be a link  down below. And if you want to learn how to use   CodeX CLI or Claude Code at least then I do have a  bunch of videos on my channel about this as well.