Startup tips from AI expert, Kushank Aggarwal

The NUMI founders talk with Kushank Aggarwal, founder of Digital Samiritan, a set of social channels dedicated to helping product people sharpen their design and marketing chops, with a focus on AI-enabled tools.

Startup tips from AI expert, Kushank Aggarwal

This interview has been edited for clarity. To view the interview in full on Youtube:

BLD:
Hey guys, welcome to Bottom Line Design. I'm Harrison. I'm Agree. And with us today, we have a very special guest, Kushank Aggarwal. Kushank, you are a multifaceted, almost like a Renaissance man across being a co-founder, skill sets, AI-oriented, marketing-oriented. Do you want to just give us a rundown of your history? And I know you certainly have a really great first question that I'd love to hop into as soon as that's done. Yeah.

Kushank:
Thanks for having me first of all. It's a pleasure to be here. I know we've been going back and forth about scheduling this for two months while I was away. So it's finally nice to be on here. Quick rundown, like, you know, I'm a typical Indian guy. My parents were like, "What do you want to do?" Dr. Engineering. So I was like, "I'll go to engineering school and figure my way out." So I went to engineering school, did computer engineering. I always liked building products. In school, I was building electronics. My last year project was like a wireless wearable posture correcting device because I slouch all the time. Vancouver wasn't super big on the hardware industry, so I was like, "Okay, I got to switch into software a bit more because, you know, there are more opportunities there." But I never really wanted to work as a coder all day, every day. I didn't want to be behind a computer, ironically as to lamp, but in different capacities. So I decided to look for roles in the consulting space. I started my first gig as a consultant and then I quit the job after 10 months because it was too boring.

They always hype up the consulting. Now it's like the product management piece, like, you know, how this is the most glamorous job. Funny enough, after that, I got into product management because I was again, maybe chasing that glamor that they always try to sell you in business school.

BLD:
At least you're honest about it.

Kushank:
Yeah, I mean, you know, it's like they say, "Hey, you do all this cool stuff. You're kind of building these cool products and stuff." So I got into product management. And then I realized that I just want to have my own thing. Maybe that's what I was chasing after all, which again: the glamour piece of it.

BLD:
Ha ha!

Kushank:
At least I knew, since my parents were also business owners, like, you know, in India, so I knew there's something, maybe I've seen that lifestyle. Like, maybe this is what I resonate the most with. Plus, I have too many interests, so, you know, it'll be easier for me to pick and choose what I want. Yeah, so that's how it really got started. And over the consulting, the product management, I was able to pick a lot of skill sets, you know, how to build products, engineering background, how to actually, you know, what to build, how to build, how to market. So I think all of that kind of came together. What I do right now is a combination of what has happened in the past. So that has really helped me where I am right now, even though... that was something I never intended, sort of thing, if that makes sense.

BLD:
Right, right. So, I mean, you're being super humble about this, but I think we need to get the scoreboard out there. OK, that, like, you're what? Seven hundred thousand followers across your socials, right?

Kushank:
Clothes probably are.

BLD:
Okay, wow. So yeah, you have, it is, we checked. Roughly the city of, like, your social following is roughly the city of San Francisco or Baltimore, right? So not a small number. I'm kind of curious to know then, like, between where we are now, and when you decided that you wanted to start your own thing, did you know what it was gonna be?

Kushank:
No, absolutely not. And what has happened right now with social media, that never crossed my mind because I used to hate social media. I was not active on my personal social medias from 2014 to 2020. Reddit was my vice. I was only scrolling on Reddit, no Instagram, no Facebook. I had, like, I built a few apps and stuff like that. The first one was called "Random Gift Idea." It was like Tinder but for shopping because you're always out of ideas.

Kushank:
I made One dollar off of it as an Amazon affiliate commission. So it was the first internet dollar. My brother and I started an e-commerce business back in 2019, which bombed. We made zero dollars in sales. So that was fantastic.

BLD:
Wait, okay, so your first internet dollar came in 2020.

Kushank:
Correct.

BLD:
Can you describe that feeling?

Kushank:
It was interesting. I mean, I didn't feel it was great because I invested $600 into building the thing and my ROI was basically negligible. But I think it was nice to get that $1.70 check from Amazon come in the mail. So it was still a big win, going from 2019 with $0. Then I had that one, which was $1. Then I had a little database product over the internet that made me $500-$600. So it has been a slower progression to getting to the next step.

BLD:
Well, I mean, look, one dollar off a $600 investment is a lot better than many SPAC companies were able to return, right? Results-wise.

Kushank:
That's true, that's true.

BLD:
So, I mean, at least I had revenue, which is, yeah, I mean, you're now you're whooping ass against a lot of publicly traded companies when you make your first Internet dollar. And yeah, I feel like that's a story with a lot of founders is that it's just like it's not even an overnight success that takes a decade, it's successively less and less failure that it just, you know, yeah, first is zero, and then it's one. And then it sounds like it was a few hundred bucks. And was all this under the banner of you doing your own thing?

Kushank:
I mean, all those products basically had different names and we're just trying different avenues. I didn't have the social channel that I had until that happened. So this happened after that slowly. 2020 is when the channel Digital Samaritan started and when I started was actually in a different name earlier.

BLD:
And so I just want to make sure I get these timelines right. You started Digital Samaritan late 2020. Your first internet dollar, is that before or after COVID really kicks off?

Kushank:
It was, so 2019 was the e-commerce business, which bombed.

BLD:
Okay, right.

Kushank:
That started in late 2019. We kind of ran it until early 2020-ish.

BLD:
What was the e-commerce business, by the way? What were you doing?

Kushank:
It was drop shipping furniture items, like home decor.


BLD:
Okay, yeah, I mean, dude, it's a bloodbath from what I've heard. Like, I think any easy dollar on the Internet just gets swarmed by thousands of people trying to do it. My favorite is the agencies that try to sell you a seven thousand dollar package. And boy, are they convincing. Wait, like agencies of what? You know what I'm talking about, like the Shopify agencies that are for the seven thousand dollars that try to teach you the course to actually do it, like the FBA courses and stuff. It's like Tai Lopez plus Shopify. Yeah.

Kushank:
But to be fair, all those failed adventures, even though investment went into it, this is a conversation my brother and I had at the time. Instead of paying for a course or paying to educate ourselves in a certain way, it's just much easier and better for us to invest that money into something that we're actually doing, putting the skin in the game, because then you actually learn a lot more. And since you have your skin in the game, you have a higher chance of success. It's just like, you know, rather than paying someone and going through a lot of materials and trying to do stuff off the material. So no-brainer. It's something I believe in right now as well. Rather than paying for a course, just honestly invest the money in what you want to learn by doing it.

BLD:
Yeah, yeah, exactly. And I mean, it's probably going to take about as much clock time or calendar time as it would to go through the course or go through school or whatever it is that you're doing. But if you actually go out there and do it, try to make it happen yourself out in the world, books are just an approximation of that. They're like that in a can. But would you rather have the fresh thing or would you rather have it pickled, right?

Kushank:
Totally.

BLD:
So, you know, I guess the question on our minds is, at what point did you see this generative AI wave coming? You were one of the first to kind of put it on social in a mainstream approachable way. Tell us about that.

Kushank:
I would say it was probably late 2020 as well. You don't have to wait until, I mean, of course, it got a lot more mainstream in late 2022, but it's funny because one of the things I was also dabbling into back in late 2020 was I thought I should probably start a podcast. So I had one episode with someone who's a product manager at freelancer.com. The second episode that I recorded, never published, was with Chris, the co-founder of Copy.ai. Just because at the time I was creating content in that space, so I did create.

Kushank:
And then we got on a call and chatted about that stuff. And that day, having the conversation, learning about how they're basically using the API from OpenAI to do all the work they're doing. And then I looked into OpenAI and what they're up to and all that stuff. So initially, that was the first time when I realized what's sort of coming. And around those weeks as well, synthesia.io was the one that also released their product and that video got mega viral, like half a million views. It was basically the video. I created a video talking about the video in the video. So the meta piece of it, I think, made it viral, which was honestly super cool.

BLD:
I mean, this is, I think, one of the things that seems to be one of your underrated superpowers is that you have kind of an uncanny sense of timing. Because, you know, to be able to call the generative AI wave in late 2020, like in hindsight, we all had the same data, right? We're all reading the same press releases and news articles. But it just became mainstream months ago. And it's like, that's the crazy part, the people that had the foresight back then. It's pretty impressive. Yeah. Were you in any founder communities that had, like, who inspired you to get on this track in the first place?

Kushank:
Like for creating content?

BLD:
No, for getting involved with generative AI. Yeah. The generative AI and design and marketing intersection beat is a very well-established path right now.

Kushank:
Mm-hmm.

BLD:
And I would say it's a very crowded path. But you've got content. You have content with timestamps from 2021, early 2021 I've seen where you're like, look, here's how you can make a better banner image using AI than you would with Canva. And it's like for one out of the thousands of tools that you overwhelm us followers with. Yeah.


Kushank:
Well, I would say I never really positioned it as generative AI. I prefer to view it as technology in general. I try to avoid using buzzwords as much as possible. Throughout my career, I've always adopted technology to make my work easier. I'm inherently lazy, so leveraging technology to save time and bridge my skill gap has been crucial. That's the reason I create content about it, to show others how they can leverage the same technology to advance themselves and add more value to their teams. So, back in 2020 and 2021, it was all about using technology to save time and money.

BLD:
So that's how you were thinking about it. Like, wait, I can actually get a banner image cranked out way faster.

Kushank:
Yeah. Or like everyone's using Canva for slide decks. Let's find something else with cool templates. Enter pitch.com. Those templates are pretty sick, and not many people are using them.

BLD:
Oh yeah.

Kushank:
If I present those templates in a meeting, people think I've done all the design work to come up with them, but they don't know I just used a different platform for the templates.

BLD:
Yeah, yeah. I mean, the number of times we get on a call with a founder and they're like, I need help with this pitch deck. And they're like, can you improve the design? And they're like, well, I'll send you the Google slides. And it's like, hold up, I'm using Word 98. Yeah, yeah, exactly. There's a program called Figma or pitch.com. And pitch.com is almost like, if Figma is a motorcycle, pitch.com is like a tricycle. God forbid their designer ever leaves. Yeah, seriously. So for you, it was really about the fundamentals. It wasn't about hopping on any hype train. You recognized the common use cases and saw new ways to address those needs, like a better pitch deck or a banner image. And AI became a logical solution for you to explore.

Kushank:
Basically, yeah.

BLD:
Wow.

Kushank:
Then, similar to all the technologies out there, it's about keeping up with it from a product perspective. I want to know what people are building, how they're going to sell it, and whatnot. Having my head in the game that way also helps.

BLD:
There are so many products out there. Sometimes it can be really overwhelming. Do you have a set of criteria nowadays for building conviction around what tools are worth your time and what's just a fad or developed by one person who maybe doesn't care that much? Ha ha.

Kushank:
Oh, that's an interesting question because I've been discussing this with my team recently. I honestly wanted to post content yesterday that, like, half the tools out there right now are horseshit. That's why I haven't been posting about it as much because I'm just not feeling as inspired anymore. Everyone's kind of using the same API and you have no-code website builders. You plug in the API at the back end and you have a product. So, to answer your question about the criteria, we're looking into the use cases. The impact you can create, if it's actually useful versus just cool. Because a lot of those tools, what they say they can do is cool, but half of them are in waitlist or not fully there yet because the technologies are relatively new. So we're trying to dig into what is something that will actually make an impact and make your life easier, rather than just wasting your time with the free or paid versions.

BLD:
So, what do the tools that do end up making an impact have in common compared to the ones that are more like vaporware?

Kushank:
Most of the time, it's the product itself. They're solving a real problem. I'm trying to think of an example that comes to mind. I haven't posted a video in maybe a week or two, so it's hard to say, but

BLD:
You've been traveling around the world.

Kushank:
Just here, just here.

BLD:
I love this guy's stories.

Kushank:
I mean, in competitive spaces, like video editing tools, there are so many out there, and the image generating tools that can replace backgrounds and whatnot.

BLD:
Mhm.

Kushank:
So those are kind of going out of the picture. I get so many emails about SEO blog ranking tools, which there are so many of. So, in those super competitive spaces, if you don't have your own model or a unique approach, it usually gets discarded right away. Whereas something that is more novel, like a tool where you can chat with your PDF, can solve a problem that many people have. I've been in situations where I have a giant PDF and need to search for information, so being able to save time there. So that helps. And you can just kind of go back to basic stuff like product management. If it saves you money, saves you time, you know, that's something that's a win, like a winnable product. So kind of like keeping those basics in check usually helps us like figure out which one's kind of worth our time.

BLD: PDF reader. Oh, yeah, I'm talking about PDF AI. It's a plug. You're talking about the plug-in in OpenAI's chat, right?

Kushank: There's this plugin, there were a couple of tools that were built on top of it as well, but it's just more like a nice use case that someone has built just using the same technology.

BLD: Yeah, I mean, I run an out-of-band new me context newsletter and vivid leaves. Agree.substack.com. You check it. And it's like history focused, right? I'm really into history. I like reading a lot of different episodes of history. And I wanted to do a link drop, where I would just post interesting links that I've And I realized that just the link in the title wouldn't be very helpful. And so I actually checked out one of these PDF plugins for, it wasn't OpenAI's one, but it was just PDF.AI, which probably is honestly still consuming the OpenAI back end.

Kushank: I'm sure.

BLD: Or the API.

Kushank: yeah.

BLD: And it was incredible. Yeah, I was just like summarized this weird article about Soviet banking history in 500 words. And it had an exquisite summary. So I feel like the product management mindset of looking like, OK, besides the hype, besides the beautiful animated gradients, will this thing actually save me time, money, you know, enable me to do new things that I wasn't able to before?

Kushank: Yeah.

BLD: When you're talking to a pro, part of the app, it's already going. I'm going to go ahead and turn it off.

When you're talking to product teams or founders and they're trying to, you know. I'm not actually totally sure how you work with product teams or founders, and I'd love to hear about that, but I would imagine that they look to you for tooling. And I would also imagine that since the underlying technology is all from OpenAI, are we moving towards a world where all the copywriting, all the design, all the marketing, all the same underlying technology.

Kushank: I don't think so. So I think, you know, how people say that the AI is going to take the jobs away or whatnot. I just feel like it's going to take away work or livelihood from people who are like the average, because, or who don't strive to be more than average. The reason is, let's say you have a website and you want to rank for SEO, you can use technology to write yourself like 30 blog posts a month without lifting a finger. But at the same time, everyone else can do it. They all have access to the same technology that you do, which means the bar of standing out just goes higher. So now instead of writing this blog post, you're like, okay, what else can I do to stand out? So I think your creativity and creative mindset need to stand out even more. So it just becomes harder and harder for everyone to do things in a new novel way, to be more creative. Even if you want to use strategy, you need to be creative, at least being able to talk to it in a way that you can get those creative ideas. So you can always be ahead of others because that's just the human way of us to always be competing with each other. That's how I feel what's going to happen. Like there will be designs obviously, but the winners would be the ones who would then go beyond that and see how they can incorporate the technology and their own creativity to stand out from the rest.

BLD: So then, I mean, you're probably in a uniquely good position to give advice for people on how to stand out because Digital Samaritan only started, like, what? 2 and a half years ago?

Kushank: Yeah.

BLD: And yeah, you now have a solid mid-tier American city's worth of population that you can press a button and blast onto all of their feeds. So how did you get Digital Samaritan to stand out? And do you think there are any lessons there for people who might be looking for ways to stand out as the middle becomes totally homogenized by AI? Like, are there any lessons for people looking to stand out like that?

Kushank: I think, it's gonna sound cliche, but just being true to yourself always ends up winning because that's something that's unique about yourself and no one can take away from you and no one can repeat it. So I think a very broad high-level cliche answer would just be staying true to yourself and your roots. Like what I'm able to do in the content comes from the background I had, being a product manager, dabbling around building my own products, and working in consulting with an engineering background. And that experience that I've had is unique to me. Plus, Indian upbringing and entrepreneurial parents. So

BLD: Mm-hmm.

Kushank: No one can take that away from me. And no one can repeat that piece. So I think that is also my competitive edge moving forward, even though everything else is the same. As you've probably seen, there are a lot more channels in the same space now, similar content, honestly, a lot of them have even better content than I do. The quality of the content, the research, it's incredible. So what I need to do now is just again understand why I did what I did. What is my goal? Why did I start it? Just staying true to that versus following what other folks are doing? It's probably what I would recommend to myself and to others as well.

BLD: And so when you started Digital Samaritan, did you know the lane that you were going to come for in terms of being almost like a one-stop shop for educating people about the intersection of design and marketing and increasingly AI?

Kushank: Absolutely not. It's a funny story. I started this as a channel to prove to myself I could be persistent. You know, the product I told you about that made one dollar and 70 cents. I thought I was not persistent enough because I wasn't able to grow or whatnot. So I was like, oh, I can just now quit because I just bail, sort of thing. And then everyone used to call me the idea guy, and I was like, no, I'm not the idea guy. Idea guys are the worst. I want to be the execution guy, the guy who actually gets the job done.


So those two things, along with COVID, everyone had their phases. I was going through the phase of getting out of my comfort zone. So in September, I was like, I hate roller coasters. Let's go for skydiving. And then in October, I was like, okay, I hate social media. I want to be persistent. I want to be executable.

BLD: You definitely took that challenge on. Yeah, a little bit more than a phase.

Kushank: Yeah, well, it's a phase, still a phase. It's just a longer phase because two years don't count of COVID.

BLD: Yeah, no, that's true.

Kushank: Yeah, so anyway, I was like, okay, let's just do 30 videos on TikTok. I'm not gonna tell anyone, I'm just gonna do it so I can prove it to myself that I could be consistent, I could do something that I'm gonna set my mind to. So for 30 days, my channel, the same channel, but it was called "doing whatever every day" because I had no idea what to do. So it was just like me pointing toward Google images. I was like, hey, this could be a cool business idea. Like one example was, you know how you have those expandable and collapsible laundry bags?

BLD: Yeah.

Kushank: So, like, that before changing at the beach. So a bigger version of it, you can just change at the beach. So it's like, oh, this could be a cool idea. So it's just talking about random stuff really for 30 days. But then it kind of became a habit.

BLD: Wait, and so, wait, and so it was literally the original TikTok channel was just kind of like Kushank's business ideas, like almost like requests for startups.

Kushank: Basically.

BLD: I love it. Are those still up?

Kushank: No, unfortunately, I don't know why, but I deleted those. I didn't even private them. I deleted those videos, which I'm like so choked about. I wanted to see those videos again.

BLD: No, those are going to be like toddler photos or toddler pictures of this growingly monster, like a social media presence. It's like, here. This is me pointing at Google image results of, like, did anybody think of making one of these things at the beach? And now, you know, one of your videos goes up and, what? I don't even know how many thousands of likes it gets. Right. So like, what was that journey then? So you started saying you want to do something, you want to do something to prove to yourself that you can be persistent. And actually, this is an insight that Harrison and I had about ourselves and about NUMI this year, was that the only things that seemed to really be working were the ones that we could just kind of do on a predictably consistent schedule.

And so now when we think about even channels, like channels for customer acquisition and things like that, we have what we call consistency experiments because we know that it's going to take a while to figure out a channel. But we just don't trust ourselves. Yeah. And so, so we don't even test before we test whether or not the channel is going to work, we actually test, can we be consistent in that channel and we define what consistent counts as and then, you know, from there we see if like, you know, results will, will bloom.


Kushank: That's really smart.

BLD: But it sounds like what you're doing, right, with the TikTok.

Kushank: I suppose, yeah. For 30 days, as I was saying, what happened was I had 20 likes over 30 videos. That was nothing. But for the first time, I was measuring my input and not my output, which, as you said, the consistency piece of it. Before with the other startup, that micro app about Tinder for gifts, where I made a dollar 60 cents, I was like, I want to make a hundred dollars in one month. So I was chasing the output.

BLD: Mm-hmm.

Kushank: But for the first time, I chased the input and actually felt accomplished. So I think that had the mental shift in my brain. It was like the aha moment. I was like, interesting. And then I just also kind of creating those quick little TikTok videos became a bit of a habit. And I think a month later, one of those videos about Microsoft Clarity, because it just came out. And I knew about Hot Jar, Microsoft Clarity was free. We talked about how all these companies use this analytics tool to track where you're clicking and whatnot, and this is free. That video blew up. It got like 10K views. And I was like, holy shit, like this is insane. Am I famous now or something?

BLD: Yes, you are. What was the hint from there then? Was that when you discovered maybe this could be your content lane? Is that when you discovered things like that?


Kushank: Basically, when I talked about that, I mentioned other stuff, and the views were going up, likes were going up. So the dopamine hit was coming in. Basically, they chased the dopamine hit for a bit until it just kept on growing from there. That's how it started.

BLD: Kind of changing lanes here. So, you know, you've worked with companies before, Notion, Zapier, Google. What types of questions are they coming to you with?

Kushank: It really depends. Sometimes even with the bigger companies, there are intermediary parties that come to you because they're interested in the campaign piece. But sometimes there's more than that. For example, a brand wants to create more awareness about a certain topic.

BLD: Hmm.

Kushank: So, how do we educate people on the topic? We need more signups or the whole campaign is about increasing user signups, sales, or adoption. It varies based on the brand, their vision, and the campaign they're running.

BLD: Wow. And did this kind of set up an open-door policy where more and more brands started DMing you as you established yourself as this product guy on social media? Or was it more like you trying to form those connections outbound?

Kushank: No, it was all inbound because I didn't even know you could get paid for this stuff. A few months in, I got an email from a company asking if I do paid collaborations. I was like, wait, are you saying I can get paid for doing what I'm doing for free? And they said, yeah, I do paid collaborations.

BLD: Hahaha!

Kushank: So yeah, it's mostly been inbound, which I'm super fortunate about.

BLD: And what are the problems that you and your clients are working on? What's important to them right now?

Kushank: Right now, it's either awareness or user acquisition. So if we're speaking specifically about the clients that come for Digital Samaritan sponsorship, it's awareness and user acquisition. Awareness would be the top of the funnel for the eventual goal of user acquisition and education.

BLD: And without giving away your secret sauce and what people pay you for, what does that framework look like? How does that conversation start? Does it start from a marketing perspective, design perspective, generative AI perspective? Where does it come from first?

Kushank: Mostly from a marketing perspective. The CME is an avenue for marketing the product or technology. So the conversation starts there, and it's just like, you know, they see, or we just create the content the same way we're doing anyway. So it feels more authentic to the users, and they're creating the awareness they desire.

BLD: And the tools that you suggest to them, do you end up suggesting tools for them to use, or are they baking it in, or is it purely just raising awareness on the channel?

Kushank: It's mostly raising awareness on the channel. And then I look into the tool myself and see what audience we can target, what is the use case, what problem can we solve using a particular piece of technology? That's what I try to talk about in the content, like here's how we can solve problem X, for example.

BLD: Question. I don't know if you have one, but like, what are some of the use cases that are top of mind for companies right now? The easy ones are TLDR, PDFs, email summaries, condensing text, like the standard. But what does that next layer look like for businesses?

Kushank: That's a good question. Right now, it's all about generative AI, like text-to-text, text-to-video, text-to-design, basically text-to-whatever, or even audio-to-whatever. So for the next thing that's going to happen, I feel like it's not going to be about having the same text-to-something technology because everyone will have it. Instead, it will be about the quality of who's better. Currently, everyone is using the same API and trying to release their product as quickly as possible, so they may not be doing their own engineering on it extensively. The companies with bigger budgets are likely building teams and improving their models to achieve the promised results in the promo videos. That's what I think the next stage will be.

BLD: It's funny that you bring that up because I was talking to Brittany's younger sister, Michaela, who just started her internship. She was asked to do some diligence on various AI tools for writing content. Naturally, I had opinions being in the space. It really came down to OpenAI and Jasper. OpenAI is the better choice if you're a skilled prompt engineer, while Jasper AI is great for layman marketers. So where is that gap, and how is it closing in?

Kushank: I believe that tools like Jasper or Copy already have wrapped prompts within their product, specifically for SEO writing or blog writing. This optimization allows for faster results. It's similar to building a website using React directly, where you have more control and can create a better website if you have the coding skills. On the other hand, using a no-code tool can still result in a great website, but with more restricted customization. The same applies to AI tools, where going directly to the source gives you more control over the prompt and input, compared to using pre-built prompts provided by tools like Jasper.

BLD: So, do you think that in the AI product space, the focus will be on productized prompt engineering? Rather than simply targeting specific use cases, the differentiator will be having a set of intuitions on how to massage the LLM to align more with user inputs and expectations.

Kushank: Yes, I would say so. Every product aims to keep users engaged within their platform for as long as possible. As a product-minded person, you would consider the moments when users have to leave the platform to perform certain tasks and how those tasks can be integrated into their workflow. For instance, Salesforce has been releasing numerous chatbots recently, likely to address the issue of users going to external platforms to solve problems. They want to make it possible to handle those tasks within their product. However, this approach brings up questions about pricing, API costs, data sensitivity (especially for governments and public organizations), and the choice between building models into the user environment or using a cloud environment. These are additional challenges to consider, in addition to privacy concerns.

BLD: Yeah, yeah, this is one of the things that's been interesting to watch. Everyone's scrambling now to apply these models to their products. And there's clearly a reckoning coming around the curve, around the bend of the economics of AI-enabled products looking totally different from cloud products, where it's like, 90% margin, baby, let's go get a couple of Stripe subscriptions and we're off to the gravy train. But with AI, you might be looking at 0.0001. I mean, yeah, exactly. For token, you might be looking at stuff that's like, 40% margins, it looks a lot more like basically an employee that never gets burned out and is always showing up on time. But it's not like the crazy high inference is so expensive, right, from a compute perspective.

Kushank: Well, I think that would really depend on the volume that's been used and how it's incorporated. Because using or incorporating AI into your product might not cost as much. If your user base is lower, the volume of queries you're putting is lower.

And let's say you're going to do 3,000 Instagram captions with social media manager on a tool. That's not going to cost you a whole bunch. But the question is how good the results are. Is it also gonna help you keep this cost lower because then people have to run fewer and fewer keyword queries onto the product. Of course, there's a cost to it, but I feel like it's not too much to really hit the margins that much. But again, that's like you're renting the platform from OpenAI, you know, like the whole Reddit drama, what if OpenAI says, "Oh, are we gonna bump up the cost for tokens this much?" Then you have a platform risk right there. So I think those are going to be interesting ones too, how companies, bigger companies, enterprise software, think about pricing those things.

BLD: I'm not sure if that's a good thing. That goes into thinking about that. It's a very real thing, right? That people could be losing their livelihoods.

Kushank: Yeah, no, I agree. It's interesting you mentioned this because my content in the next few weeks is gonna have a little bit of a pivot, addressing more upscaling from that perspective for folks to be more competitive with their skills and retain their jobs. If you're hiring, there will be some jobs that will just shift. They won't be eliminated. New ones will be created, and old ones may disappear. It's like any revolution, you know? For example, we have elevators that operate by themselves, but in some countries, there's still a person pressing buttons for you in the elevator. They probably got displaced by technology, but hopefully, they found something else. Now, they're escalators.

BLD: They repair escalators.

Kushank: Right, now the point is, as long as you're open to learning the new tricks and leveraging the technology, I think you'll be okay. To use these models and tools, you need certain skills. If you're the one setting up those things in your team, your older bosses won't be able to fix them. And you won't hire an external consultant for that. So if you can use those tools yourself and save time, that's probably a better bet. Also, I'm totally against the idea of having to work 40 hours a week because it's not the industrial revolution. Your input is not proportional to your output in terms of hours. We're not manufacturing things. It's all creative and thinking work. So if you spend more time thinking, hopefully, you'll generate better output and productivity for the team with advancing technology.

BLD: And where are the immediate areas in a business where AI could provide lift?

Kushank: One depends on the function. So let's say if you're in operations, you can combine Zap here with OpenAI, and there are lots of opportunities. A way to think about this is to break down the tasks you have and the tasks you collaborate with other teams on. Think about if you can do those tasks yourself using technology. For example, if you're a product manager and you need to work with a design team, are there things you can now do using technology instead of bothering the design team? That way, you can be a more contributing part of the design team rather than just a liability to them. That's how I would think about adjacent functions and how you can contribute to those teams instead of asking for help.

BLD: Right. One thing that always interests me about the generated content is the quality or the gap, so to speak.

Kushank: Yes.

BLD: Like the example you gave of unblocking yourself using AI so you don't have to bother the design team. Let's say you want to create a piece of social media collateral, and the design team has a brand book. This is a place where I found it hard to fully dive in headfirst despite all the excitement around AI. If I ask Mid Journey to produce social media collateral but can't manipulate the layers because it only spits out an image...

Kushank: Yeah.

BLD: How do you bridge the gap between what the brand book says the collateral should look like and what Mid Journey can produce?

Kushank: Well, hopefully Canva's new feature allows editing of Midjourney layers or something, or maybe in Figma.

BLD: Oh, so you're suggesting there's a new feature?

Kushank: No, I mean hopefully Canva should have something that can edit Mid Journey layers, or there might be a way to split the layers within a flat image.

BLD: Ah, got it. I thought you had beta access to something we didn't know about.

Kushank: No, no.

BLD: Yeah, I was like, did they? Did we miss something? Because I'm quite updated on the design software front, and I haven't heard about this.

Kushank: No. I'm sure there's some way to do it within a flat image.


Kushank: Somewhere they're thinking about how to separate those elements and be able to do that. And I think that's a great point. Going back to your earlier point about where these tools are going, it could be more usable and editable. For example, as you said, you feed your brand guide to it, generate something, do a check if it matches the brand guide, and then you can manipulate the layers. So it makes sense to be able to manipulate. You can easily manipulate the text generated content, but what about video, image, and audio generated content?

BLD: Yeah, one example that comes to mind is Framer. I actually put numi.tech through it.

Kushank: Oh, really?

BLD: Yeah, it was just an exercise to see how good it is. I always check up on Framer because they kind of ebb and flow. It spit out a less than ideal representation, but I saw it being built in real-time. What's super cool is that it gave somewhat of a brand Bible on the right-hand corner. I was able to select what I wanted, select new colors, add copy, define tone, and it would regenerate. I mean, this is pretty impressive stuff, but it doesn't have the same pedigree as our designers, of course. Is it just a matter of time? I mean, I was saying years before, but is it now a matter of months? This brings me back to the wide spectrum of use cases that AI has shown me.

Kushank: Right, right. I mean, at a manager level, you can declare the need for a work artifact with specific qualities, and you can either do it yourself and fine-tune it or hand it off to an LLM. Then you can come back as a human and bridge that quality gap. But going back to the guy pushing buttons on the elevator, there was a phase where elevators had steam levels to manage and gear shifters to work. We needed that. If I'm renting in a building, I don't want to know how to operate the elevator. But there was probably a phase where it wasn't that bad, but still not something a typical tenant could handle.

Kushank: Yeah.

BLD: I wonder if there's a rising labor need to take the raw outputs of these LLMs, learn how to work with them, and bridge the gap between what the stakeholder wants and what the LLM can deliver. It's like the last mile logistics concept. The breakthrough technology of shipping containers revolutionized the world but can't get you an Amazon package in 48 hours. It still requires fine-grained human involvement.

Kushank: It's like what we discussed earlier, where everyone can get the same designs, but it's your creativity on top of it that makes the difference.

---

Kushank: It stands out from the rest. And then even Firefly from Adobe is making waves. But to think about what the background should actually be, that's still something you can get ideas from. Chat GPT or genera can provide inspiration, but actually making a difference, whether it's for artists or graphic designers, requires a human touch. User studies and feedback cycles supported by tools can help analyze and improve, but a human element is still necessary. We'll try to operate with as much human element as possible, but we'll also push the limits and see what we discover.

BLD: Yeah.

Kushank: It's like a calculator. It can be in the hands of an eighth-grader during math class, but it can also be in the hands of a famous mathematician solving complex problems. Long division by hand is not commonly done anymore, except maybe in gifted and talented math classrooms in fourth grade. But once you learn it, you realize it's just arithmetic practice. They weren't teaching you how to divide complex numbers in the real world.

I feel knowing the basics helps you understand what's possible. If you know math, you know how to approach problems. If you don't know what's happening behind the black box, you don't even know what's possible.

BLD: So true. I want to step back because I found it interesting when you advised founders or teams to start by leveraging AI in the way teams communicate with each other. What other areas come to mind in that same vein?

Kushank: Another area would be looking at repetitive tasks and leveraging automation. We use Zapier for many automations, even for our newsletter. We have a campaign that goes out to every listed tool, so the tool builders are aware of it. It's like an extension, a freelancer, where you start with tasks lower on your priority list. It's not as impactful, but it's better than nothing. Start there and if you get the desired results, you can scale up and improve. That's one way to approach it. But the real power lies in merging AI and automation together instead of just focusing on AI. With automation, you're not just providing input. For example, Coda's new feature allows auto-filling a table with blog post ideas based on a formula. So using those tools in conjunction is the superpower.

BLD: I love that. It's not just about automation; it's also about prompting AI to be creative based on your own creativity and merging those worlds together. Do you have any examples that come to mind when you think about the impact of AI in the workspace?

Kushank: I think if we're unable to improve the quality of the generated response... There was a time, around 2017-2018, when Alexa apps were really hot, and people thought it would be the next big thing. Building apps and making money from it, but it never took off. The user interface and practical use case didn't thrive due to the friction involved.

BLD: You don't want to accidentally order 48 Tide Pods by voice every night.

Kushank: Exactly. And with Alexa, we witnessed collective societal patience evaporate. Thousands of engineers working on it, but the user experience was frustrating. It asked for specific commands and explanations, putting the burden on the user.

BLD: Okay, let me remember this through my Black Mirror seed. That's never going to happen.

Kushank: Right. I remember spending a whole weekend at a voice app hackathon in 2017. We were trying to build things there. If we can't eliminate human input and improve the quality of the output, some fads might disappear. Chat GPT is here to stay, but other failed ventures might disappear.

BLD: So it seems like inputs are the future. The better the inputs, the better the outputs.

Kushank: Yes, but we all want shortcuts. We're not good at everything. So being able to generate better output without doing the hard work is what we desire.

BLD: The prompt engineer plugin, yeah.

Kushank: Yes.

BLD: I feel like that's a recurring theme in our conversation. The quality of the outputs will determine everything from here. There's a gap between what people want or need in a production environment and an LLM that may be hallucinating or acting like a skilled bullshitter in a seminar class.

Kushank: Totally.

BLD: Yeah, they still have the disclosure at the bottom saying it may not be factually correct.

Kushank: It's like those AI video apps like Synthesia. One of them is now a billion-dollar company. I'm sure they have customers, but what are the use cases?

They raised another round and now it's valued at $1 billion. What are the use cases? Customer support, internal training, maybe sales. But how effective do you think that's going to be? It still feels very robotic. If you're in customer training, it's already boring enough. Internal training with a robotic voice lacking emotions will put me to sleep. From a corporate perspective, it makes sense to produce all these training materials in video format for employees, but will they actually watch and learn from them? That's the question.

BLD: Yeah, what was the big legacy company that recently replaced their HR department with generative AI bots? I'll have to look into that, but yeah, I'm sure that went very well. Oh, sure. And then Buzzfeed, we all know that story. They literally replaced droves of people with bots. There's a time and a place, but HR is probably not one of them, right? Yeah. I also saw today that one of the British newspapers, maybe The Guardian, was going to start using AI-generated text in the suggestions or recommendations, not in the body of the articles, but maybe in titles or summaries. That's a good use case, but it's interesting. Journalists saying this is probably coming for our jobs, and either we can get eaten by the tiger or we can ride it, right?

Kushank: Yeah.

BLD: I had a conversation with the former chief financial correspondent of the New York Times, and questions came up about people taking AI-generated output as fact without dispute. When you ask the AI to cite its sources, it's not accurate. It may make up names or provide incorrect information. There are serious implications and hazards there. In a business environment, what happens if too much of the workforce relies on something that may not be factual or accurate?

Kushank: It becomes a liability issue. I wonder how it will be addressed in contracts, especially for customer-facing roles or when sharing potentially harmful information. There have been talks about publicly mentioning if something is AI-generated, which would be a good precaution. However, humans also make mistakes, possibly more than robots. It's an interesting debate on which way things will go. Ultimately, it boils down to liability, and corporations will try to distance themselves from it as much as possible.


BLD: That's part of the other MO, right? They're like liability minimization machines.

Kushank: Yes.

BLD: Good time to be a lawyer.

Kushank: It is.

BLD: I don't want to politicize this, but there's something to be said about how Sam Altman carried himself in front of Congress compared to, let's say, Zuck. Clearly, there's a dialogue or discord around the people behind these technologies. They're the innovators, but they also bear the responsibility for world change.

Kushank: True.

BLD: AI is going to be the topic for the next 10 years, touching every part of our lives. Companies need to go to Congress for policy or things could go really dark. What's your take on private companies without oversight? Pros and cons?

Kushank: The pros are obvious—they can innovate without restrictions. But I lean towards the cons because most of these venture-backed companies are driven by profits and revenue.

BLD: Hopefully.

Kushank: Hopefully.

Kushank:
So I think whatever decisions we make, companies like Facebook repeatedly violate privacy policies and end up paying fines. They know the ROI of violating policies, so they're happy to pay and make the problem go away. I don't know if policy makers can catch up with what's going on. It might come down to us being more conscious of how we use and let them use it. Companies won't shy away from violating privacy policies if it helps them make profits.

BLD:
The only examples that come to mind are Manhattan Project, which gave us nuclear power, and the Navy, which gave us the internet or Tor. AI was started by an open-source project and has now become privatized. We're entering a dangerous place where private corporations are not aligned with what a government should be. What are your thoughts on that? Starting from Tor and the US Navy, the Manhattan Project gave us the A-bomb and nuclear technology. Now AI, the most mature version, is coming from a privatized for-profit company.

Kushank:
A good example is the voice cloning apps out there.

BLD:
Yes.

Kushank:
We considered using that tool to create content faster without voiceovers, but we decided not to. I feel scared about that technology. Social scams are already happening, but imagine if my parents received a call with my voice asking for money.

BLD:
I've heard about this.

Kushank:
That's terrible, exactly. Those companies in charge could sell the technology without doing due diligence. Even shell companies could exploit the technology for scams. Selling the technology is different from the restricted access in the Manhattan Project. Now, both bad and good actors have the same access. We need to be conscious of what we get ourselves into, both as a company and as individuals.


BLD:
Yeah, wise words. Companies can land themselves in a world of hurt.

Kushank:
They could sell data for profits or whatnot, so you never know.

BLD:
Yeah, yeah. That totally became a thing during the later innings of Web 2 and socials. They were talking about those shady data brokerages like Cambridge Analytica. That whole debacle was a shining star example. So, closing things out then, Kushank, what are you most personally excited about in terms of the product landscape and the AI landscape over the next 12 months?

Kushank:
Ooh, that's interesting. Honestly, I'm super excited for the adoption. Digital adoption has been going on for several years, but we might just be hitting a mark where the awareness is at a level where more companies will accelerate their digital adoption. So, that's something I'm excited about. Also, I hope that we can realize that people don't need to work 40 hours a week at a company to do what they do. The extra time we get can go toward balancing work-life and making more features for this product.

BLD:
Famous last words of the Industrial Revolution. We don't have to work anymore. Greedy people are going to work more and more efficiently.

Kushank:
Yes.

BLD:
Best believe that.

Kushank:
Yeah, I think the adoption of the technology, bridging skill gaps, and ensuring job security while increasing efficiency.

BLD:
That's where you come in. You sell upskilling. Could you tell us a little about what you do and how you can help?

Kushank:
Of course, thank you for the plug here.

BLD:
Ha ha.

Kushank:
Yeah, my content focuses on using technology to improve job performance. Future content will also cover how people in marketing and operations can make their teams and themselves more efficient by leveraging technology. They can make themselves more hireable instead of facing job loss.

BLD:
Awesome. Amazing. Want to take us out? So guys, that's it. Thanks so much, Kushank Agrawal of Digital Samaritan. This has been Bottom Line Design. I'm Agree. I'm Harrison. And we'll see you next episode.

Kushank:
Thanks.

BLD:
You're welcome.

Kushank:
So much for having me, guys.

Tap into the most driven engineers and designers on the planet