welcome to the AI tools episode of building Jam we're going to talk through what are the eight tools we actually use week to week to build Jam let's go let's see the Tweet here we are the first tool on the list here we have in product API drum roll please Gemini and open ey okay yeah we used we used two so this this created a lot of discussion in the tweets like why are you using two of them so we have this feature called Jam GPT it's a debugging assistant right in Jam we're also building some other AI features that there's an early access happening right now but top secret well so so why do we use two it's because they're spiky and good at different things so the Gemini's ability to analyze video is like Far and Beyond everyone else for everyone else you have to sort of you have to like break down the video into its components and send frames and transcripts and all like it's just really hard to work with video and Gemini just works the video great out of the gate um but it's the quality is not you know as up to Snuff on other things as you know like CLA or open Ai and so uh we we end up using multiple cool okay prototyping number two yes okay so hey building an AI feature is very different than building other types of features because uh usually for other types of features first the designer can go in and like plan all the user flows and then the engineer can like look at the figma and be like got it I'm gonna build that but with AI the user flow is generated and we need to prompt it and and so like actually the text that is generated is the user experience and it needs to be designed like um for example imagine that jam wrote the bug report for you but it wrote it in a way that that sounded very corporate it would change your company culture just by using this product right so we actually care about the tone we care about the length like we don't want you to read a lot of stuff so um so that's that's a designed experience how does the designer play with prompts and try stuff out so uh glyph is like zapier for AI you can sort of like build these workflows and plug and play and so like Chris as he's been working on these like new AI features will just like create a glyph for the feature where he can try different prompts it'll be like he'll like sort of deconstruct a jam and he'll send the various parts of the jam and he'll send different prompts and be able to sort of like figure out what would be a good prompt to suggest to the engineers oh I see is it can you give a a concrete example about how we'd how we look at one part and how we'd make a or change something based on it like I feel like it'd be I'd love to see like specifically or hear more about like a concrete yeah like here's an example so um there's a lot of details in every Jam right there's the device details there are the console logs there are all the network requests there's user actions there is a transcript of the video walkthrough there's the screen itself there's just a lot what a when a human receives a bug report we're very good at like just knowing what to focus on logically based on the bug but if we just send everything to an AI sometimes it will overly focus on one thing versus another like maybe you have an ad block on so you have a bunch of network errors and it's overly focused on your network errors even like the bug is about something else and so it's really important for the designer to sort of prototype like I'm going to send here are the conso ons here are the network requests here's the device information here's the screen and then tweak The Prompt so that the AI is always focused on the right things that's like one example that's hard to do if like only the engineers have access to like code and prototyping that's why glyph is such a great tool cool okay number three code let's talk about co-pilot there was a day where pet was like on a loner laptop and he didn't have access to co-pilot and he was like at the end of the day he's like I didn't get anything done like this the like team has been using this now for like a year and a half it has totally changed their workflows um Oscar says he can't imagine going back to a time before co-pilot he says he thinks that makes him 20% faster writing code um but my suspicion is as a team it's probably higher than that and the reason is that um one of the things that co-pilot is really good at is it will name your variables for you very logically like like that's like a hard creative task for a human it's actually quite EAS like it's a lot easier for an llm to name a variable it can write more code comments than you might be willing to spend time writing um and and so it helps us create more readable code and that has like compounding effects as the code base grows and as more Engineers join the team okay prompt engineering yes number number four we're talking about prompt Fu prompt Fu I I've actually I've never heard of this one myself when we started um building AI features we were first like tracking our different like ideas for prompt and then like um their sort of eval rates like as we would test them against like test jams how well they would um perform we were all doing all of that tracking in a Google sheet and then we learned about this tool called prompt food which just basically helps you track and test all your prompts and then like then you can man you can like manage your prompts in one place so you don't to like keep track of which prompts do we try and how did how do they perform okay podcast production something I'm very interested in myself yes let's talk let's talk wondercraft so so there are two Jam podcasts there's this one which is like how like trying to share what we're learning building Jam um we have another Jam podcast which is called this week in JavaScript it comes out every Sunday at 10: a.m. and it's what happened in JavaScript news in four minutes or less and this one is uh is like the voice is not either one of us it's an AI and it's it's done with wondercraft AI um we met this team because they demoed wondercraft at an AI event we hosted in London um they are such an awesome product team um and it's it's cool to now use their product wow that's and and I guess I would add here the AI features or what we're using to produce this podcast is the tool descript and I want to show you just some of the these features because that that sounds like a really AI native um platform and this incorporation of AI feels a bit different but um so like yeah let me see here so this is what I was working on before this is the episode that came out last week and some of the cool um things that I use almost every week is just like um over here this uh rep repurposing features like for example creating clips and creating a highlight reel from a longer recorded episode um and so for example like often I'll find that these are out of the gate aren't what I need to actually be able to to share and to to they aren't sharable out of the gate but they give you this first like initial draft that then I can work with and so as I'm like trying to find moments to create a highlight reel like creating this highlight reel here I mean come on might as well might as well show you but after I create a highlight reel here I'll find these little moments that then I can stitch together like and it it just lets you take like more shots on goal like as you're looking to identify key moments um and so I found that to be really helpful myself that's so cool so I use another um this might actually just be the next tool in the Tweet um I use another sort of similar AI tool um called Opus clip um because we're trying to share with the jam Community things we're learning building Jam like like like we're doing now and um so sometimes like I'll go on a podcast somewhere and we'll do this like hourlong conversation we cover a lot of topics and so I want to figure out like what did we cover and what might be interesting to share more about and so I'll take the link from that podcast I'll like put it into Opus clip and it's actually really good at like it brings out moments where someone goes oh wow or like people laugh or whatever and so then we get sort of a transcript and I use that to write like other content that we share about how we build J actually using it as a way of finding highlight moments or looking back on things that you didn't record yourself that's a that's really interesting yeah maybe this one I'll share so like um I went on this podcast I called founder real talk it's by one of our investors and so we have this like hourong YouTube and so I put into Opus clip and 20 minutes later it's got like here are the best clips for you to like share and like okay here's my feature request for like Opus clip like I can't exactly share this because like this like doesn't look quite polished like the people are cut out the font is like you know very tick tocky but I can be like oh we talked about how we did user feedback we we talked about this moment that we had a saster and I can be like okay is there something for us to share for this example no but like that's that's how I use it okay where were we now we're at meetings let's talk about grain HQ yes we record all of our calls with grain and grain has two AI features that are really really helpful so one of the most important things we do with the grain recordings is we clip up the most important points for team discussion like if a user had great user feedback or is using G for really interesting use case we'll create a clip and in the past we had to like name this clip well but now grain has like an auton name your clip feature that's helpful the other thing is um oftentimes recordings are really good for like sharing information so uh I'll go do an advice call I'm like oh my gosh ero I wish era was there he would have learned so much or vice versa we can send each other the call um and grain just does like meeting summary key takeaways it's pretty good um and so that's a helpful AI tool we use in the stack yeah I mean I feel like I I'm always seeing grain Clips all the time in our slack like it's just all it's all grain Clips which is which is great let's okay prds by the OG chat PRD yes um okay I think one of the hardest jobs in every startup is to be the first product manager we have an amazing first product manager name her name is Maya um and she solely is responsible for writing prds and making sure that they are awesome and clear for engineers um and so chat PRD is just one helpful tool to help you make sure that like you haven't forgotten something something could be clearer uh something could be formatted better like just great to have assistance when you're the one person product team um managing a product product that has 156,000 users Goa and the last the last on the list writing okay three Triple Threat here okay this one I feel a little spicy about like part of me thinks like when I read stuff that's written by an llm I can tell I have this like sort of very um like I have this sort of purist approach where all the writing should come from a person how however now that these things have been out for a while sometimes they are really helpful editors for your writing and the different LMS are are the different tools are good for different things so for example if we're writing a guide for running a bug bash um and we're like there are probably a lot of tools out there besides Jam perplexity can help us go find out a lot of those tools add some details about them and then like combine it into one article right and then we're like okay we want to rank well in SEO um we should find out like we want this structured in a certain way we go to Claude and we're like Claude write an outline for this SEO article that has these tools in it right so so we use a mashup of all of these depending on what it is that we're doing or like sometimes it's even really basic like um you know Ivana and I are writing like an email subject line for our latest feature announcement it's the unhinged hour which is like the evening before the 7 a.m. launch we've exhausted all of our ideas and now we're like what is the AI think um and and maybe something's like it just Sparks idea so we use all of them basically yeah this I mean I I think about the parallels just in my video workflow too with just like the the generating options like it feels like a very productive um way to brainstorm or like just generate all the explore the The Sur the surface area of possibility around what you could uh what you could make and so videos is like what what could all the clips be with writing it's like what could all the different titles and subject lines be and it can be very helpful as a part of that process even if it's not making the production ready one writing AI task that I've seen era do and I thought was just like really really cool is so if you look at era's calendar it's just like you know back to back to back toback sales calls right and so um but he needs to follow up on each sales call and say like I'm hearing you you're saying these things and here's how we are planning to support you and so what he does is he takes the transcript from grain and he puts it into Claude And he says and he's like write me the three bullets of like what did I hear from and of course then he goes in and edits but it's a lot faster if he has to write like eight follow-up emails after a long day of meetings um to get a starter like that so I thought that was a great like AI to help assist in writing so that was the list okay wait I want to discuss one more thing yeah so there's one response to tweet that's like in 2024 these are all tools next year all the AI quote tools that you use will all be agents like you won't be using um you know a writing tool you'll be using an AI writer you won't be using um a a coding autocomplete tool you'll be using an AI coder right I actually think that this is how Jam will be used 10 years from now because if you like if you think about an engineer that's managing all these like engineering agents that are going off and doing these sort of discret tasks you have to then review their work and like quickly share with them feedback so they can iterate because they won't get it right the first time it's like a junior engineer and so like people use Jam for that I think people I think Engineers are going to send jams to the agents to be like fix this this is broken this is not right you know change this um I'm really really excited for when we have um you know Creator accounts viewer accounts and agent accounts on our pricing page cool thank you Ian thank you