How AI Can Grow Your Business in 2025 (Expert Advice)
Jay Singh, formerly at LinkedIn and Microsoft, walks through how to use AI for workflow automation and product development in 2025. He covers specific tools like Claude, Cursor, and v0, plus patterns like RAG and collecting-synthesizing-generating pipelines. The biggest friction point: getting teams to trust AI output when hallucinations and accuracy matter.
Chapters
I am completely obsessed with AI right now, but I was looking for someone to bring on the show who's more obsessed than I am, and you popped right to the top. My guest Jay Singh works with Microsoft and LinkedIn on AI implementation strategies. He then turned that insight and experience into an agency that helps founders design and build awesome AI products.
I think where we're going with all of this is more personalized software. If your voice or authentic stories are the beginning stage of any AI product, it makes the response is so much better. 2025 is going to be the year of a lot of—
Oh, that's interesting.
I think to simplify how AI could really help, are there parts of your business that have the following 3 steps? You have a lot of data and you need to be able to collect it. You need to be able to synthesize it. That could ideally leverage expertise. And then can you generate something from it? A lot of very simple applications can be built with that in mind.
What actionable workflows we think creators should implement in their day-to-day?
The external product development, there's Cursor, there's Lovable, zero for designs when you're talking with creators about how they can monetize their audience and monetize their business, uh, maybe actually the best way in the future just—
right, Jay, I am completely obsessed with AI right now and what can be done. Uh, I want to dive into all the projects that I've done recently, but I was looking for someone to bring on the show who's more obsessed with AI than I am. And you popped right to the top. What's been your background the last few years diving into AI?
Speaking of being obsessed with AI, my friends and my partner call me JPT. And so the more, the more I get made fun of and make names that are similar to different AI products that exist today, I think that could kind of give you a sense of maybe some level of the obsession there.
Yeah. Oh man. Yeah. There's so much of that where they're like, oh, you're just using this for everything. Yes, I did have a moment. 'Cause I try to use AI all the time and bring it into my workflow. Someone, a little tip that I heard was add a second monitor that's just dedicated to AI. And so you always have it like off to the side, like do a little vertical monitor that's just like 3 or 4 AI tools right there. I haven't done that yet, but I had a moment the other day where my team wanted me to record a video and it was like, there's a Slack thread going back and forth what the video should be. It's leading up to our team retreat. and I almost, well, I was, I knew there's a lot of great stuff in there and all the info I needed to write a script was in there, but I just said, hey, could one of you summarize this? And, you know, just gimme like the 5 bullet points, cuz I'm gonna record a Loom for the team. And like a few minutes later I got that back and I was like, oh, perfect. That's excellent. And I started reading through it. I'm like, this is Claude.
Yeah.
And I just realized it was like that, let me Google that for you moment.
Right, right.
Where I asked for something and my team member Mick, who is amazing, she was like, Sure. And she like pulls all the context into it, including my request, puts it into Claude and comes back and is like, there you go. There's what you asked for. And I, so I imagine this moment where she's like, you know, yeah, I can use AI for you if you want, dumbass. Like, or you could do it. But she was so nice about it. But it's just that thing where there's still a few moments where you're not ingraining every little thing.
Mm.
Because basically what I found is 95% of the tasks that I set out to do are made better with AI, but I'm still probably only using it 50% of the time, 60%. You know, I'm trying to level that up. Anyway, that's a tangent, but I want to hear more about—
Well, I want to ask a question on that. As the CEO, I think sometimes you want your team to be able to use AI, but then sometimes you may feel like, okay, like, did you actually think about this? Did you actually think about this work enough? And so I want my team to be able to use AI all the time. And sometimes if I feel like they've used only AI, then I have a negative signal towards it subconsciously. So did you feel that in that moment when you were like, oh?
I didn't feel it at all. What I felt in that moment was obviously I should have, like—
I could have done this myself.
I could have done this myself and I shouldn't have bothered them with it because AI is very well suited to the task. There is a moment where, like, if people take out that taste or quality filter, or if you can tell they just use the most vanilla implementation. Yeah. Where, like, because ChatGPT is a terrible writer.
Yes.
By default.
Right.
Like absolute garbage.
Yep.
You know, it's the 7th grader who found the thesaurus and is really excited about it. And so like that is unacceptable. Right, because we have bars for quality and work.
Exactly.
All of that. But someone who's saying, hey, I took all of, you know, we recently rebranded the kit, so we have all these messaging guidelines, how we talk, all of these different things. And it's like, I have carefully created, you know, this custom GPT or, you know, it's projects in Claude that writes just the way that we do. And then that, I'm like, hell yes. That all day.
Do you, as someone who's written a lot, previously, now when you're writing with AI, do you feel like you're able to be as creative as you once were?
I think you can be so much more creative.
I love it.
Because you just move so much faster. Let me give you an example. So I'm working on a book that requires a lot of stories about creators. And something that we've done really well at Kit is we tell a ton of stories. We profile, you know, dozens to over 100 creators. But to have that and to access that is really challenging. And so Isa, our storyteller, I used to DM her in Slack and say, "Hey Isa, I need an example of a creator who stayed really consistent for 2 years or more and then had their breakthrough." And she'd be like, "Oh yeah, how about these 3?" And all that. But what we did is we just took the 150 stories that she wrote, loaded them into Google Notebook LM, And then I ask it the question and it'll say like, yeah, here's these 4. And it'll cite the example. 'Cause I can be like, really, that one? And I click through on the citation, read it and I'm like, oh, that is a good example of it. Then when I'm stuck in my writing, then I'm like, I'm getting unblocked and unstuck so quickly because I have access to all of this, not just the broader internet, but the source material that I provided it, you know, all of my own research. And so I move so much faster.
I think, I think you're getting on an insight where if your voice or authentic stories are the beginning stage of any AI product, it makes the responses so much better.
Right.
And so I think that's why, that's why I think 2025 is going to be the year of a lot of voice integration with different products. Because just think about your— have you used ChatGPT transcription or advanced voice mode before?
Yeah, a little bit.
It's just easier to be able to share exactly what you need done when you're speaking. Versus typing, whether it's with another person or whether it's with an AI product. Think about the different examples you have within work. Is it easier to be able to just like write out a really long Slack note? Sometimes if you're really feeling the typing energy, but a lot of times it's easier. Hey, do you mind if I can just hop on a call quickly?
Yep.
Or send a voice memo, send a voice memo.
And so I was going to type this out, but you know what, let me just think about it.
Exactly. And so then there's just think about how much context is now going into the prompt. There's a lot more when you're speaking. It's easier to be able to share more context. And so, and so I think, I think that like voice integration also creates a little bit more authenticity. And so a lot of the stories that, you know, you're plugging into NotebookLM, I'm assuming a lot of that came from originally podcast or conversations.
And Isa, our storyteller, did, you know, she interviewed all these creators extensively, asked them tons of questions and then distilled it down to the essence of the story.
And so those data sources, if you can plug into them, whether internally or if you find them externally, ends up becoming amazing ground for creating really good AI content. It's actually via your stories, your voice, et cetera. And you could tell then versus just like a quick prompt on ChatGPT.
Oh, that's interesting. Okay, so I think that AI is transforming like almost everything in how we run Kit. But your business is to go in and either create products, AI-related products for other companies or transform services inside of it. So what are some of the things that you're seeing and doing?
Yeah, so just maybe a quick introduction on my end. Spent 5 and a half years at LinkedIn. I was on their product team there. I helped scale up their verification effort while I was working within that team. Got access internally within LinkedIn and Microsoft to understand like what the heck is going on in AI, right? This was in the summer of 2022 and just felt like I maybe had like a quarter of a head start with the world to say, whoa, this is going to really change things. And so I was always someone who wanted to kind of hack on it.
LinkedIn being owned by Microsoft.
Sorry, yeah, exactly.
Microsoft being very closely tied in with OpenAI.
Yeah, Microsoft has a couple investments with— A couple small investments. Yeah, exactly. And so at the time it was, it was like high-level overview of, you know, what the market looks like within LinkedIn's ecosystem. And so we can share back some of those insights to the Microsoft team. To understand how they should be thinking about the space as well within their portfolio of companies, LinkedIn being one of them. And so it kind of gave me the head start to be like, whoa, this is going to really change the world. Like, how can we start to take advantage of this ourselves? We ended up kind of developing a few products just off the side of our desk outside of LinkedIn with a few friends. And then one of them ended up kind of like getting a decent amount of usage. And we use that as a way to be able to say, hey, we kind of know how to design and build AI products as much as we can know. It's a new space. Yeah. Let us know if we can help you. As well. And so that has been kind of the business where we're helping founders to be able to design and build AI products, or we're using some of those same skill sets of building product to be able to help businesses implement AI into their workflows. And so when maybe, maybe I could place some like interesting things that I've been seeing over the past, past year or so. I think, I think to simplify how AI could really help, I think it's easy to get a little bit lost in the noise. Are there parts of your business or products that you'd like to create that have the following 3 steps? A, you have a lot of data and you need to be able to collect it. And then 2, you need to be able to synthesize it in an interesting way that could ideally leverage expertise in some sense based on your industry or based on your company or based on your use case. And then can you generate something from it? So a lot of very simple applications can be built with that in mind. So pulling a lot of data in, AI's really good at that. Whether it's scraping information, whether it's uploading a large body of text, synthesizing it, which in that case you can put your expertise, your prompting into the product. And then it then gives you like the right response.
I think a lot of our accounting and finance processes, as we work on budgeting, annual strategy, all of those things, it's an area that I haven't, we haven't touched bringing AI into yet. Partially 'cause there's this idea, obviously, as you're dealing with money, that accuracy is so important. And we've all had it where, you know, you throw out a prompt and it's like, oh, it's this. And you're like, okay, perfect.
I don't know if that's right. I don't think that's right.
And if I didn't have just a little bit of context, I would have missed that.
Yes. Yeah.
But yeah. So how would you think about bringing, right, like automating a huge amount of work, like monthly close, like all of the stuff in accounting and finance?
Yeah, we kind of run like a role play of some sort. So like what, what within accounting and finance, what are some of those manual processes that exist today that maybe take, you know, a team or an individual like a few hours to be able to kind of go through, review things like what workflows are kind of coming in.
Yeah, I'm probably not close enough to it to know for sure. But as we, so some of this will be speculation.
Yeah.
But you think about monthly close for a company that's spending $4 million a month, right? You have to reconcile all the receipts. The IRS still requires that every transaction has a receipt, you know. Do we pay everyone out properly? Is it all like categorized correctly? You know, are you depreciating and amortizing all your expenses? And then reporting on it, like how did we, you know, we had a forecast, how do we perform against it? You know, profit was $100,000 higher or lower than expected. Why was that? Well, it's not one reason. It's that this was up by 10 and that was down by 50. And this is, you know, there's a bunch of different things in that.
Which one of those, that's great. Which one of those do you think takes the most amount of time today?
I think probably reconciling all the transactions and you know, making sure that things are depreciated and amortized correctly. And I guess—
and today it's like tracking that happens how? Like, is it, is it being done within like a QuickBooks or like a Ramp or— Yeah, or individually, like literally looking at different receipts, like Brex and QuickBooks.
And then we use a tool called Runway, which does a lot of our modeling and that sort of thing.
Yeah, that's great. And then, and then from there you take all that information and create like different reports from it as well.
Yeah, exactly.
Got it. I mean, some of, some of these things and some, some places that we actually help is recommending off-the-shelf tools that you could use to be able to just do a lot of these things. I know Runway has like different AI functionality that they're launching. Brex has AI functionality that they're launching. Um, and so for that, for that specific use case, um, I'd have to probably go a little bit deeper into like, yeah, you wanna sit down with the expert in the business, which is not me. Yeah. But like, like quite literally, like we will either physically go and monitor how they're going through their day.
Like show me how to, yeah, exactly.
And like, it's like myself and, and potentially potentially one of our engineers and we're seeing, you know, let's use another example. You know, we were, we were helping a client in the verification space. They, whenever they have like certain checks that are kind of coming through their system, they have a few that they want to manually review. There's a person that manually reviews those today. That takes X amount of hours per day and they need to be able to reduce that because they're going to see a lot of scale coming up in the next few quarters. So, so what, what do you do with that kind of problem statement? It's sometimes simpler than you think. It's literally just downloading their ways of thinking and looking at a manual process that they may not even really understand themselves. It's asking the right questions. Why are you, why are you like flagging this or why not like doing this? Like just almost understanding what the right process is first. And then some of them actually you may not want to automate with AI. You were mentioning some of the different financial use cases. You need to be really crisp in terms of making sure that, that those finances, those numbers are accurate. And so maybe you don't want to actually automate all of those processes. You just want to do a few of the pieces that, you know, like get to the point where the human is actually just reviewing the last final number and then you kind of alleviate everything around it. But that thing you don't want to automate, you just kind of give up to the person to be able to review.
One thing that I've done is I come from the development background years ago with design and dev. And unit tests are something that are really interesting. Right. Um, one, anytime a developer is skeptical about using AI, they're like, no, it can't replace all that, you know, my expertise or all that. I say, why don't you just get help from, like, just have it write your unit tests.
Right.
Do you love writing? No, you don't? Okay, great. Have AI do that. And they're like, all right, fine. Like, the code is my craft. This is like a thing to just help me with.
Exactly.
And so that'll be a good gateway drug to get them in. But I also think about accuracy and like the hallucinations being so important. Yeah, I haven't done as much with this as I want, but I want more ways for, you know, say in this accounting process, verification for AI to be able to function as the unit test to say, hey, as a human I did this or as an AI I did this. Now this other separate agent or AI verifies, is this correct?
Yeah. Yeah.
How would you set something up?
Yeah. Yeah. So first is Uh, after you kind of go through the process of, um, downloading their brain, sounds a little bit, uh, interesting, but it's, it's just making sure you have documented all the different processes that person is doing. Um, help them, help them like understand, like, you know, in, in this use case, what percentage of the, of the, of the different verifications were you actually passing or failing? So you then have a benchmark that you can revert back to. Um, you can then backtest some of the, uh, kind of previous pieces that you may have run through that manual process as you're building out a small proof of concept to start running data through that product as well. And so then you can start to see what percentages are things getting passed or failed, or what percentages are things bubbling up for manual review. So I think the process is, it is dependent on the individual use case, but generally it's really understand what the manual process is, document it. To even maybe sometimes they may not even have like the right measurement themselves on how they're measuring success, help them do that and then use that as a way to be able to pinpoint where in those manual processes you can start to implement AI and then measure it against the previous measurements that they had. Ideally, if you have data that they've already done before and there's some audit log or some review of the percentages of this thing that led to a success or failure, you can start running those tests through that. Kind of like new proof of concept as well. So it got a little bit technical on that piece, but hopefully that's relevant.
I want to talk about tools and getting really specific there because people, I think, understand how to do the basics with ChatGPT.
Sure.
And they know how to make a custom GPT or some of these other things, right? This is like the 100 level, get your feet wet.
Yeah.
You know, a 20-minute tutorial on YouTube and you're off to the races. And then there's a lot on the custom development side of like, I'm building full software and Cursor is going to help me do that, that sort of thing. But these like in-between business processes and all these other things, it's like, what, what tools do you use? And I, as you're implementing these, I'm really curious, like a very practical application of something and then the exact toolset that you used to bring that to life.
Maybe even taking a step back of how I've been thinking about, uh, this, this space, uh, and the landscape, because there are a lot of platforms and tools that are popping up that are helping people to to be more efficient to develop or design AI products, to kind of like implement the right workflows. Like, the, the reality is, um, I don't— as an, as an agency owner, I don't know which one of these platforms are actually going to be the winner, right? And so tactically, what I'm doing is developing strong relationships with all the founders and CEOs of a lot of these different platforms, whether they're no-code automation platforms, you know, a lovable cursor, uh, Mind Studio. There's like Code Giant. There's all these different platforms that you can use to some part of the stack, you could be able to kind of build something or implement something. And so the reality is like we're trying to stay, you know how people say like you're model agnostic, we're trying to stay platform agnostic just to be able to see who's going to actually kind of take off and be the winner and, or the winners and whoever those are, we can position our team to almost become an expert in that platform to then become an implementation partner for other customers that need that support. One company that's doing a pretty good job, I helped advise them for a few months, is a company called Mind Studio. And so the way I would frame it is there's external product development with AI and then there's internal AI implementation workflows. I think people sometimes get confused of which one they want and why. I have a list of questions that I try to ask companies of like, are we talking about internal automation? Are we talking about external product development? Because you need to be clear about which one you're doing and why and prioritize it accordingly. Um, it actually changes a lot of what you end up developing, as you can imagine. So for the external pieces, there's Cursor, there's Lovable, um, v0 for designs. For internal, you could use platforms like Zapier. Mind Studio is a really good one. Um, so Mind Studio, you can kind of jump in there, uh, similar to like setting up a Zap on, on Zapier. You can set different parts of your workflow. You can say, you know, uh, if I want to be able to create LinkedIn content from my transcripts from my meeting notes, here's like a quick automation that I can do. And so automatically, you know, connects an API from Firefly, puts it into Mind Studio, takes a transcript. I go in the backend of Mind Studio, update the prompts, it sends it over. It like, then you can add a little block and say, okay, now create a next thread, create a LinkedIn post, and then email me that. And so that like pretty small automation, you could set up pretty easily within a platform like Mind Studio or like Zapier. Those are kind of the two that I've seen that have been helping a lot in terms of internal automation, at least.
Yeah. What I'm realizing about that is getting access to all of your information formatted in the right way makes a huge, huge difference. Like we had an example of a team member who we're doing biannual reviews, right? People are writing out, hey, how did, how did I do the last 6 months? They're getting that feedback from their manager. It's very, very important. And it can be tedious. And on one hand, you don't want to like just have AI write it all because it has to be accurate. But the people who have really good notes and have formatted them well and can put that into AI and be like, hey, and then ask it questions like, hey, what did I do really well? Right? And so Georgia on our team, she's an executive assistant. She talked about, you know, she's got great notes on everything, all of her weekly meetings. And so she plugged all of that in and she could just query it. Yeah. And she's like, oh, I totally forgot that happened 4 months ago.
Yeah.
I, on the other hand, have my notes scattered around between Apple Notes, Notion, probably some pen and paper, you know, on my desk, like all of this. It's a disaster to feed into. Like AI can't help me yet because I didn't follow good practices to begin with. And so what you're talking about is you're saying, hey, I could have my meeting transcripts in a queryable format from ChatGPT automated because I set up something with Mind Studio or Zapier to make that happen.
Right, right. Yeah. And then maybe within the master prompt that you set up within these, uh, applications or these workflow automation platforms, you could say, you know, it's gonna come in in a relatively unstructured format. Uh, but here's the format that I need you to structure it in and then run the analysis on it. And so for your example, maybe her name is Georgia.
Yeah.
Um, maybe Georgia can help you understand what is the, the structure that then you should dump in all your other kind of like, you know, messy notes so that it structures it in that form and then it takes it and, uh, it, it does, it can kind of create the right performance reviews or performance reflections for you.
Yeah.
At the same time, you know, that is the beauty of a lot of these large language models. They do take very unstructured information and then they help you structure it. And I mean, you could probably still take a couple pictures of your notes, like, you know, copy-paste your Apple Notes, and it still gets you to a decent spot without even necessarily having like the perfect structure going into it.
It's wild. Like, one example, it's crazy. We did a brainstorm. Yeah. Of all the dream creators we'd love to have on our platform. We did this in 2019 and we had names up there like Ellen DeGeneres and Hasan Minhaj and all these other people. And we went back through the list and like they're customers now.
That's amazing. That's huge.
Putting that out in the world. And so it's like, all right, let's do that exercise again. So you have 80, 90 team members putting sticky notes up on the wall and all of this. All right, someone's got to Copy all that down and it's like, take a picture, take a photo.
Done.
Do you want comma delimited? Like what format do you want it in? No problem. Here's the data.
Isn't that crazy? And like, we just forget like how fast of an innovation that was. Like within 3, like 3 years ago, that would've taken 3 people on your team to sit there and review every single note and it would've taken maybe 2 days to, to do that.
Yeah. It's, it's wild. Yeah. Okay. So you're talking about using those studio or those products for internal processes. You're right. What's the other end of the spectrum look like?
That's where I'd say a good 60%— most of our work actually started with helping founders to be able to design and build an AI product. Now it's becoming more AI implementation, using those same principles and helping midsize or larger companies. This is where we cut our teeth over the first year of business. I would maybe break down the product development process into different stages. I think you could use AI for all those different stages. For us, it's discovery, design, and building. So discovery is the process of just learning and understanding, doing research, speaking with potential users, customers, you know, looking up search terms, like just getting a good understanding of is this idea that you have in your mind something that you should actually work on?
Yeah.
And, and while, while it is becoming easier to be able to just take that idea and build something, I think it's going to— we need even more of a premium on that discovery process to make sure that idea has a little bit of relevance and people want it. In it, especially when everybody can just create products now or will be able to in the future. And so for discovery, that's a lot of the meeting notes and transcribing, asking the right questions, then synthesizing that information correctly. Those are simple stuff, maybe what we even just spoke about, asking the right questions, having all the transcriptions done, feeding it into a similar place, and then being able to create some sort of product requirements document to say, Here's Nathan, here's all the stuff that he cares about. Here's what— here's a product that he wants to be able to develop. And it kind of acts as your source of truth before you could use other AI products to then feed that into like a larger prompt to start designing or creating a product. What I like to do with clients is— so maybe like what our process was before some of these AI tools and after. Before these AI tools, I'd sit down with a client, I'd understand their needs, I'd kind of act as a product therapist of some sense and just understand what they want to build and why and who they are and I would take that, I would write down notes, I would send over a product requirements document to my designer. My designer would maybe spend, I know you've been a designer yourself, maybe like 2 weeks or maybe a week in Figma, spin up some mockups or wireframes. I'd then bring that back to the client and say, is this kind of what you're thinking? We'd go back and forth for a couple days and if they get approval, we get the deal and we kind of start working on it. Now with AI tools, especially tools like v0 from Vercel, I could, I could finish a meeting. Those transcripts are already being sent somewhere else as an internal automation. And then I could just create a product requirements document from that meeting. And if I've asked the right questions and I've done a good job—
Just to pause. Yeah. You would have a custom GPT that is creating product requirement docs in the way that you want?
The way you implement it is, you know, you can, you can, you can be pretty flexible with how you implement it. But the, the, The TL;DR is like, I have a transcript and I want to create a product requirements document from it. For us, I have a little, I have a Firefly, I have my meeting notes in Firefly, I have a Slack bot that then pulls out the meeting notes and then creates the PRD within my Slack. So that's like my implementation of it, but you could use platforms like Zapier or Mind Studio. You could copy paste a transcript, put it into a custom GPT.
If you wanted to old school like a caveman. Yeah, yeah, exactly, exactly.
It's not automatically integrated into your Slack. Like, what are you doing? I was like, so 6 months ago. Yeah, but yeah, it's moving so quickly. And, and so now I have a product requirements doc roughly that I can go and edit. I then copy paste that, I open up v0. Have you heard of v0 before?
Yeah, I'm familiar with it, but you should explain it.
Sure. Yeah, it's, it's a tool that helps you get to your v0, your first version of a potential product visually. And so what it does is you can copy paste, you know, your product requirements document or ask it to design a product for you. And then it's really cool. You'll start to see, you know, like you start, you put your prompt in, you start to see like a bunch of code and you're like, whoa, like what's happening? And then, and then boom, you just start to see some mockups being generated. And so now instead of that 2-week whatever cycle that I had with my designer, I can then, you know, if the process is happening well, I finish the meeting, I like, you know, leave the office, I walk back to my house, I already have the Slack Purity done. I just, you know, put that into v0. I start kind of like cooking and like creating some designs. And then, you know, within maybe an hour and a half, the client has a first version of their mockup done. It's not super high fidelity, it's not beautiful yet, but it kind of, it starts the conversation. And so then they're looking at it and you're like, no, no, but it's not like this or it needs this or that. I'm like, great, now we're having the dialogue. How much faster of an iteration cycle was that? Right. And then I give it to my team.
Best case, 7 to 14 days down to few hours.
And time kills all deals. And so that is a way to be able to reduce the amount of time that you can get to something fast. And then after they give some feedback on it, then I'll give it to my designer to be able to polish up. He takes that into Figma. He can then kind of create something a little bit more high fidelity. And then I bring that back to the client.
How do you think about V0 cursor will come up with some great designs, all of that versus Figma? Like I see this debate going on where someone's like, why would you even need Figma? And it's like, well, I mean, I still feel like I need Figma, but I also don't know where this is going to be in 30 days, let alone— literally 30 days now.
Yeah. I think, I think, uh, in our workflow today, you know, this podcast recording, things could change within like 2 weeks. Um, I like, I like having our team use both. Um, so it's, it's using v0 to kind of spin up the first versions of the product and then taking that into Figma. Um, you can actually do the opposite way as well, where you then take your Figma back into v0 to make quicker updates.
Like v0 has an import from Figma.
From Figma option. Yeah, exactly. You could even be scrappy and just take a screenshot of it and put it into, into v0 itself. So like Figma has more customization as a designer, as you know. Yeah. And so it's, it's nicer to kind of work in there and then, like start in v0, go to Figma, come back. It just kind of creates this circle. I'm really surprised that Figma doesn't have. you know, their own like AI tooling like that. Yeah, I'm sure they're, you know, working on something there, but, um, I bet the bar for implementation for them at their scale is very high.
Sure. Yeah. And so they, though they, well, they could do a Figma Labs kind of thing or some beta or something.
Yeah. They're, they're, I mean, I, that product is beautiful. So I'm, I trust that, you know, maybe they're following like a similar like Google to OpenAI model where they're kind of waiting, learning, and then, and then implementing whatever they need. But I still think you can use Figma as well to kind of like really like build out all the flows. Yeah. There's so many things in there.
I'm curious on the different platforms. Yeah. This is something that people ask me and I don't have a great answer for is if we're doing development, right? Lovable, Cursor, right? V0, right? Replit. Yeah. Like what are you seeing? Yeah.
And what would you recommend? Yeah. So, uh, V0, we use pretty religiously for designs. And then, and then Lovable, to be honest, I don't use as much ourselves. We end up seeing Lovable a lot whenever someone kind of runs out of the capacity that the product has for them. And then they want to work with us to be able to actually like build up more of an advanced product, I suppose. Cursor, our engineers use internally to just move much faster. Replit, I haven't used as much myself, to be honest. And so I think you can speak more coherently to some of the work that you guys have been doing. Yeah.
So on the Replit side, what I wanted is, I mean, going back to the use case, right? With Kit, we've got an app store, right? And so we're following the footsteps of iOS, Shopify, WordPress, etc. The line that I always say is like, imagine your iPhone, but only with apps made by Apple. Like, kind of sucks. I mean, at the time, greatest invention ever. Groundbreaking. Amazing. But like, you know, the App Store, completely changed the game. And so that's what we're doing with Kit is saying, hey, every creator, because our customer base, they're all creators, they all— creative people, they all want to make things happen. And so saying, hey, you can build and extend Kit however you like. Yeah. And a few weeks ago, I was really thinking like, I think we can make this process way easier. Not just that any developer can build apps on Kit, but as I start to play around with Cursor, I'm like, I think Anyone who is like slightly technically minded, you know, who, if you say JavaScript, their brain doesn't shut off, you know, who's like, who has some curiosity around it, could go and build an app. And so we set out to do this project to make it as easy as possible. So test tutorials and prompts and all of that. And I was looking for a platform that would be the most all-inclusive. So Cursor, I felt like was best at building out, like maybe building and extending my full platform codebase. Right. Exactly. Like a better developer tool. Yes. And Replit, I felt like was better at going from zero to deployed. Yes. And it's hosted and it can have a database and all that.
Yeah, it's pretty solid.
So what we got it down to is a single prompt that will get you an OAuth application. So authentication done, deployed, all of that done. And then more prompts take you through like, okay, here's how to pull in all your account information and go from there. And then the app that I just did over the weekend, because I wanted to learn, Kit has these editor plugins. So when you're writing an email, you can pull in content or do dynamic things and it feels totally native. And so I was like, well, hold on, can I make an app? That generates custom fonts, right? So you want to render your signature or header in a custom font that the email clients don't support. Well, can I make that render as an image? And I had to get on a call with one of our engineers to be like, hey, how does this all work? And he's explaining. He's like, no, no, no, it's just this endpoint that goes and asks your server, like passes information and then returns whatever your server generates. I'm like, Oh, this is actually really easy. And so it was an hour call with him and then 90 minutes on my own. And then I had an app that was fully functional that would take any font that I gave it and, you know, render as an image with that. And then it probably took me another hour or two to clean it all up and get it production ready. And so I'm like, which is still not that bad. I think I'm 4.5 hours in, like as a novice. And so incredible. What I loved about Replit in that case is that it's a fully hosted environment. Exactly. And it was great for that.
Right, right, right. I think it's speaking— I'd love to get your thoughts on where you think the future of the Kit app marketplace goes to, because it's a broader, it's a broader trend that's just been happening in software, which is, you know, products are becoming more and more personalized. And we went from, you know, large companies that were just vanilla and like supporting everything. And then, you know, we've kind of verticalized, like we have products that now support creators who want to be able to build a business. You have products that help designers who want to be able to build mockups. And I think, I think where we're going with all of this is more personalized software, right? And so each individual should be able to, if they want to, just to be able to spin up a piece of software themselves for their own use case. For you, you're doing it, you're doing it yourself for your business. And so I wonder, like, I imagine that trend just to continue. So how do you think, do you think, do you think that creators themselves on Kit can come in and be able to just spin up their own apps that can, you know, maybe that first they use it themselves and then they're like, why don't I actually make a little bit of money from this and like externalize this? Like, how are you thinking about that?
Yeah, I think that's absolutely what's going to happen because creators constantly want a process to be better or something to be different. Hey, could you add this feature? Why don't you?
Yeah. I mean, you could do it too.
Go ahead. Uh, I think it'll also allow us to experiment faster, right? Like for us to plan out and build this, you know, this custom font generation app, like, right. We got to prioritize it a bunch against a bunch of things and all the stuff, but I can launch it and be like, oh, this is just a, you know, a KitLabs thing or a Nathan side project. And really I'm just doing it. One, to see what's possible, but two, to show everyone else what's possible. Right. And the tutorials, because I went through— it's kind of funny, I bought Nat Eliason's course on build your own apps with AI. Right. Because I wanted to see what all was possible. And I got through part of it and I was like, on one hand, it was the best $400 that I've ever spent. On the other hand, I was like, I don't actually need this, right? Because I just needed someone to show me here's what's possible, like how to use Cursor. And then I'm like, oh, I got it. I get stuck. I just ask Cursor. Yeah. And so though he had lots of little tips and tricks where I was like, now I would have learned that like 3 weeks from now and knowing it 30 minutes in. Yeah, it was a game changer. Like how to get the right context. Oh, here's how to upload the API docs. Yeah. Doesn't hallucinate. Yeah. All that stuff. Yeah. But I think that just having that model of showing people this is what's possible, here's how you do it. Yeah. And then they'll come up with all of the things. Yes. Like someone was showing me this was Brendan Dunn, who owns a product called Right Message, which is really good at taking your website traffic and everyone on your email list and asking them gradual survey questions and then storing it in the profile. And what's cool about this is you can gather so much information and then he would do this really advanced personalization. That, hey, if you're a designer who has this goal in your business, I'm going to talk to you this way. Well, AI is remarkably better at it. And so he's building an app and watching this play out that, you know, is way better than any of the personalization stuff that we have built into Kit. But he's just doing it as a Kit app because he's like, hey, I have this great data set. And so I think the pace of innovation is going to go so, so quickly. And then you're going to see a stark difference between, uh, Kit and then these other platforms where they're saying like, oh no, we're a walled garden where only developers employed by our company will be able to build features. And it's going to be like, why? We have 60,000 creators that pay us. If even 200 of them become obsessed with the Kit app store and start bringing things to life, then the pace of innovation is going to go absolutely wild.
Successful platforms make more money for people on the platform than the company, like the platform itself. And so, you know, is there a world where when you're talking with creators about how they can monetize their audience and monetize their business, maybe actually the best way in the future is just to be able to build a Kit app? Right. That could be interesting.
I think it's going to be huge. One thing that I struggle with with Kit is thinking about what functionality to build into the core product with the pace that AI is changing and innovation. Like, I don't know what jobs will matter and I don't know what jobs to be done matter. So like taking an example would be like a landing page editor. We're at the point in our life, like product lifecycle where it's time for us to rebuild our landing page editor. And I'm like, what does that even look like? Is it entirely prompt-driven? I think what Lovable is doing with some of their new like visual editing and prompt-driven, like you can click on an element and then prompt something for just that element, is really interesting. But I'm curious how you think about or advise your clients on this, like, wild rate of change, but you need to keep leveling up your product, but you don't know where things will be even in 6 months.
Yeah, the real answer is that no one really knows. And I think, I think that's how I start off all of the conversations. Um, and it's important not to it's important not to kind of like trick yourself to know where the world is going, first of all. But what are, what are some steps that you can make to be able to kind of like prepare for that future? So some of the, some of the stuff that we've been, we've been helping our clients with, especially when they're thinking about like, how do they design like this next version of an AI product for their business, is really focused on like workflows and jobs. And then just like rethinking like what that could look like today. Ideally creating some flexibility for it to change in the future. So for the landing page editor piece, like, you know, given today's technology, how would you be able to create a landing page? Maybe you sprinkle a little bit of voice. Maybe there's a bot asking you like, hey, what kind of vibe you're looking for? Like what kind of colors? Like why? Like, oh, I've already actually like pulled up all this information about your instance on Kit, here's what I'd recommend. Here's a couple examples. What do you think? And it actually, it turns into more of a dialogue with the user to share insights, speak to it, and then it kind of just automatically starts mocking up these landing pages using some of the techniques that v0 does. And so that's one example of can you design something that is flexible, it has a conversation with the user, you can quickly iterate using these like off-the-shelf technologies. Um, so that, that's like, that's like one, one thing that I would think about, um, is how do you, how do you try to collect as much information from the user as possible through a conversation, right? Which then can lead to you generating something really quickly and just like creating that feedback loop with them. Maybe, maybe it's the same principle that I'm trying to articulate, which is, uh, with v0, right? I individually like go speak with a client. I understand things. I come back to them fast with certain things. That speed of iteration and feedback cycle increases, increases the value of that product because I can say, no, that's not it. Like a little bit of this, a little bit of that. So maybe that, that experience can be tested out within the landing page feature as well. Yeah. And just thinking about all of the context that you already have on—
Yeah, someone's been running an account for multiple years. Yeah. You know, there's some version of this where you could say like, You know, I'm trying to make a new landing page for this product. And it could be like, oh, the product that you've talked about here, here, here, and here in these broadcasts over the last 2 years.
You're like, how do you even remember that? Yeah, right.
Of course you have all the context of every broadcast I've ever sent.
Exactly. You know, but yeah, like surfacing all of that automatically, I think is really interesting. That's where some like agent workflows can come in as well, where if you have context on all the things that they've spoken about and maybe some of the people that they're influenced by or why or some, you know, You know, you'd have to take a look at each like individual dataset, but you could maybe run research flows where you're saying, okay, you know, Nathan talked about how he really enjoys like this type of website. Go like search online, go pull up those different websites, you know, grab it, screen, you know, like those are, you can kind of run these things in the backend too to recommend to the user right away what they should be building towards.
So as business owners, we can talk a ton about ways we're using AI and what we're experimenting on. Most of our listeners to the show are creators. And so I'd love to wrap up with maybe 4 or 5 actionable workflows that we think creators should implement in their day-to-day. I love it. Yeah. What, what comes to mind? We'll kind of riff off of those.
Yeah, I think the, I think, um, the one that, so, you know, I, I identify myself also as a creator. Um, and, uh, it's hard to know what to write content about or to share content about. I think And this is a founder building a company called Tone, which is focused on being able to take meeting recordings and then creating LinkedIn content from it. And so that I think is like a part of a workflow, especially if you're writing written content and you are maybe a part of a team or you're having like consistent meetings. It gives you that idea bank of things that you should take from meetings and then creating content from it. That's a simple workflow that like it just helps to kind of get some more authentic content out there really fast. So meetings are a thing that everybody has.
You either have internal meetings with your own team, with clients, or as a creator, you're meeting with your friends. Exactly. People you want to collaborate with.
It could even be podcasts or, you know.
Yep. All of those. I mean, especially podcasts, right? I think what most people are doing is they're saying, cool, I'm using a Grain or a Firefly or one of these. Right. And then it just goes off into a library and there if I ever want it. Yes. Yeah. But what you're saying is, hey, this is actually gold. Yeah. And we need this to be available. Yeah. For prompting. Yes. All the time. Right. So what's, what's the mechanics of doing that? Right.
Uh, I think there's a role in the future called transcript management. Uh, kind of like there is product management in the previous generation of software, because that is, there's just so much, there's so much interesting information that's just sitting within that transcript. Um, the mechanics can start pretty simple. It could be, You know, have the Firefly API or, you know, Otter, I think has like their own APIs connected into Slack, into your Slack instance. Ask Firefly to be able to start to create content for you. It gives you a content idea bank. And then from there you can kind of edit it a little bit and then share it. Again, platforms like Tone are like helping specifically with LinkedIn content as well, but it doesn't have to be overly complicated. That's how I have my kind of setup today. What I'm thinking about is using podcast transcription.
Oh yeah. It's almost— I wonder if I could get to something that shares my new ideas where it's basically like, look, here's everything you've ever published. Yeah. Or like I have 10 flagship pillars of content or something like that. Yeah. And I want to know 2 things, right? I want to know, did I talk about one of those established pillars of content in a new way? Yeah. Or 2, did I talk about something new that I haven't talked about before? Yes. Yeah. And then like, interesting. Surface both of those. And that would be fascinating as a content creator. Yeah. Yeah. And like, it's one of those things that from a first principles perspective, I'm like, I don't see any reason that couldn't, couldn't work where it's like, hey, what's new that I haven't talked about before? Totally. I know what you've talked about and I know what this was. And these two things I haven't heard you say before.
Yeah, it could be, it could be like a database that has all your content that you've written before. Some of them might be kind of manual to just like copy paste and like put into there to like have that initial data source.
Um, but like an export of your WordPress blog, an export of your Kit broadcast. Exactly. Exactly.
That'd be really cool. Um, and then, and then yeah, you could, you could store that. You could store that. Maybe you can use like Mind Studio or like another application like this. One quick thing.
What do you do when the context window is like, when it just can't handle things? Like the reason that I use Google's notebook is that it was the one that could handle, you know, 200,000 words. Yeah. Yeah. Yeah. Whereas Claude and ChatGPT would tell me, oh yeah, I ingested all of that. Right. And then it would only answer questions about like the first 5 creators in there. And I'm like, wait, you didn't read this whole thing.
Right. This is where some of the more technical implementations I think would come in. Um, there could be products that do this already today, but, um, this is where like we'd be using like, uh, RAG to be able to like—
Oh, what's that one? I haven't heard about it.
Um, Retrieval Augmented Generation. It's like a process to be able to store, uh, a bunch of information and then for future queries effectively, um, you know, all the AI nerds are gonna tell me I'm wrong in this specific specific explanation. I'm trying to simplify it. It's just like you have a database of all your content, it's exported from Kit. And then for future queries on, let's say, a chatbot, it first takes a prompt, it runs it against this larger database, it tries to extract out all the summarized list of information, and then it tries to answer the question based off of the existing data that you already have. Got it. And so it would be a little bit more of a technical implementation, likely. I'd have to sit down and this is where a lot of artwork comes in where people people get to a certain point of building these AI tools on no-code platforms and then they're like, oh, it's not really getting us there anymore. Uh, which is great for us because then, because then we can come in and be able to implement something for them more custom. Um, but for you it could be, uh, export, export kit, maybe use Zapier or Mind Studio to like upload all that data. Mm-hmm. You might run into some context length issues. Um, but you still, you'd be able to get like a few of these like, um, pieces of information. Yeah.
And you could distill it down and store it as a file.
You could summarize it, store it.
Yeah, exactly. Yeah. Yeah, that makes sense. Okay, so that's, that's one workflow as we're managing everything with meetings and transcripts. Yes. Yeah. Uh, what's something else that you think?
Yeah, maybe, maybe quickly I'll turn it back to you and ask like, what are, what are some of the, what are some of the pain points that creators are talking to you about these days? Um, not AI related, just like, um, like, like when I speak with creators, it's like, how do I know what to create content about? And that's like a, that's like a, a big pain point. But what else, what else are you kind of noticing? And maybe we can back into a good workflow.
Yeah, I think the The deliberate— actually, we see this a lot with flywheels, right? You and I both nerd out about flywheels in business quite a bit. And as we've seen people go through the, like the flywheels course in community, what happens, I think most creators don't understand how to distill their business down to like the core essence of what actually drives things. And flywheels help them do that. And then you can automate most of a flywheel, right? So I think what I would want to do is I would have an agent or, you know, an AI automation just around reporting, right? Pulling those metrics that you check on. Because something else that creators do is they tend— creators being human tend to focus on what's exciting and what's in front of them. And so they need this like reliable consistency. Of looking at the same metrics in the same way every single week. Right. And not, oh, I'm, you know, I was obsessed with LinkedIn growth, but now actually I listen to a podcast and someone's talking about Instagram growth. Now I'm going to flip over here, you know. Guilty. And you, as you jump around between all these things. And so probably that next workflow, which doesn't exactly answer your question, but the AI implementation that I would do next for creators is consistent weekly measurement of all of your metrics. And make it so that that is a 5 to 10 minute activity because the generation pulling of all that is totally automated, right? Where you'd want to know, hey, what's my top performing post? Yeah. Across all these platforms, all these platforms pull in this data and then ask, yeah, you know, have the agent automatically tell you, here's what I noticed about what you did this week. Right, right, right. Your frequency has dropped off. You're, you know, whatever else.
Right, right. That'd be great. I think you run into some issues around like pulling the right insights and metrics from all these different platforms and some of the kind of issues, you know, you run into there as lack of it for a lot of times.
And so there's, you know, there's workarounds for that, but then like, what's an example of, of a workaround that you would tend to implement? Like, are you going to scraping? Are you?
Yeah, there's, there's a lot of, there's products that you could use to be able to kind of like pull information from different platforms. I can't speak about that so much given some of my work at previous employers, but But yeah, there are, there are tools that you could use, uh, to be able to kind of like aggregate up, uh, relevant insights across different platforms. But that, that, that is, that goes back to your walled garden piece. Um, platforms have made it really difficult to be able to externalize information to other developers. And so do you think that the, um, once you get in, you get, once you get it, it's great. Cuz then you can like, and then you could build all these automations for yourself.
Do you think the like operator from OpenAI and you know, about being able to browse the web and then screenshot anything and pull the data out of it, right, is gonna make a lot more of this possible?
I don't know, it's a good question. I think, I think the terms of service— the, the reason, the reason why, uh, the reason why a lot of these platforms, uh, don't externalize their information in the first place is, um, A, because they're maybe like, uh, giving too much information out, which then could, uh, maybe hurt the reason why you would want to come to that platform in the first place. Reddit famously like cut off their APIs a couple years ago because there were products that were doing super well and effectively competing with like Reddit's main platform. And so they cut down the APIs. And so it could, but it would likely still run into terms of service issues.
And so just because it's technically possible doesn't mean it's legal.
Exactly. And so it, I don't know, I don't know if Operator itself actually fixes that. Maybe it actually maybe it actually hurts it even more because everybody puts up their defenses even more, right? Because you're like, oh no, like, you know, now it's— if you go work, if you go look at any of these like large social companies, like that might be one of their top priorities to defend against, right? Where before scraping, it was one of their kind of things to defend against. But now you're like, it's so blatantly there, you're like, okay, like, how do we, how do we not, how do we not make this happen to us? So there's a world actually where the ideal of operator doesn't actually end up operating. Yeah, that's good. So if you're looking for just other ideas of ways to be able to implement AI, if you're a creator, just take a step back and think about what are the things that are taking up the most time for you, especially pieces that are more manual and rote. For me, again, it's related to creating content, but it's a part of my workflow internally, which is creating a proposal document for a customer. I, that would be my time sink for the most part prior to automation. And now with some of the pieces that we're talking about from taking meeting transcripts, like creating documents from it, that actually has reduced the amount of time I was taking from maybe like a full day of effort to maybe an hour. And so, but that effort took some reflection and just awareness of what the heck am I doing every single day? What are the things that are taking up the most time? And then starting to go back to ChatGPT and ask like, hey, what are some ways that I can automate even these pieces? Right. Yeah.
I'm just thinking about all the things as a creator, you know, we're spending time like, how do I make things? How do I get on the different platforms and all that? But you know, you spend much time on finance and accounting, maybe have an AI-first platform like, uh, Kick in the finance world, right? That's specifically targeted at creators and that's going to help you streamline all of that process, right? Do your taxes much faster, relieve a bunch of pain. So I think it makes sense to just look through. Your calendar. A lot of creators talk about having this wide open, you know, I don't want to have anything on my schedule or all of that. And it's like, that's great, but you got to know what you're spending your time on. And even if you say, hey, just this week I'm going to audit it, you know, I'm going to use a time tracking tool or something else. I'm going to audit how I spend my time. And then you can see, okay, this is a manual process. This is a waste of time. Not a waste of time. This is important. But it took an— probably anything that takes more than an hour and you do it more than 2 to 3 times a month. Does that sound right? What is a candidate for automation?
Yeah, I mean, uh, if you're only spending an hour on that and doing 2 to 3 times a month, like, pretty good. Maybe if you're spending like at least, at least like twice a week spending, you know, 2 hours on it, uh, to— yeah, I'd focus on that piece first. Um, it's, it's hard. It's, it's hard to actually I think because we hear this a lot when we're working with different businesses at, you know, decent scale, it's hard to work with those employees that are really like doing that process manually and then say, hey, don't worry about it, we can automate it. And I think as a creator and as an individual business owner, maybe it's easier to do that because you're, you know, you can, you can see how that can impact your bottom line right away. But it's this whole idea of like, you know, implementing AI into society. It's going to take time because it's, it's a little bit scary. It's a little bit Oh, but like, no, I can't do that. Like, I could do it better. And sometimes that's right. Sometimes it just takes a little bit of like kind of inertia to keep going on that piece. But we actually notice that a lot internally where we get put, you know, an executive will say, hey, I think that manual process can be automated, but the person doing the manual process doesn't think so. And it kind of creates a little bit of tension, which is hard to work through. And so that could also be running through people's minds themselves. Like, no, no, no, no. Like, that's the thing. Like, they're never going to get automated away. Um, sometimes that might be true, and sometimes you may actually just want to just create that human creativity and not automate that thing away. I think that's important too.
It reminded me of just everyone having a different perspective, because this is probably a year, maybe a year and a half ago, so a lot of AI stuff was very different, more nascent than it is today. And my friend Joey was talking about a dinner that he had, and he was, he was with a doctor friend, a lawyer, and an accountant. And at first I I was like, Joey, is this a joke? Like, did they all walk into the bar? Like, what happened? He's like, no, no, no. We were just like, we were just hanging out. The group hangs out often. But he said every single person in that group could see how the other person's job was going to be automated by AI, but could not see it for themselves. They're like, but my job, you know, we're good. Well, I'm good. Like the level of expertise and all that. And people would be like, no, I'm pretty sure like all of this is gonna change. And so interesting from that, you know, someone saying like, oh, this process has so much bespoke work into it that it couldn't be automated. But someone else is like, yeah, absolutely could. And so what that means for yourself is to go and seek out that outside perspective and to say, hey, Jay, will you like look over my shoulder as I'm doing this? Or asking a friend, hey, here's how I spend all my time, right? I can't find anything to automate in this, right? And I bet someone else would be like, what are you talking about? This, this, and this could all be done by AI.
It's— but it's like, yes, and it's difficult to do that because we associate a lot of our identity around work. Like, like, work is a collection of different manual things that we've been used to doing. And so, like, you know, even, even being on the other side of wanting to help companies automate a lot of their internal processes or build AI products, I can empathize with, with people that are like, well, like, then, then what am I doing? Then what? Like, what am I meant to do next? And even at the enterprise level, what I hear from, you know, folks, folks within like Microsoft or some of these other like large enterprise implementation software providers, a lot, you know, the sale happens, but the implementation doesn't really happen because you have like internal strife that says, hey, you know, I don't want to implement this thing. And so this is going to take some time and a lot of like supporting those people to find other jobs and like, you know, right. Addressing the concerns of, you know, automation for them too. I think, I think that's going to be continued to be important too.
Yeah, that sounds good. Well, the whole world is changing very, very quickly. Mm-hmm. Thank you for walking through this episode with us, diving into it. It's unfortunate that it'll be completely out of date by the time it releases in like 2 weeks.
Yeah, yeah.
You know, just remember that this is the worst AI will ever be. Exactly. And so anytime something doesn't work out, it's like, yep, it's getting better at such a rapid pace. Right. But for those who are interested in what you've been talking about, you know, want to hire your agency or follow any of your work, where should they go?
Yeah, you can go to CasperStudios.xyz and you can see our website, or you can follow me on LinkedIn under J-A-Y-S-I-N-G-H, Jason. You can find me in there and DM me. Sounds good. Thanks for coming on.
Yeah, thank you. If you enjoyed this episode, go to YouTube and search The Nathan Barry Show, then hit subscribe and make sure to like the video and drop a comment. I'd love to hear what some of your favorite parts of the video were and also just who else you think we should have on the show. Thank you so much for listening.
