Pick a product you use often and vibe code an MVP - Course.mp4

Well, excited to have this conversation and do a vibe coding case with you. How does that sound? >> Yeah, sounds good, man. >> Awesome. Well, in typical sort of case frameworks, I'd be curious like can you tell me about a a product and a problem space product or something of that order that you're passionate about that you'd be interested in exploring in this case? Yeah. So, um I always care about, you Know, user experience and um one of the most I think just annoying user experiences is when you are trying to get data into a

new application, especially when you're onboarding and you're trying to just get that application to start working and you have to put data in. Um, and so, um, one of the things that I see a lot is just like even just as easy as like a CSV uploader, right? Many apps will like they'll say, "Here's your template. You Have to download our template. You got to move all your data into that." And then so you're spending all this time just to like get your data ready just so you can upload to to their system. And so

I thought maybe we could uh do something with that. >> Cool. Can you tell me um I like the idea. It feels kind of broad. Can you tell me how you might like who is the scoping of it? Users, is there a particular domain you're thinking about? Is it just data data sort of users in general? >> Well, yeah. So, think about like um I worked in martekch, right? And marketing users, they're not usually super technical. They often are having to deal with spreadsheets. They don't want to necessarily have to deal with data types and

having to figure all of that out, importing and exporting. They do a lot of it, but you know, they're more worried about getting the data into the System so that they can go do their job, right? And so if we can make the technical part of the import and the export a lot easier so that they can just bring the data as they have it and then the system can make it easier for them, that that's the thing I'm thinking about. >> Fascinating. Um, and I guess just to double click a little more, any thoughts

on like the segmentation of this? Is this for marketers in general? Is it Like a marketer who doesn't have an analyst? Any sort of thinking about that? Like who your who you think your audience is and how many users there might be there? >> Yeah. So, you know, let's go with like I used to work at an email company, right? And so like let's just think about we can just go easy on this one. And it's like it's an email list with some basic persona data, right? So they're going to upload an email list for

a segment of Their users that they want to, you know, send out an email campaign to. And so we can do something easy with that. >> Cool. And what what do they currently do? I know you think you painted a picture. It hurts today. Like what do they current do? >> Uh so a lot of them will they'll have their like email collection form. They'll use any number of systems for that. Sometimes it's a spreadsheet that they you know collected emails at a Conference or you know one of their salespeople collected it. Sometimes this goes

into a HubSpot. They may or may not have that connected up to their email system. You know, maybe they have a web collection form that they set up, you know, some sort of ClickFunnels thing or or some app somewhere or something on a website. They maybe made a landing page somewhere in some other app. They, you know, it's going through Zapier to a spreadsheet. Who knows? They've got all These things. They're all disconnected. um they're pulling these things down and now they have to go and put it into their email system which could be you

know send grid mailchimp again HubSpot could be anywhere and and they need to upload it and and now they've got to map these fields that were defined by somebody to this system that is pretty prescriptive and they have to get it in there and you know they want to go get these emails out because sometimes the Timing matters. And so they don't really have all day to go and mess around with all these fields. >> Well, I think we think we could probably move on to the next section. Thinking about like assumptions and constraints. Can

you kind of walk me through any assumptions you might be making as you think about your approach to this? >> Um, I mean, I'm going to make the assumption that they've got the data in Uh spreadsheet or in a CSV file. Like they know how to get a CSV file out, right? Because Google Drive can do that, Excel can do that. that's pretty easy. I think I'm going to make the assumption that they're using a web browser that they can upload and that the system that they're using can upload. Um, and that they're, you know,

they're technical enough to be able to handle that. And I think for the purpose of, you know, for this use case, I think I'm Going to make some assumptions that like they're not using a screen reader that, you know, they're able to use a mouse and a keyboard. But there are things that we could do if we wanted to to build that in uh and make sure that you know we were VPAT W CAD compliant all of those sorts of things. >> So you're getting at like accessibility. >> Yeah. You know with some you know

if we wanted to we we could go there. It would take a little bit longer but but we Could definitely get there. >> Yeah. I'd be comfortable saying I'd be comfortable saying this is for the nonaccessible starting point for sure. Yeah. That's a great point. Yeah. Yeah. Yeah. any thoughts on like privacy or security on your mind when you think about this? >> I mean, always like in a production environment, I think for something like this, like for a shortened timeline, we would need to, you know, we're probably Going to do a proof of concept.

Probably not going to actually like hook this up to a database, probably just worry about the the front-end processing of the data and being able to like match fields and that sort of a thing. But if I was going to production, I would want to make sure that, you know, we were uh because we were doing this in an application, they were doing this with an email app, they would already have a login. We would probably want to make sure that um you Know, when they upload the file that we encrypted it at rest, that

all the data was flying over the wire encrypted. if they were uploading, you know, personal identifiable information. Obviously, we have emails. So, we're going to want to make sure that um you know that we're encrypting that and storing it in a way that's compliant with their geographic zone. If we're working with customers that are in what, California, Australia, Canada, Europe, all the places that have Compliance zones that we're covering that um you know, and making sure that their account is set up for the right zone for storage and for uh you know, for over the

wire flight. All of those things are really important, but that that's going to be more in a production environment than >> Yeah. >> today. Then I'm curious like thinking about constraints like you mentioned, I've I've lived in like no code Ecosystems with Zapier and stuff like that. Are there any constraints on your mind of where you might not try to solve for things or where you might limit what you're endeavoring to do? >> Oh, yeah. Yeah, that's a great point. I think uh for right now, I mean, I'm probably not going to Let's see. We're

talking about spreadsheets, so you know, I'm I'm living in the world of spreadsheets, but the like you could Put a whole document in a single cell. So I'm making the there are some assumptions about like what you can export to a CSV and so those assumptions are there in that like it's going to be you know UTFA textbased content probably and there'll probably be some you know in in a production environment we're probably going to have some like text constraints for like different field storage and things like that. Um and We'd probably want to put

those constraints on the fields when we went to a production environment. For this test case, I'm probably gonna stick to, you know, uh, smaller fields just to show what we're doing. But yeah, we would probably want to do that like in a if we were gonna actually take this thing alive. >> Cool. And then before I think next we'd get into flushing out requirements and your perspective on that. But I'm just Curious, any risks on your mind with something like this? Anything you'd want to kind of watch out for? Yeah, you'd want to, you know,

make sure that you're not allowing people to upload anything other than certain file extensions or um, you know, because zip files are always fun so that you can't upload zip files. Um, file size constraints, so somebody's not uploading like a two terbyte file that has a, you know, CSV extension. Yeah, I mean, there there are definitely things like that with file uploads and and those are pretty standard things. Um, let's see what else would we want to do. Um, you know, you probably don't want to allow for like you you don't want to allow any

path for like code execution on the file name or anything like that. So, you know, you're going to have the normal um probably out of the box use a file up open source file upload, you know, uh package that's Well um that's well known so that you know uh all of those sorts of things are already taken care of. Uh, cool. Sounds great up to this point. How about we jump into defining the requirements and how you might measure success. >> Cool. So, yeah, typically what I do at this point is um especially like with

the vibe coding thing for like a proof of concept. If if it's cool, share my screen. I'll show you what I do. >> Please. >> Yeah. So, I'm going to jump in. I typically just go straight to chat GPT um because I'm gonna just start um just riffing with chat GBT to sort of get my ideas out and ask it to sort of you know build out a uh like a requirements doc for me. >> Um part of this is because you know I'm I'm we're vi coding so I can be a little bit lazy.

they don't need to write out the whole document, but also because um It's going to be a lot more efficient and then I can start to actually ask it to um you know find some holes, do some research for me, that sort of thing. Um makes me a lot more efficient. So, um from here um actually we can we can actually even cheat a little bit further. Um let me get Whisper Flow out. Um which is a pretty cool app that I have for my computer. Um, we could use the chat GBT one, but I

like this one, too. Um, see if it's working. Test, Test, test. All right, perfect. Um, all right. So, uh, Whisper Flow just lets me hold a button and talk. Um, and so I'm going to do that for a second. Um, I want to build a a web-based CSV uploader. Uh, we'll probably use Bolt.AI AI to actually um do the initial coding for us um using I don't know Nex.js uh Tailwind CSS and open- source packages. Um if you can help me find the um open source packages for that that would be Great. What we want

to do is have um an email contact list uploader for CSV files. Uh this will allow the user to take a or to upload a CSV file um through a web UI that they have downloaded from uh their application. It will have um various fields in it like email address, first name, last name um and then persona based fields that will match uh whatever it is that they are sending people emails about. Um, we'll get that going real quick. Make sure that it gets that in there. Um, [clears throat] so now let's see. We want

to make sure that we've got the email upload form, that we've got the ability to match the fields. Um, okay. So, from here, what I want to do is start thinking about the sort of um basic use case things here. And so the things that I really don't like about CSV uploaders is that they're Pretty they tend to be pretty strict where you know somebody will give you the um the template and you have to match it perfectly. Um so what I want to do is I want to build this into the prompt where um

we will tell the user we'll give them you know an example CSV that they can download and they can match but I want the uploader to actually give them a good experience. >> And so um I actually want to do some of the matching for them. Um and so this is Where things can be really interesting. Um, so I'm gonna start talking to it again and give it some more um, information. Um, with the uploader, one of the things that I want to do is I want to make the uploader slightly smart in that when

the user uploads their CSV, I want to try to match the fields from their CSV to our uh, CSV fields. Meaning that if our field is something like email address, first underscore name, last underscore Name that we try to match to the fields that the user has in their CSV. So their field might be email or fname, lame, first, last, things like that. We should try to match to those things. Then when we show the user in the next stage, we can give them a field mapping UI where we show them the fields that we're

expecting. We show the fields that they have and we allow them to create a map from their fields to our fields Including personabased fields where we will allow them to um create personabased fields that are arbitrary for our system from their CSV. This will allow them to um insert those fields from their CSV into our system. Now, our system will be just a prototype. So, we're not going to have a database on the back end. Um this will be in memory for this proof of concept, but we want to show them what This looks like

for the proof of concept. So, um, we need to make this work for the session of the user in this proof of concept. >> Cool. Is this like the line in the sand for your your proof of concept or your MVP? >> Um, I think there's one more piece, um, which is that once it's uploaded, I do want to show them the data in the UI. >> Got it. >> Um, okay. So for the last piece after They've mapped their fields the um the last view is that we want to actually show them the data

in the UI to prove that they have actually uploaded their CSV into our system. Um this again will be in memory and uh will just be part of this user session in the app. >> Cool. Well, can you describe any just think I use the word line in the sand, but um anything you're explicitly trying not to include in this through the lens of a PC? Like what is the least you can Do here? Anything that you you have ideas you have that you might not be including right now? Um yeah actually so um one

of the things that I have always wanted to do in a CSV uploader um and like if we have time it would be really cool to try to do it um is that if there are rows that fail, I would love to actually have the app capture those rows um like in a side object from the CSV um process all the rows that work Completely independently and then show the user a UI of the failed rows and allow them to fix them as part of the upload because so many CSV uploads will actually do it

as a transaction where it's either everything succeeds or everything failed and I want them to be able to fix it in line as part of the upload process and then save those failed rows. I just think that would be such a cool experience. Um I don't know if we'll have time but it'd Be really cool to be able to do that. Do you think this would be usable without that? >> Um, I think what we need to do is uh potentially probably what we need to do is have it do the normal thing of if

the CSV is no good, fail out of the box. Um, but in our test case CSV, we'll probably have it use a perfect CSV the first time. Um, and so that we won't have the failure case. Um, and then what we would do is Like if um, if we wanted to, we could probably generate a CSV that has a failure case to see what it does. Um, and then build in um, failure handling. But again, we're we're talking about a a proof of concept right now. So um, depends on how much time we have. >>

Cool. And would you would you include success metrics here or are you imagining that later in the workflow? >> Um as far as like how to handle uh failures or how many like um emails Could be uploaded or >> how you would how you would know this worked, how this was a valuable thing that you made. >> Yeah. So right now I think like for a proof of concept my success metric would be can I upload a CSV, get it through this process and see it in the UI. Um and then after that I would

start to think about um scale failure handling um and then putting it in front of real users to see are they able to do are They able to use this without handholding at all? What feedback do we get? That sort of a thing. >> Any opinion on like you described like the pain or the status quo? Any thoughts on how you might measure this is less painful? >> Yeah, absolutely. I would probably um you know as a product person the thing that I love to do is actually watch people go through what they do today

and and actually like do a user test session With them um the way that they do it today and have them like sit there and do um you know monologue as they're going through and then have them use the new version and monologue as they're going through. um maybe even as sometimes as one after the other and sometimes as two separate sessions um with different users to see what the different monologues are. Uh because sometimes when you have the old and the new in the same um context, you get Different results than when they're done

in different contexts. And so it's it's interesting to see how people respond um in those different contexts. Cool. Do you think we're ready to Oh, you're Looks like we're ready to include these um uh requirements. Um after you uh let me know what questions you have for me. Um so this is always a thing that I include in every time I Start building requirements of chat GPT. I ask it what questions it has for me. um especially when I'm on the spot um because I'm gonna forget stuff and so um I want to see what

it will ask me um just to see if it'll fill in some blanks. Uh I'll answer those really quickly. It'll kick out our requirements and then we'll go throw it in bolt and see what happens. >> Let's see. So you're typically have you have you tried doing this all in one Like using bolt for example for requirements or using a single like what you would use to vibe code you also use to build the requirements or do you kind of always do that? >> Yeah. So I I would oneshot with Bolt when I first started

um where I would like start to chat with it. Um, the thing that I saw with Bolt originally, now they have Claude integrated, it's a little different, but um, found with Bolt that it has a propensity for action And so it starts building >> before it has everything. Um, and so then it will have to undo things. Um, Bolt also just in general has a tendency to like build a bug in every time. Um, and so there's a there's a high likelihood that we'll have to have it fix something. Um, uh, it almost it almost

always has like a weird bug. Um, and so there's a chance that we'll have to do that. I don't know what's going on here. Chat GPT is just really slow right Now. Um, but, uh, you know, what I've seen that it can do really well, um, is that when I give it a really nice prompt, it takes out so much of the creativity that the model wants to to take on its own. Um, and it it just gives it a a punch list and it handles that punch list really well. >> Um, and I find

that with a lot of a lot of the models and so um, chatbt with its chat interface because it doesn't really Keep track of tokens for you in the chat interface. I do a lot of ideiation here where I'll give it just full-on, you know, conversational stuff like this where I don't have to type and have it usually it responds by now. What is going on here? But yeah, I'll usually like do the ideation here and have it like beat up the idea, ask a bunch of questions. I'll answer those questions and then I'll have

it create a canvas doc um that I can then just like hit a Copy and paste it into bolt and go um because that tends to work pretty well. Um I've even made an MCP where I can have like um Claude or Cursor take a plan that they've written and pass it over to OpenAI and have it review the plan and kick back uh notes. Um which is a lot of fun. That sounds like a valuable case as well. That could be a cool product. >> Yeah, it's nice. Um, it uh it usually like it

gives a pretty good rating on Those plans. Um, but it'll be like, "Hey, you didn't think around this corner or hey, you didn't do this." And then cler cursor will then go and update their plan um with a couple of the holes that it had. And then, you know, I have just a little bit of a better plan because two different models have now looked at it, which is kind of nice. All right, cool. It's doing a lot. All right, so it's gone through. It's looking Yeah, it went through and looked For a bunch of

packages, which is nice. Um, it's looking at schema validation, which we want because we're asking it to do that with the CSVs. Um, it's actually looking for um, CSV header libraries and because we said it was for an email upload, it's actually building an email validation which is interesting. Um, now it's planning the uh, questions and the implementation which is nice. Okay. Uh, canonical field we will say so I'll Start answering the questions as it goes. Um, for one we'll say email address. to wonder. Do I have something running that's making this so slow? We

don't have anything running. Yeah, it's just Chrome. Check everything off. Okay. So, it's just the internet. Okay. two, max CSV size. Let's just say uh 5 megabyte for now. Three, Um uh CSV only. See, so now it's asking me a lot of the questions that you asked, which is really nice. Um uh any required validations between valid email. Um, let's uh yeah, let's ddup by email trim whites space. Um, drop empty rows. That's all nice. Um, by what's nice is that it'll be able to do this really quickly. Um, preferred data grid vibe. Uh, let's

do Batteries included just because that'll be fun. six uh uh display only is fine for this PC. Um okay, open source pieces. Uh let's go with that's fine. Below is ready to paste bolt new prompt. Um update the prompt and provide in a canvas for easy. Copy paste please. Um, and I will go and hit the canvas button so that it does that. It'll do this quickly. >> Do you ever answer those questions using Whisper still? >> Um, yeah. Sometimes the only thing with Whisper is like I haven't figured out how uh like spent enough

time with it to like do the new line stuff. >> Mhm. >> Um, and so sometimes I'm just faster uh that way. Usually I'll do whisper for like the um Stream of consciousness stuff when I just don't want to write paragraphs of text. >> Yep. >> Um or like I'm self-conscious about somebody watching me type paragraphs of text. Okay, let's see. This thing should be So this is nice. So it's giving us all the things that we need, right? Um and it's going to give us This is why I really like doing this. So, it's

giving Us tons of things that we need um including like all of the um various pieces of functionality, the the sample data um like readme file, everything. Um which is great. It's going to give it to us in this canvas file. So, we can even open that file just by itself. Um it we don't really need all of this. So, I can actually pull this out of the canvas file, which is nice. Um, and so now we don't have to like paste everything below. I can actually go and just copy This. Let's go over to

Bolt. I have this ready to go. I can paste it and we can go. Um, and so now it's just going to start building, which is great. Um, Bolt is going to it's going to sit here and think about it for a second. It's going to make a plan, which is nice. Um, it's, you know, every agent's going to remake the plan that we just gave it because it's got to sit there and chew on it for a second. Um, but luckily we've boiled that plan down. Um, but it's basically Regurgitating what we just had.

Um, three-step wizard. Upload, verify the fields, output the data. Um, and it's going to start building. It probably already has. Yep. So, it did its basic Nex.js environment for us, which is great. um it's already created a lot of things um and it's going to they do a lot of stuff um async which is amazing um so it's pretty quick and um you [clears throat] know so we have a lot already going for us and so like It's already like built out a lot of table stuff I think they have a lot of this already

built and they're just copying it into the environment so that they're actually optimizing their token usage which is pretty cool Um because like why would you recreate these components every single time? >> Yeah. >> Especially if we're just saying tailwind CSS and next.js and I don't care what the colors are. Um you know um and They're rebuilding the same thing a thousand times a minute basically right now. So um which is awesome. And we're not going to have anything here yet because um they're still building. But if we sit here and watch this it'll start

throwing more content in. I don't think we're going to have very many pages. Uh, but let's see what else it's doing. Yeah, it's sitting here doing a bunch of actions. Oh, here we go. Uh, Yeah. So, drop zone's where we're going to put the file. Um, it's writing a parser. So, we can look at that lib pars ts. Where's that source? Oh, not in there yet. That's funny. So, it's it wrote these, but it hasn't put them in here yet. >> While we while we wait while we wait for it to build, like how would

um what's your experience or your your take on this sort of prototyping PC phase into How you would actually like productionalize it? How would how would how might engineers take hold of this for example? >> Yeah. So often what I do is um even you know in some of the projects that I've actually taken to production um I will do this first I'll actually connect it up to GitHub um so I'll actually go up here connect it to GitHub if um you know if I'm using a database or something like that I might actually connect

it to Uh Subabase or something else um but then I will pull it down on my machine from GitHub and I'll connect it to Cloud or Cursor and from there I have uh rules set up. I have agents set up. Um that's where I'm actually going in and doing all the things that we talked about before where I am I'm going through and making sure that um it's secure, it passes all of my test requirements, it passes my coding requirements, everything is well linted, um it builds Right, um you know that everything's typescript strict. Um,

if I was actually going to do this thing for production, I would have spent a lot more time on the requirements doc. Um, I've written 20page requirements docs before and actually only asked Bolt to give me um what is effectively a proof of concept that is just a design system and component library as a hello world next.js JS app that has my routes And components mocked up for each page but not functional. >> And then I put that into GitHub and pull it down and then I actually use u you know a hands-on um IDE

or um CLI tool with agents where I'm going page by page and wiring it up building out the services and doing all of that um with a lot more attention to how it's being built. um so that I'm making sure that it's following the rules and guidelines that I need for the Environment that I'm going to. >> Does it give you a sense of where it is on the journey and how far it has to go? >> Um so right now um it it will be pretty close because we're almost there. >> Like there's not

a lot to build here. So but it it's not telling us like we're 60 70% right now. Um, if we were in a different tool, typically it would like give you its to-do list and show you. >> Um, now it's just saying, "Look at all the stuff I did." [clears throat] >> Wiping the sweat off its forehead. >> Yeah. Yeah. It's like, "Look at the hole I built for you." >> Um, but we're getting we're getting pretty close. Uh, it's building the project right now, so it actually might have something for us to look at

soon. Oh, just copied everything to the source directory. >> And is that uh is that database icon at the top? Is that showing the data model? >> Um, so this actually would connect you to uh if you wanted to actually connect it to a database. So >> if you wanted to actually go to like Subabase or something like that, you could do that. >> Um, you could actually connect it up to uh Superbase O clerk. You could put server functions in if you were serverless. You could do a lot of Different things in here. >>

Is there a way for you to expose the data model with what's going on right now? >> Um, yeah. So, if it put the types over, which it hasn't done yet, um, I'd be able to do that. >> Got it. >> So, that would be in the source file, the source directory. U, but it hasn't it's not giving me that yet. There it is. Okay. So, here. Come on. Minimize. Cool. Good job. So, here we should be able to go and see uh components books. Um where are my types? Not sure where it's putting the

types. Interesting. Maybe there transfer. Looking to see where it not totally where it put the types here. Oh, here we go. Okay. Um, so it's giving us the upload stuff, but uh I don't know. Let's go over and see what it what it gave us in the UI real quick. Um, the app is ready to use. I don't think it is here, friend. Okay, let's see. Three-step wizard blah blah blah. So, it's Yeah, it's saying that it is uh Bolt pulled a fast one on us here. >> I've been there. >> Yeah. Oh, it actually

didn't put the actual wizard component in. That's why it gave me a my bad. Oh, cuz it put it in the source directory instead of moving it to the app directory. Yeah, I mean this is one thing about using tools is that you are at the mercy of the tools. Um, and when you have somebody watching over your shoulder, you know, you're sort of responsible for What the tool is. >> Oh, they acted up too. I think they can perceive it. I think they know there's an audience. >> Are you on a Are you on

a basic plan or is uh >> No, this is a pro plan. Okay. Oh, Pro 50 is the next level. >> Yeah. So, all right. I'm going to download this CSV file. Oh, it gave me sample.html. That's neat. Um, the sample file that downloads downloads is Sample.html, not sample.csv. So, I'll have it fix that. Um, I'm going to open that really quickly, uh, just on my own real quick and try to fix that. Um, really quickly, um, no, my text edit app. Uh, I'm not going to share that. I'm just going to, um, make a

CSV file. Um, fame, lame, um, let's see, puppy, um, And birthday are the fields I'm going to give it. And then I will put a fake email [email protected] Matt Berner um Cricut and one one 1900 uh test two. Uh I'm just making some garbage data here to upload. Okay. I have file save as that's one sample.csv. Okay. Okay, let's see if it will let me do this. Okay, I'm going to drag a file in. Oh, it's still editing. So, let's not do it until it's done editing that. Okay, refresh it. Okay. Uh, let's see if

this will let me open this file here. Can I do that? Nope. It's not going to let me show you the file, but um I can put it in here and see >> that's like a template it's created. Uh so yeah it should have created a template here um >> to download sample 1 CSV and then this should be a file uploader but the drop zone is not giving me a file upload uh control um and it is not working for dropping a file u so I do have an example file Um, but it's not

working. Okay. And then here, so this is a CSV file. It's a little bit small. >> I just did tested example um with the Headers so that you can see that um that's the one that I built. So now I'm waiting for this to restart. Soon as I do that, we should be able to upload and I test it out. Um, one thing that I do like about like vibe coding too is like if you've never used some of these components, right? Um, the agents can get the docs and figure out how to use the

components and put it together and then you get to Learn how they work. And so what's really cool is like it's it goes right here. It says um I had to make sure it's rendering correctly and then let me look and make sure that um it needs proper event handling. um verify the imports. And so you can go and like see, you know, all the different components of this. And so for someone who like had to go and figure out how all this stuff works as a Developer, you know, like I know how all that

stuff works, but for someone who's learning how to code, being able to have like an agent build the code and then going into the code and seeing, excuse me, what it built and then translating that I think would be so valuable. I kind of wish I had that when I was, you know, learning how to develop. >> Yeah. Are you are you ever also like feeding it GitHub repositories you found Like hey this does a cool thing can you put that into this app? >> Oh absolutely. Or I or I'll be like hey I want

to try to do something like this and it's like oh you just need to do this this and this. And I'm like cool do it. [laughter] It's like all right cool. And um like I have an app that um like I've never set up Reddus before. And so I had an app where it just was like, "All right, I'm going to go set up a Reddus uh database At in a container and do all of this stuff." And I was like, "All right, sweet." And it just did that. And it just worked. Um, yeah. So

it is not doing this file upload at all. Let's see. >> I haven't heard Reddus in uh like eight years. So that's like in heat like storing >> in Yeah. Yes, >> cache, >> a key value store for caching, you know, Something like that. Yeah, this is just not working. >> Um, so, okay. So, what I would typically do at this point, um, because Bolt is not doing this, right? Um, is that I would probably Okay, let's see what's going on here. Perfect. So, we found a couple of problems here in the logs. Um, so

I would either give it these logs or I would go over here and I would look at my JavaScript console and figure out what's going on. It looks like there's a Couple of problems here. Um, so I'm going to pull this up full screen and check this and try like clicking here to see if there's something going on. So we definitely have some console problems and we have some um server problems. Um, so here and we have this. I'm gonna copy this and give it that and just be like, there are some issues. Please fix

these. I can't upload and I am in interview. I would love for this to work. Please help me look good. Um, And so, uh, you know, like that that's where I would, you know, I would start there and then if Bolt can't fix this, this is typically where I would be like, "All right, we're going to GitHub. I'm going to push it up there." Or I would download the code um, and like just export it, open up cursor, and have like start working through the issues of cursor, like build it locally. so that I can

see the errors without Going through the like bolt proxy here um and start going through and actually fixing the issues. But um now that it can see the errors that I gave it instead of asking it to try to like guess at what the errors are because I guess it wasn't looking at the server logs um hopefully it'll be able to fix this and and we'll get it going. >> Do you find that it'll uh respond to emotional appeals like hey please help me look good? >> Sometimes it does actually. Um and and actually what's

really fun is when you get excited um with some of the agents and and you're like f yeah, it'll it'll actually respond in kind. It'll match your energy. Um it'll even match your language. And so like I usually take screenshots when I get it to curse back with me. Um because it's just kind of awesome. Um >> I don't know if I want it talking like me sometimes. That's funny. >> Yeah. It'll uh it'll give emojis. It'll do all kinds of stuff. Um, and like you know when when I'm like that was amazing like we

just did something really cool and it's like yeah check this out right and um sometimes you can get it you can get it going really cool. Um actually just showed somebody that in chat GPT you can actually like get it to start like doing pirate talk and and other things just by changing your >> um which is pretty neat. Okay, it they're saying it built successfully with no errors now. Um, it did compile, so that's good. Um, I think at this point I'm probably just going to like force it over to GitHub and then pull

it exit because I'm I'm about done with Bolt. Um, let's see here. Oh, here. Open a separate preview tab. There we go. Let's see if this will give me the SA. Give it one more chance here. Uh because it's just not doing what? No, that's not it. Oh, it's locked. Okay. All right. Uh Okay. We're going to go brute force. Um, let's see. This is going to be CSV uploader create repository. All right. Um, I'm going to change my screen sharing for a second here. all that CSV uploader. I have a sneaking suspicion that doing

npm run build will tell us I need to do Npm install first and then we'll run the build and we'll probably find out what's wrong with it and then fix it. Um, also another thing that I find with Bolt is that um, almost every single time there are six vulnerabilities in whatever libraries that they're installing. Um, and if you fix it right away, you end up with some problems. Um, oh wow, it's actually going to build this time. That's amazing. That almost Never happens. Okay, fine. >> NPM rundev. Um, can you see my screen? Okay.

Is it too small? >> Yeah. Yeah. Yeah. So, we're in cursor now. >> Yeah, I'm over in cursor. so that I can actually like see things happening. Um, is this going to it actually works now? I might have just been over there. So, let's get the file and let's try to upload download sample go. >> Nice. >> It might have just been data preview. Continue to mapping. It was just running on their system that it was having trouble. So, email, first name, last name, add persona fields. So, um, okay. So, this is where it didn't

do exactly what I wanted, um, because I have to know the field that I added into the, um, into my CSV. Um, so I had a different Thought in my head than what it did. Uh, where I like I wanted it to show me the fields from the CSV so that I could add them as persona fields. Meaning, so just just a pause like the idea you had these additional sort of descriptors and you were hoping this would say, "Hey, I found these additional descriptors. Do you want me to pull them in?" But instead, it's

kind of making you hunt for them. >> Yeah, exactly. Um, and so like it what I Would do like right now is um so like uh depending on like what time we have like I would almost even just go to cursor and be like hey right now where it says add persona fields uh instead of making me add them suggest the fields that you found that aren't already mapped and allow me to add them. Right. Um but right now what I can do to do this birthday add and then map um and then let's go

to review and look um and so what it does is now it has um All the rows unique emails duplicates reviewed invalid emails and so what we can then do is let's get no uh this Now, let's get sample CSV over here. Let's add in an extra row. Um, and then let's go back here. So, it was just the fact that um it was in Bolt. So, that was on me being uh being rude to Bolt, I guess. >> Huh. >> Uh, so project CSV uploader. Let's go. sample CSV open. So, see how it has

duplicate rows now. >> Yeah. >> And let's go to continue to mapping. Um, so we don't have those fields. So, let's add puppy back in. Let's add birthday back in. Let's go to puppy. Let's go to birthday. Let's see if it actually gets rid of the duplicate fields. Duplicates removed two. Unique three. Total rows three. Invalid email zero. and it fixed it. >> Cool, man. Would you be able to demonstrate how you might on the fly iterate a solution to that problem where it didn't quite do what you were hoping? >> Yeah, absolutely. Okay. So,

there's two things right now. Start over doesn't work. Um, so let's uh I want to fix that and then let's also do the um fix that uh that fields thing. So, back to mapping right here for persona fields. Okay. So um with this CSV uploader There are here we'll do a sheet. Uh two problems I want to fix. One is that the start over button at the bottom of the last step does not actually start over. Um it should reset us back to the uh first step with no memory. The second problem is that when

we go to the um data review after we upload uh for the persona fields, what I would really like is instead of it asking me to add fields in, I would like it to show me the Fields that it found in the CSV and allow me to choose which fields to add from the CSV file rather than asking me to type fields in. >> I've got this like reminder on my computer to start using Whisper. You're definitely inspiring me. >> It's super cool. Um I got it through um Lenny's podcast. >> Yeah, I just saw

it there, too. >> Yeah. Um, and that's actually like I got a bunch of tools like my Bolt Subscription. Um, and a bunch of other things through there and Whisper was one of the ones that I was like, "Oh, this is the coolest thing." >> And is that 200 bucks price tag for all of that? I can't remember. Something really reasonable. >> Yeah, it's 200 bucks. >> And and then they just added a new tool that I haven't had a chance to play with um that I just noticed in there yesterday or the day before.

>> Might try to sneak in expensing Lenny's podcast. >> Yeah. Um, it's worth it because I mean even just like um Granola, which is like a um a notetaker, you can add like your whole team to it once you have it. Uh you can add as many people to your account as you want and each team member that you add is like a $15 subscription. And so it's it is more than worth it. A Okay, we are almost there. Um, so and and maybe my um crap talking with Bolt is just me not fully understanding

how Bolt works and so fully willing to accept that uh uh rudeness that I had there earlier. um cuz it might be running serverless and just doesn't want to do file uploads or it like they're they've got a security thing that won't let them do file uploads and so that's why it wasn't Working. Um >> well, you are gracefully getting around that in a way that many others wouldn't be able to. Well, and and like how frustrating, right? That like it didn't tell me that. Uh and that the only way that I figured it out

was that I downloaded it to my computer, tried it, uh because if I was sitting there and and wasn't technical and was using Bolt, um and was trying to do this with Bolt only, I would have sat there for a day Trying to figure that out or would have gone to a different application or something and then that would have been really frustrating just from a user experience standpoint. Um, the one thing that is nice here is that you get to see the actual like file changes and things that it's doing. Um, >> and just

a nerdy observation, someone was telling me, I think someone was telling me cursor's default LLM here is clawed. Is that right? >> Um, it is when you are not in auto mode. Um, so you can like choose your LLMs and you can like modify your models >> and choose a bunch of different ones. Um, but when you're in auto, um, which is the like this is just included whatever is included in your monthly plan. Um, that is, uh, whatever is cheapest for them, I believe, at the time. >> Um, and I mostly stay on auto

and I've Had like two million lines of code months before. So, um, it works pretty well. Okay, let's go and see. I can't remember. Can I drag that over? No, I tried that. Okay, let's get sample CSV. Okay, continue to mapping. Perfect. So, uh, now it's actually So, now it overcorrected. See, it put all the fields in. >> Yeah. would do is I'd say, "Hey, don't add the fields that we already have." Um, but it did do these, which is great. Um, and it automat them, which is awesome. And then we can continue to review

and we're good. And then if I go start over, now that works. Um, and so I would just go here and be like, "Hey, um, uh, this is great. Uh but we now have all the CSV fields and only needed the net uh net new fields that are not auto provided by the app. So email fame ln name Don't need to be included because automapped. Oh, you know what? Okay, I know why it did that. And and so I'm gonna This would be a product decision that I would like as a developer, I would go

back to the product manager and ask. Um as a product person, I would probably then make the decision that if it automapped, I would want the UI to handle this. Um, and so check out what happened because it Automapped to our actual application fields. Our application fields are email, fname,ame,ame,ame,ame,ame,ame,ame,ame,ame, lname. The CSV or sorry, application fields are email address, first name, last name. Our CSV fields are email, fname, ln name. So, I bet if I change this to email address, first name, last name instead in the CSV file to match that. And I go back

over Here and start over. And then I re-upload this. I bet it doesn't suggest those fields. Oh, it did. Okay. So, I was wrong. Um, so what I need to do is tell it um if it matches to not suggest them. And then we need to figure out what it does after that. But that's >> what you Well, just just walk me through what were you what would you frame to the product manager for that decision? >> Yeah, what I was thinking is that um There's a couple of options here. We could say if they're

an exact if they're an exact match, don't put them in here. If they're mapped, we should label them as mapped. And if they're not mapped, we should label we should show that they're not mapped or like separate them out. Um, and then if they get mapped, they should auto update to show that they're mapped. Um, that way it's very clear and you don't end up with like two email address Fields being put in there, right? Or two first name, last name fields. Um, because that could actually end up causing problems inside of your application. And

so we should we should add little UI things here, but I think that would be a product decision um to make that clear. >> Got it. >> Yep. >> Uh how do you how do you feel about where it's at right now? >> Um I think for a PC and then identifying the little tiny things like those little UI improvements, um I think it's not bad. >> Do you have any draft ideas on where a road map might lead with this? >> Oh, I have a million things. This is a problem that's been in the

back of my head for years. >> So, getting to work on it's great. And like I would totally like fix all these little things. I would like add some Color. I would probably turn this into like a um open source package um and make it so that it like doesn't actually care about what your UI is. It's more backend. Um there's a lot of things that I would do with this. And how about like with respect to users, how you might how that might shape your approach to a road map like how how might you

take could you could you take this right now and um start learning With users and how might you approach that? >> Oh, I mean I would do like kind of what we talked about before, right? I would probably um I would map this like onetoone feature-wise with the like with an actual application for like the fields and functionality. Um, so I would actually map it, you know, actual fields. Um, probably make this UI a little cleaner and, you know, maybe um, horizontal instead of vertical or Something like that. Um and then I would I would

do the you know as a product person I would do the um user testing um where I have separate sessions with some users um direct one after the other session with some users and then individual sessions with users that haven't um touched the original one um with me and then uh try to get feedback from folks to say like uh you know like run them through a use case and say like look you're uploading a um you know your Email list um this is how you're going to do it from whatever system you're exporting from

um you know does this actually help you do you know what are what are you thinking about you know asking them all the questions about the various things that they're clicking on where they're pausing um doing all of that as like a user research step um and then I would probably sit down with if I have one my PMM I'd sit down with my um designer um because this has not been Touched by somebody who actually knows design. Um, and I would say, okay, how do we make this better, right? How do we clean this

up? How do we make this more user friendly? Um, because like these could be small cards. This could all be bumped up. This could be made a lot nicer. Um, so that it is a lot more user friendly, especially for a marketing user. Um, and then how would we integrate this into our application? Um, and then I would probably do another Round of user testing to to just see if we're starting to get the same response from folks in a consistent way. >> Cool. That's awesome. Just one more curiosity on my mind. Do you have

any thoughts on like what the business model for this could be? >> Um, I think that uh yeah, so if you're looking at like if you have an email tool, you're looking at how can we get as many uh contacts from um other systems into our tool, right? and make It as easy as possible. Um, and so that user experience is great because you like obviously you're well I don't obviously but you're trying to get people to use your tool as much as possible and marketers will talk to each other and so if you can

get them to use your tool um and then talk to each other about how awesome and easy it is that is a huge win. Um, and [clears throat] then if you can once you have the data in, you have the emails in, you can start to Show them how valuable your tool is, they will keep putting those contacts in, right? Um, and so you you definitely want to do that. The um, as far as, you know, going from there, it it just depends on what the the platform, the product is, and and how you would

go um, and expand, you know, um, from there. Let's start off with um a really amazing interview from both you guys. Awesome work. Uh let's start with Scott. Scott, when you when you mapped out this Problem, what do you think is important that you want to see from candidates? It's not well, it has to be like grounded in the reality of real users, real needs, evidence-based to the extent it can be. So Matt instantly grounding it in, hey, I worked in Martekch. I've saw firsthand that this this problem really sucks. Let me tell you about

it. And then I tried to direct a little more just to say, well, can we double click on that a bit more because all marketing Teams aren't the same. Like who are you actually who is this person who would be using this? Uh, but he was he was pretty much right there. You don't want it to be vague. You don't want to be like, oh, it's really hard to do marketing or whatever. You want to be this is like a very discreet thing. It sucks to merge email lists. It sucks to do this. >> Yeah.

Um, so you don't want to be like, oh, everything's painful. I'm going to solve everything. It's just like, no, This discreet thing is where I'm going to start. Yeah, I think I could stop there. >> Cool, cool, cool. Did you Did you have gates that you you expected the candidate to get through on this problem? >> I have I have kind of notes I've been cheating with, but um they were directing me. Yeah, I was just kind of thinking like if it was just shiny object, which is a very common product Engine. It's a problem

for everyone. Like, oh, it'd be really cool if we could do this because it's really cool. Yeah, that does sound cool, but does anyone want it? Like, that's a checkpoint. the reality of it. Like, is it does it does he seem like he's I don't know if I can swear on this episode or whatever. Does he seem like >> Let's say no. Let's say no. Yeah. >> Does he seem like he's full of crap? Um, >> yes. >> Is he just saying like >> does he seem like he's just making stuff up? That would be

a thing. So, trying to probe and be a reality check, which is kind of the foundations of product interviews in general. Everyone can talk about stuff. Everyone can be creative and just be like, "Oh, I've done things like this." So, you just want to probe and be like, "Do you really know what you're talking about? Do you really know What you're talking about? >> At what at what point did you say to yourself, "This guy knows what he's talking about." >> Pull this book off the shelf, right? >> Yeah. >> Yeah. Empowered or I've got

three or four other books up here that you can just pull off your shelf. Inspired, sacred, seven, trust, and inspire. Like you can regurgitate stuff out of a book. That's easy. >> Whether or not you can actually apply it, that's the difference. Did did you did did you have a moment Scott could think think of the moment where you said to yourself wow this person knows what they're talking about. >> There's multiple layers to it. So if we say like the problem space he he kind of put his martekch credentials on the table. I've worked

in martekch. So I was like all right all right you can make the claim but then he started Demonstrating like no I've got like fluency in this. He starts saying I know the ecosystems like >> CRM to Zapier or what I'm like all right that's credibility. And then watching him >> and then the other layer is kind of the vibe coding itself. So as I said like watching him first go well even his workflow just like hey I'm going to start in chat GBT because that's where I like to do requirements. It's like well My

go-to is Claude. I'm very personal with Claude. We're best friends but I get it. >> So it's cool to see your workflow and then to go to Bolt and say all right I'm going to start defining this. Hey I almost swore again. Hey crap. It's not doing what I wanted >> to then have the resilience and the experience to say you know what I'm going to go into cursor. I'm going to look at or for no first I'm going to Download it and run it locally be like yeah it's not working all right now I'm

going to go to cursor I'm going to look at the repository I'm going to see exactly what's happening and I'm going to like troubleshoot this like that's he just kept unfurling like yo I'm pretty sophisticated at this I've been doing this a lot >> yeah can can you say a little bit more about the resilience a lot of these interviews involving AI nowadays I Imagine candidates get very finicky it doesn't go wrong it's not responding someone's was watching over my shoulder. I feel like Matt encountered about 10 roadblocks. >> Yeah. >> And was resilient to

hold. Can you just say a little bit about how with these new tools, it can be very easy to just get flustered. >> Yeah. >> Can you you say a little bit about >> very Yeah. It's kind of the interviews in general like someone asks a question, you're like, "Well, I wasn't expecting that." Or they or they they kind of poke you a little bit. They might I I kind of conceptualize things as like boxing. Like someone tries to punch you in the face and be like, "How are you going to react to that?" here.

It's like your tools are punching you in the face. So, how are you going to react to that? These are these are non-deterministic. You're not going to get the same thing every like he could do the same thing twice and get completely different outcomes. Chad GBT server was very sad for him. It took four and a half minutes to even parse what he gave it. And that could have been a moment where someone would be like, "Oh my god, I quit. I can't I can't handle this." Like, yeah, it's you got to stick through it

because this job inherently I think it's kind of the subtext of product. It's very Lonely. It's very entrepreneurial. A lot of people want to just point at you and be like like I this happens to me all the time. Like that's Scott's idea. I don't even even know if I believe in Scott's going where I I get it. He seems competent, but I don't really trust him until I see it. So that's kind of the vibe. People don't believe it until they see it. And if you >> don't have the experience of managing that pressure

and just believing in Yourself and just like being in the moment and like for me it's like my spiritual journey of deep breathing and like pranayyama and stuff like >> Yeah. Oh, I you after you go through this stuff, it's the same with like look speaking, we're musicians, whatever. The first time you're on stage, you're so self-conscious like, "Oh my god, this feels terrible." Then you just keep doing it. It's the same deal. So, I would argue just keep as as clearly Matt Has. >> Keep practicing, keep trying, and be like, "Oh crap." As he

said, like a normal person would have hit that wall, spent eight hours being like, "Why won't this drop zone work?" >> But he's like, "No, I already know how to troubleshoot. Let me let me go to routine number two. Then we go to routine number three. >> Yeah. >> Pushes that building up the familiarity To navigate because it's always going to be ambigu ambiguity. It's always going to be chaotic. It's always going to be things not behaving like you want. >> What would be your hiring decision and why? And maybe say something about leveling. >>

Yeah. >> Well, is this is for a product role? Yeah. Well, I guess he he was saying I guess part of it could be the ownership of it. Like I would talk to a PM about This. I guess if if you were explicitly in the mindset of this is a product role, it' be like, well, I'm going to I'm the PM. I'm going to make a decision here. This is what I would do. So, that struck me, but I know it's like we're in a weird context here, but that would be one where I'd be

like, well, maybe you're not maybe you're not like the most senior for this. Maybe you do need people to kind of bounce ideas as you that's a weird way to say it. We do need To bounce ideas, but more like I want to I want to own this. I want to be autonomous versus I'm not I'm not sure what I want to do here. I want to work with someone else who kind of owns this with me. But with respect to like the leading edge of vibe coding, the state-of-the-art, uh, the podcast I've been watching,

I think Matt's right there. >> I agree. Yeah, I agree. So, just like overall just hiring for a role at a Company, maybe it's product, maybe it's engineering, maybe it's something that in that like what would be your, you know, strong hire, hire, no hire. >> Yeah. Like kind of kind of where we're at right now. The zeitgeist is I think we said at the top of the call, we know vibe coding is happening. We know like the the top tier, the Google, Netflix, maybe there's a few more name brands are starting to bust it

out, but it's not pervasive. And it seems the zeitgeist is People are saying, "We need AI." Companies are saying, "We need AI here. We need people who know this, but no one knows it." So, they're looking for leaders who can come in and say, "I know how to do this." And someone like Matt >> just clearly demonstrated, "I know how to do this." So they're looking for people where sorry to have people looking over your shoulder again, but they're looking for people to come in and like teach them what what is the art Of the

possible now because everyone's handcuffed by what we've been doing and they can't even see what they don't know. So it's kind of like open up your eyes, have someone like Matt come in, prove themselves in an interview like this and then say, "Oh, you've got all these things where it doesn't have to be three months of PRDs and debates. It's like 10 minutes or as we saw here with GBPT maybe 45 minutes and you got a thing and you can all react to it. >> Incredible. Yeah. Incredible. I want to want to redirect a few

questions to Matt here. I thought this was an absolute master class on vibe coding. You made this really interesting comment earlier when you were like, you know, this is the thing about using tools. You're at the mercy of using the tools. What is your mindset watching having someone watching over your shoulder and knowing that in an interview? >> Yeah. Uh, so I I think one thing was When I'm in an interview, I am always looking for how the other person's reacting. >> And so one thing that um that I appreciated about Scott was he was

being as patient with me as I was trying to be with the tools. M >> um because it's really easy, especially, you know, when you're on a time crunch, when you're trying to interview a 100 people for a role, especially right now when you're getting thousands of uh Applicants, >> to be really impatient, be like, "Come on, >> get it together, right?" And and the fact that like Scott, you were watching over my shoulder. You saw Chad GPT was just sitting there being a dog, >> right? And like it's it's super easy to be like,

"Come on, Matt. What's going on?" But like at the same time, that wasn't me. like we're on a video chat. You saw that like my video is working so It's not my internet. Um and so just like that there was appreciation for the fact that like you you weren't like egging me on to keep going and like you know figure it out. >> Just to jump in. I suspect in these interview cycles you will have a bulldog who's going to like just as we were saying they're going to try to knock you off like what's

the problem man? You don't know how to use it? Like what what's what's going on? I'm sure there Going to be people out there who like poke you. See if they can knock you off your comfort zone. >> Yeah. And you know, I was thinking that too. I was thinking, man, I could really derail him right now. I could be like, "What's the problem, dude?" Like >> and and I can handle those situations, right? But that's a big flag for me. >> And >> and that is a that's a situation where I'm going to be

like, I work with this Person >> because if they're going to do this during the interview, what are they going to be like working with them every day? >> Yep. >> Yeah. I love that. I love that. I think almost What do you guys think about this idea? The tools are going to punch you in the face. So, as a shortcut, be ready to talk shop in the downtime. I don't know if you guys noticed, but when the Tools were messing up, you guys were like, "Hey, let me just get curious. What's your favorite LLM

of choice?" Oh gosh. Yeah. Sometimes cloud does this. Is that is that a decent tip for people? >> Oh, yeah. Oh, yeah. You have to be able to distract. Um, and that's that comes from doing the product, John, a lot >> because you're going to have situations where you're in a room with people >> and somebody's going to lose their mind, right? And somebody's gonna you're gonna Have to distract from the executive or the person in the room who's having a moment or from the, you know, the demo that's not going well. You're going to

be on stage giving a presentation at a conference and something's not going to be working. You're going to have to vamp, right? And you have to be able to right there, pick that up and handle it. And that's part of being a product person. That's part of handling those situations. >> Awesome. Two more questions and then we're done. Awesome stuff. I just think there's a lot of misinformation out there about newbies getting cooked by LLMs. And you said something really cool earlier and I wonder if you could just expand on this where it's like you

can use these tools the way that you are using them to see okay first you do this then you do this and then you do this but what's your take Matt on newbies that are new to vibe coding and your Kind of best practices for them so they can avoid what people are saying where it's like if you're not a code don't even start. you're going to be paying somebody 10 times more in the long run to go and fix your your vibe coded stuff. Yeah. >> Yeah. Um so the thing that I have done

that open the door say that the thing that opened the door for me with all of the agents is the agents have no judgment and you can ask them any Question. And so when you're sitting there trying to figure something out, just ask the agent because it will like half the time it responds, man, you're the smartest person in the world for asking that question. Here's how you do it, right? And then you get this like nice little ego bump for like learning how to do something. And then it's like, yeah, of course you can

do that. And then it's like, here, let me help you. And you know, like we were talking about Reddus earlier, right? I wanted to do this thing that required caching or like quick lookups. And it was like, oh, cool. I'll just add this in your Docker compost file and we'll add this whole server thing that you've never touched before and now we can do this whole new feature set and it just did it and then I went and figured out how it did it and now I know how to do that. Right. >> It was

like it was your first prompt to chat GPT. You said did you have any Other questions for me based on this and I thought that would be a best practice for newbies. Really smart. >> Yeah, absolutely. Yep. And >> I've heard Yeah, I've heard that. Sorry to cut you off and jump in. I've heard that from many who are kind of been doing this for a long time is like don't just start building get clarity ask it like what what what am I missing and I I do that I try to do that in general

with whatever it is irrespective of vibe Coding just like here's my take or what I'm doing I did I miss anything like can you do some research do you see any gaps here >> the the other thing I'll add too is >> a lot of the time I will give it a prompt >> and it will give me a result and then sometimes a lot of times what I'll do is I'll say, "Okay, I gave you that prompt. You gave me this result. How could I have given you? What prompt could I have Given you

to get a better result?" >> That's so smart, >> right? Um, and like the other day I was working on a project where I was like, "Okay, we got this thing. How could we have done that more efficiently?" >> And it was like, "Oh, well, if we did it this way," and I was like, "Cool, rewrite the prompt to do it that way." And it did. And then I was like, "Okay, now how do we design the system so that it's even more efficient?" I was like, "Okay, let's do that." And so then I ended

up rebuilding the whole thing to do it more efficiently. Cut out like 20,000 tokens per request and two minutes of processing, right? >> Amazing. >> And it was just because I asked the question. >> Yeah. >> And and it works. Guys, last thing here about Whisper AI. I just saw this post earlier today from the pragmatic Engineer and he was saying that Can you guys see my screen? >> Yeah. >> Um I didn't believe it till I saw it. I'm going to go again. I didn't believe it till I saw it, but it really works.

um you can hear devs whispering. So here's an engineer using whisper and what they found is that based on their data of 20k users, developers who use it do 18% whispering first typing in month one and then 72% whispering versus 28% Month typing in month five and it grows almost linear linearly until that point. Can you just tell us what you think of this new tool and h how you see it's um yeah your thoughts on this new tool how it helps you? Um, I mean, I I use it for, like I said, for a

lot of the like um stream of consciousness stuff where I'm just like, I need to get this idea out of my face, out of my head into the computer, and I just don't >> I don't have the um >> I I I just can't sit there and wait to type it. It's not going to be fast enough and I need to just get it out. And so I'll just hold the button down and just, you know, uh, verbally just give everything I can. Um, and then let the AI just sit there and parse through everything.

That for me is great. And I do know that like when I open up Whisper, it has all of these other things that it can do. And I haven't Even dug into that because I have so many tools to dig into. >> Nice. >> It has hooks for cursor. It has hooks for all of these other things that it can do. And I haven't even dug into that. So I know that it like this totally makes sense to me. >> Awesome. Yeah. Apparently, it takes the meaning of what you said, not exactly the transcript, which

is next level. Thank you for watching this master class On vibe coding.