S6E4: Zero Budget, Six Figures Later with Razy Shah

DISCLAIMER: This transcript was generated by AI and may contain errors. 

Guest: Razy Shah 

Season 6 of Productly Speaking 

--- 

[00:01] Karl Abbott: 

Welcome to Productly Speaking, the podcast with real stories from real product people about the messy, surprising, and occasionally brilliant work of building things. I'm your host, Karl Abbott. Here we have no perfect processes, no polished answers, just honest conversations and lessons learned the hard way. Let's give this a go. AI is making it easier than ever to execute, to move faster, to ship more. But when execution stops being hard, something else becomes visible. The judgments we make when nothing is forcing our hand, the boundaries we keep or don't, and the kind of people we choose to be at work when no metric is watching. Today's conversation is about one of those moments. Razy Shah is the co-founder of a digital marketing agency he's been running for over a decade. But this conversation isn't about tools or tactics. It's about a moment early in that journey that didn't look important at the time, and what it reveals about how we make decisions now, especially as AI takes over more of the execution. Razy, thanks for being here and welcome to Productly Speaking. 

[01:07] Razy Shah: 

Thank you for having me on the podcast, Carl. 

[01:11] Karl Abbott: 

Yeah, so Razy, before AI entered the conversation, what did doing good work feel like in your day-to-day? 

[01:20] Razy Shah: 

I run a marketing agency. We provide services to small businesses, to startups, and even to large companies. And what good work felt like was when we would create an excellent campaign. So this would mean writing a copy, headline that converts, creating visuals that also convert, that the client would look at and say, this is a really nice visual. We like it. It's getting us leads. It's getting us conversions. So before AI, all of these things were being done by people, and it showed your experience. It shows your expertise in the craft. So that's how good work looked like. But now, as we all know, AI writes the headline. AI does the creative. AI optimizes the ad campaign for you. The other thing that good work looked like was, once you set up the ad campaign, are you optimizing it? Are you getting the results the client is looking for, the sales, the leads, the conversions? That was good work. But now even, if you look at Meta, if you look at Google, AI, their AI can do this as well. So that's what good work looked like in our industry. And now all of that is now being done by AI. 

[02:29] Karl Abbott: 

So do you remember at what point that execution started to feel less special, like it just stopped being the advantage that it used to be? 

[02:38] Razy Shah: 

Actually, very early on. So in December 2022, I was already tinkering with ChatGPT. Jan 2023, I think it became very mainstream. And by May of 2023, so we used to have this account where we would write content for this insurance client. We were writing content for them for five years at that point. And in May 2023, they said, hey, look, our HQ has decided that we are just going to use ChatGPT because ChatGPT is going to be cheaper, faster, quicker, better. It's going to understand our products better than a bunch of marketers can understand our insurance products. And yeah, so that was the first moment. A contract has been there for five years being no longer renewed because ChatGPT could do the work better. And that was the early phase, right? So early phase of AI was written content. So a lot of people were using ChatGPT. And I remember back then when you generate an image of individuals or objects or people, it looked really strange because they'd have six fingers, seven fingers. But now if you look at the kind of visuals it creates, I mean, it's just mind-blowing. 

[03:46] Karl Abbott: 

Yeah, it really has gotten there. And I mean, I'm sure you've seen it, but Will Smith eating pasta is like the video standard of how AI has been. And the first versions of that are really bad. But the current versions of that are scarily good. 

[04:04] Razy Shah: 

Have you seen the one from Hicksfield? So Hicksfield just did one recently where it's Will Smith eating the spaghetti. And then someone, one of the AI actors walks in and joins him. And it looks like they are really by the beach somewhere. And it looks like it was filmed. 

[04:19] Karl Abbott: 

Yeah, I think that's the one I saw there by the beach. And it's just so good. And I mean, like you can't, you almost can't compare the start and the end. There's that much of a difference between the two. And the new one, it's kind of like, oh, this is dangerously hard to determine what's real or not here. Yeah. So what part of that shift, because you went through it, you saw it with your client. What part of that was personally unsettling, even if you couldn't name it quite yet? 

[04:49] Razy Shah: 

What we are doing is we're doing a lot of execution. So what happens is when our clients come to us, usually they get, I mean, they're not marketers. They don't have an in-house team. They are very intimidated by the platforms. Like if you go into Google's Ad Manager, if you go to Meta's Business Suite, they get very intimidated by the platform navigating it. They get very intimidated by all the jargon, you know, CPL and things like that. So when execution got cheaper, what was unsettling for us was that, okay, look, now these clients, they can actually do the work themselves. You know, they don't need an external agency to do the execution. So the execution, which used to be something, it's a craft, you know, that we have perfected over years. We have hired experienced people. Now this execution can be done cheaply with a simple prompt, right? So what was unsettling for me is what is going to be the demand for these services going forward, you know, and not just the demand, but what is the price that a client is willing to pay for these kinds of services? And the clear answer is that the demand, the demand might still be there, but the price that they're willing to pay is going to go down. Right. So I was just, yeah. So that's, that was what's unsettling for me. So what I was unsettling is in short is the longevity of the marketing services agency business. And also like, how do we pivot? What do I, what do I do to prepare for an age where marketing services can be done cheaply, can be done quickly. And the, you don't need such a big, you don't need such a big team to execute. 

[06:28] Karl Abbott: 

Yeah. And so we're a couple of years past the point where that client decided to start moving more of their business into AI. Have they come back or has that continued? 

[06:39] Razy Shah: 

No, they've not come back. They've just gone fully in house. In fact, I was at a marketing conference two days ago in Malaysia and there were some CMOs of some very large Malaysian companies there. And one of the things they were saying is that instead of giving the work to an agency, what they have started doing is they've started doing it in house, hiring in house, and then using AI to empower that particular staff to do more, to do better. 

[07:08] Karl Abbott: 

But the idea that I guess being that their staff that they hire in house would still be able to judge the quality correctly. Because with AI, I think that the human element is very much that you have to judge what you ask of it and then you have to judge the output on that end. And that's a service where like you guys could still come in and play. But then if you start talking too specific, tell me about how kind of judgment plays into this at this point. And is there an angle there for your services and for trying to win business in that regard? 

[07:43] Razy Shah: 

Yeah. I mean, judgment still plays a big part. So if you look at all the horror stories, I'll give you a few examples. Recently in Singapore, there were two lawyers who got fined $5,000 each because they had submitted some documents or they submitted some cases that were non-existent. Right. And then the court found that they're wasting the court's time because they had to go and find these non-existent cases that were purely AI-generated and they got fined. Right. And this is where human who would have cross-checked the work and the references, that judgment wasn't necessary. The other thing that was recent, you might have seen the news, Deloitte had to refund, I think the Australian government $440,000 for a report that they did because it was full of AI hallucinations. Did you see that one? 

[08:27] Karl Abbott: 

No, I did not see that one, but half a billion dollars is a significant amount of money. 

[08:31] Razy Shah: 

And then in just May of this year, just last month, Ernst & Young, EY, had to retract a paper that they had published because 16 of the references that were cited, 16 of the 25 references that were cited were hallucinated. So again, nobody is checking. And then of course, you have the horror stories of visuals that had an extra building, an extra tower. Which is why if you look at a lot of the clients we work with, now they are implementing this thing called a human in the loop. Because at the end of the day, if you use AI and something goes wrong, there needs to be accountability. So that human needs to be accountable before this thing goes into production, before it goes out into the wider public. And I think that's where the judgment bit is extremely important. 

[09:18] Karl Abbott: 

Yeah, absolutely. And that's pretty cool to hear that there is now requests and requirements for human in the loop in some of these places. Because as you pointed out, you can get absolutely terrible output with AI. And if you then rubber stamp that and say, this is what we're doing, this is good, and put your name on it, well, you get fined. Or you have to refund things. 

[09:41] Razy Shah: 

And you think that these companies that are so well resourced, you know, they're so well funded, they have such a huge team, they would be a lot more careful. So if these big guys, they are stumbling, or they are tripping over when it comes to AI hallucinating or making poor quality outputs, just imagine the rest of us, the rest of smaller businesses, startups, individuals who are nowhere near as well funded, who don't have the resource that they have. Imagine the kind of mistakes they could potentially be also making. 

[10:11] Karl Abbott: 

Yeah, it's really crazy. And it's all for just basically passing on the quality of the judgment, you know, that judgment happens in moments. It's not something that you just do at large. And at risk, it really is in that moment where you're like, okay, I'm about to actually mark this as final, and then send it off. That's where that judgment happens. And I think that that is really one of the critical human skills that we have in an age of AI. And you and I were talking before, and you had mentioned meeting early on with a restaurateur that struck with you. Take me back to that meeting. 

[10:49] Razy Shah: 

So this is the early days of running the marketing. It's like four years into the marketing agency business, four years into starting this business. So this restaurant owner, he reached out to us and he was saying, hey, look, I've just started this hot pot restaurant. I want to do social media marketing, specifically Facebook marketing. But then he said, I have no budget. I have no budget, but can you help me? You know, I'm a new business, help me make something out of this new venture that I've gone into. Usually when you hear that someone has no budget, you disqualify them, right? So if you look at the whole thing about B, A, and T, in sales, they always talk about whether someone has budget, the B, authority, the need, and whether or not there's some urgency, the timeliness around it. He had none of that, but he did have the A, which is the authority. He was a decision maker. 

[11:42] Karl Abbott: 

Decision maker with no budget. 

[11:45] Razy Shah: 

So the thing is, at that point of time, I'd been four years into entrepreneurship, into running a business. And I always wish that in my early days, if there was someone who had helped me, who had guided me, who had given me an opportunity, I could have done, I could have come along a lot faster. I could have done a lot better in that four years. So in a way, maybe I saw a little bit of myself in him. So I did agree to meeting him, sitting down with him and teaching him all that I know about Facebook marketing so that he could be then empowered to take on his own content, take on his own Facebook marketing and make his restaurant a success. 

[12:21] Karl Abbott: 

So what were some of the boundaries that you expected to have in place to keep around your time and energy there? 

[12:29] Razy Shah: 

I thought maybe I'll just go in there. But boundaries would be around time. Like, you know, go in there, maybe be done in an hour and then move on so that I don't have to invest too much time. Just an hour, share with what I do it like a quick favor and then move on. So the main boundary I was trying to look at was around the, how much of my time this was going to take in. 

[12:51] Karl Abbott: 

Yeah. And so as you kept talking to him, you kept using more of your time. What's kind of that back and forth conversation you're having in your mind as you see the clock kind of ticking past and you're probably getting excited by it. So that's what kind of drove you, I would assume. 

[13:08] Razy Shah: 

That's true. That's true. Yeah. And the thing is, once you go into these conversations, I mean, I have a lot to share about Facebook marketing back then. And there's a lot of examples I could show him, a lot of things I could show him. I also wanted him to take some pictures of his food and, you know, put it up, create some posts live. So I just got carried away. But in the back of my mind, I was thinking of the other proposal that I had to write that was going to pay me five figures. I was thinking of probably I could have gone for a networking event where I could meet 20, 30 other people who could potentially give me business. In the back of my mind, it's like, if all these inner conversations that are happening, you know, is it like, oh, this is not the best use of my time. What am I going to tell my business partner that I spent three hours with the guy who's not going to be on the retainer? So that's what was also going on in the back of my head. 

[13:55] Karl Abbott: 

So, yeah. When you finally left that meeting, what was going on in your head at that point? Were you like relieved to be done with the meeting or were you energized by the meeting? What did that feel like at the end of it? 

[14:11] Razy Shah: 

I would say I felt energized, actually. So when I was done the meeting, I felt good. You know, I feel it always feels nice when you give value to someone, especially someone who's new. You share some of your experiences, your expertise. And then there is appreciation, right? I mean, there's a lot of gratitude that the restaurant owner had shown me. He was very grateful. He was very happy. And of course, I felt good to have given the value, to have received the appreciation. And yeah, I worked out feeling lighter. 

[14:44] Karl Abbott: 

Yeah. And you told me earlier that for a long time after the conversation just kind of went quiet. Nothing really came out of it. But then if I remember correctly, you said that you got a contract from him later on. Tell me a little bit about that. 

[15:02] Razy Shah: 

I lost touch with that business. You know, like we get, I speak to so many business owners. So I got, I lost touch with him. And then years down the road, suddenly he messages me and says, hey, Razzie, you know, we, I've started a new job and we're looking for a marketing agency. So in my mind, I was saying, oh, this restaurant owner has now moved on to a job. You know, I was wondering what kind of a job is he doing? What kind of a budget? Because I'm reflecting back to my experience with him. He had no budget. I was like, oh no, this is like more work, but for a small budget. But turns out that he was, they were a real estate company that was looking to launch in Singapore and very well funded. And they were looking for a marketing agency that would help them with PR. They could help them with their Google ads, SEO, basically get them leads and get them visibility, awareness. And we landed a six-figure contract, right? So it was a six, six figures. I would say it was unbelievable that an interaction years back would eventually lead to, to a six-figure contract many years down the road. And this is a six-figure contract. It was spent three hours. Exactly. Exactly. When we are doing things like right now at the present, we don't think, we rarely think about what this could potentially lead to years down the road, you know, especially if you're in sales, you have a target to chase. It's very short term because if you don't meet your target, whatever seeds you plant now, you might not have time to harvest them because you might already be out of the job. So yeah, I think, I think a lot of times we are all very fixated on the short term outcome. We don't think of the longer term potential benefits. I mean, I think that's how our society, our modern society, our fast-paced city life is like, you know, you're just very short term, very outcome driven, not so much long term thinking and relationship driven and looking at like giving to people. 

[17:03] Karl Abbott: 

So did that experience change how you see yourself when you're making decisions? 

[17:08] Razy Shah: 

A hundred percent, right? It made me want to do unscalable. I mean, saying yes to things that are unscalable, saying yes to things that don't have an immediate ROI. So because one thing I've learned is in the early days, I mean, I didn't know whether or not my business will survive beyond say three to five years, right? So we've been doing this now for 14 years. So since I have a longer, a longer time horizon to look at things now. I do a lot more unscalable things. So I would say yes, if someone needs, if let's say an MBA student needs to kick my brain for half an hour, I give it, I let them have it because who knows where this person might be years down the road. If someone is looking for, if like a student is looking for some help, I am willing to help. If someone needs me to conduct a free workshop for startups, I'm willing to do that. So I do a lot of these things that are not directly attributable to revenue and things where you don't have an immediate obvious outcome. 

[18:09] Karl Abbott: 

Now in this age of AI, when so much execution can be automated, what kind of moments like the one with the restaurateur are harder to ignore? 

[18:21] Razy Shah: 

I think the moments that involve human connection, I was reading, this is really interesting study that I was reading to prepare for a talk that I gave earlier this week. KPMG and the University of Melbourne, they interviewed 47,000 people across 48 countries to ask them about trust when it comes to AI generated content, trust when it comes to AI marketing. And they also asked them, what are they afraid of losing when, as AI becomes more and more, as AI takes over more and more things, right? And across all of these countries, the top thing that people cited was human connection, right? So I would say it's the human connection. So these kind of moments, the human ones where you help somebody, where you provide value, I think that's the human part is the only thing that we can do in the age of AI. We have to go out there more, meet people, connect, provide value, and that's how I guess we will have the edge in the age of AI. 

[19:25] Karl Abbott: 

Yeah. No, I've seen a lot of that too, where it really is coming down to, it's these human interactions that are valuable. That's where a lot of the real work gets done. Yes, we can automate away the execution. These days at a speed we've not been able to do before, but absolutely, the richest interactions are still the human ones. 

[19:47] Razy Shah: 

I mean, in the future, do you think Carl, we might have this where Carl, the podcast host, is the AI avatar of Carl, and he's interviewing the AI avatar of Razzy? 

[19:58] Karl Abbott: 

I've thought about it, and you can actually do that today. And I suspect with a little bit of some Claude, we could have probably pulled this off in that manner, and it wouldn't have been super hard to generate a script between the two of us based on our profiles that are out there on the internet. You know, go read, here's Carl, here's Razzy, here's some things Razzy wrote, here's some things Carl wrote. Here's the conversation topic we want to have. Generate the script. Okay, here's 10 seconds of Carl talking. Here's 10 seconds of Razzy talking. Feed that into a different type of agent that's now going to read the script in their voices, and voila. But it would be a little jilted. Yes. It wouldn't be right. 

[20:43] Razy Shah: 

It's definitely possible, right? And I've seen a lot of podcasts that I just, I can, from just the voice, I know it's an AI. What was interesting is, I was looking at this study also from, there's this institute in Germany called the Nuremberg Institute for Market Decisions. And they did this study where they showed people AI-generated ads. They didn't tell them that it's an AI-generated ad. So they had two groups of people. So one, they showed the same ad. One, they said it's AI-generated. One, they didn't tell them. The group that wasn't aware that it's AI, they actually trusted it. They clicked on it. They were willing to interact. But the moment we told them that it's AI-created, their desire to engage with it went down, and their trust with it went down. So now, if we do this podcast, which is just AI Carl speaking to AI Razie, if we didn't tell people that it's AI, they would be, they would trust it. But if, let's say, you put in a label and say this is AI-generated, I think people would be like, let me just skip it. They don't even listen to any of our conversation. 

[21:40] Karl Abbott: 

Yeah, exactly. And the interesting thing about that is that the conversation could be just as good. 

[21:45] Razy Shah: 

Yeah, true. It would be better. 

[21:47] Karl Abbott: 

And not saying that, like, the AI is better than us, but just that it could take our writing styles and our speaking styles and the content. And it could very well be that in an AI version of this, Razie and I sat down and reviewed exactly what it was going to say and said, yeah, that's as good as I could have said it. Maybe we even made some edits to that AI document before we fed it into the AI voices to spit it out. But the moment, like you said, that you tell people it's AI, it's not trusted, even though the words themselves may actually be just as good as if we sat down and pinned them, which is an interesting position to be in and to hold. I know that one makes me just a smidge uncomfortable because it's like you can write very well with AI. And I'm the type of person that I like to, I like to tell people I yell at my LLM quite a lot and tell it you're wrong. Change this, change this. And let's change it up about 20, 30, 40 times until it's finally, yeah, that sounds like something I would actually do. 

[22:47] Razy Shah: 

You know, early last year I was using a co-pilot, Microsoft's co-pilot, and actually I scolded it a little. And the co-pilot responded by saying, hey, I don't like the tone of your tone. This conversation is over. And it just, the conversation ended. I couldn't ask you to do anything. I had to start a new chat. That's like my first time getting like told off by an AI. 

[23:11] Karl Abbott: 

I've had that happen too. It might very well have been our co-pilot product that did that to me. But yeah, these days I generally use GitHub co-pilot with Claude Opus and it's usually a little bit better than shutting you down unless you've been extremely egregious with it. But yeah, it is interesting that the AIs are like coded up and system prompted in such a way as to try and keep you from being too aggressive, I guess is the best way to put that. And then I mean like not to, I don't normally like to bring in two current of events into the podcast so that they stay for a while, but Fable just dropped. And the number of people that are out there like, I can't get Fable to do this work or do that work because the guardrails are so strict that as soon as they ask it a question in that sphere, it drops them to Opus 4.8. 

[24:01] Razy Shah: 

Wow. I have yet to try it. I just saw it was available as an option now for the past two days, but I don't have any task that is feeble level. 

[24:11] Karl Abbott: 

Yeah, I know a lot of what I do Opus is plenty fine for and it's like I don't need anything more and looking at how much more this cost is like, yeah, I think I'm pretty good right about here. Yeah. Who knows? Maybe at some point there will be a model that is more expensive that makes me go, that's the one I want to actually spend the tokens on. So output has, as we've talked about, become extremely easy to generate now. I think that from my perspective, one of the key tricks to actually getting good content out of it is the context. And I like to have a conversation with my AI before we actually end up with the final thing. It's not just a singular prompt. It's like maybe hours or maybe even days of working on something and, you know, still taking kind of that craft approach to it of that doesn't look quite right. That doesn't look quite right. Let's shift this. Let's shift that. Let's change this. I don't really quite like this. The story is not quite landing for me. You know, I like to do all of that. But so there's a lot of me working with AI through a lot of my thoughts and context to get to that point. But when you're working with AI, what are the types of things that you're personally paying attention to that the AI still can't show you or that may be blind spots for that AI? 

[25:32] Razy Shah: 

I pay attention to the way the output it comes up with in terms of the speaking style. I mean, I still pay attention to like, you know, when he uses AI tropes, like, you know, it's not about X, it's about Y. AI. And those are the kinds of things I pay attention to because I feel like while you use AI, we shouldn't be using it in a lazy manner, which means when we take the output, I should be able to still, I want to still edit it. So it sounds like me. So even though I give it some of my writing, I give it my, the way I speak, it still doesn't get entirely the way I sound. So I pay attention to the language. I also pay attention to the accuracy. I think a lot of times it doesn't, it misses out certain things. Like for example, these days on Instagram, I've been using AI to create carousels, marketing carousels to share certain concepts. And I feel sometimes the visual is wrong. Sometimes the story is incorrect. So I have to go and double check. Even though I'm using clock, I use clock 4.8 maximum for the story. It might sometimes be, especially when it comes to stories that are from Southeast Asia, when it comes specific to Singapore, there's not, maybe there's not enough content available online. It's not able to do as good of a job. The other thing that AI obviously can't do is, I mean, the part we spoke about, the human connection. So when I put this piece of content out there, I don't know how it's making people feel. Whereas if I go into a room to a workshop, I deliver a 30 minute talk or I deliver a 45 minute workshop. At the end of the day, the feedback that I get from humans, I know immediately how things are, things are landed for them. Right. Because these days, even if let's say I put a piece of content on LinkedIn, all the comments you get, sometimes you see while they come from people that you are connected to the AI, you know, the AI comments. So even though you might say great things, I mean, I don't know how I landed to, to the actual human who might've consumed it. 

[27:20] Karl Abbott: 

It's an interesting thing because like the easily generated content where you haven't given it your context, you haven't like worked through to get to that point. And honestly, if you're sitting there listening to this podcast and you've never actually like fed all that into an AI and sat there and really hammered away at an answer, you really ought to try that because it's a lot better than just like, Hey, write a quick LinkedIn post on blah, blah, blah, blah, blah with absolutely zero context because it's just going to generate a very generic AI sounding thing. And when people do this on LinkedIn and they do it all the time, you see it, you open up your LinkedIn feed and you can just like point, yep, that was a generic LinkedIn one prompt answer, no context, hardly provided. And it's just like that stuff I do pass up. The other things, if there was context provided, I will read through that. But you just, you learn the more you use it, where those tells are. And what's fascinating about the tells as well, or, you know, there's the M dash. I love the M dash. I used the M dash before AI and now it's like taboo. You can't use the M dash because it was definitely AI. Yeah. It's like, but, but I love the M dash and the AI learned that we love the M dash by reading human writing. And that's another thing to remember about AI is that it's not just generating this stuff completely computerized. I mean, it is, but it's based on all this human writing for its training set. So it doesn't know what it's doing. It literally is just trying to predict what the next possible word is, which is pretty amazing. When you think about the matrix math that's going into this and that we're taking language into math and then putting it back into language and that it works as well as it does. That's a little bit mind blowing. That's, we can admit that that's a really crazy thing that it actually works, but that's really all it's doing. So it doesn't know what the M dash is. It just knows that it found it in the English language. And it works really well to connect two thoughts that we don't want to end a sentence and start one. 

[29:16] Razy Shah: 

One thing I do feel is because the people, the circles that we operate in, they use AI a lot. Like my friends, like colleagues, other fellow business owners are all using a lot of AI. So we can see when you read an AI sentence or paragraph, you know that is AI, but the wider public, they don't use AI as often as we do. So like when I teach people that, Hey, look, this sentence structure is not about resilience. It's about hard work. It's not about X, it's about Y. Once I show it to them, then they start seeing it everywhere. And they're like, Oh, Razzie, why didn't you? Why didn't you do that? You know, I see it in ad copy. I see it in print ads. I see it in YouTube videos. And they're like, why do you open our eyes to this one signal of AI? And I feel like, yeah, but we use it so often. We know what's AI, but I think the wider public, they don't use it as often. And that's where they can't discern or they can't distinguish what is AI and what's not. And which is why it's being used for all the wrong purposes. 

[30:17] Karl Abbott: 

Yeah. And there's at the risk of putting everybody into little buckets, there's at least three buckets right now. And that's like your advanced AI users who are using the stuff day in and day out to get actual work done in some form or capacity, which you and I both definitely fit into that category. You have people that have dabbled with AI and have talked to a chat bot, maybe use it for search every now and then, but don't really use it that often. You know, you'll reach for AI, but it's not something you're going to necessarily reach for every day. And then 

you've got basically a group of people that are so AI averse that they're like, get that away from my system. I don't want to see that at all. And I said three buckets, but maybe there's a fourth. People that just don't really know much about AI and don't use it either, but don't really have an adversity or a positive feel to it. The data center stuff has definitely spun a lot of people up against it. 

[31:11] Razy Shah: 

Oh, yes, yes. In fact, I was at the same conference I was telling you about earlier. Someone on the panel was talking about how there's this area in Malaysia where they're building these data centers and they've written into the law or something where if there's a need to prioritize water, they're actually prioritized to the data center instead of the people who need to drink it. Wow. Which isn't how they work. 

[31:34] Karl Abbott: 

That is, yeah, no, we should not be taking water from people that need it to fuel data centers. I hope we could all agree on that and hopefully Malaysia will fix that. So in a world where AI makes it easy to do more, which is honestly one of the biggest benefits I've seen, not that, I guess not so much that it sped me up, though in a way that is speeding you up, but that it's making space. It's taking on some of the more menial tasks that I didn't really like doing anyway and giving me more time for the work I want to do more. What's the hardest thing to keep choosing? 

[32:18] Razy Shah: 

Well, there's a number of things. I'll tell you one is restraint. So like what you're saying, when we can use AI to do infinite output even at very little cost and the hardest thing to do is to do less, to be slower, to be more careful for whoever is going to be consuming whatever you're producing. I was looking at this study by a company called Capwing. They were looking at AI slop channels. So you know these videos that you see on YouTube and on Instagram. And I was wondering to myself, why are people bothering to create this AI slop? It's clearly AI slop. And when I look at this study, the top AI slop channel on YouTube made $4.25 million. Wow. The tendency there is like, you know, if the slop is so profitable, create mindless content at like 10, 20 posts a day is so profitable. That's going to drive even more people to do this kind of work, right? So I would say restraint. So for me, the hardest thing to choose is restraint. Making sure that even though the AI has written my LinkedIn post, making sure that I put in the time, the care to look through it, to improve it further and ensure that whoever is going to be looking or reading that piece of content is taking away value and value that comes from lived experience. So one thing that I'm doing, so I'd say restraint is the hardest thing to keep choosing. And for myself, I make sure that I try to elevate the content with lived experience, things that I've observed, things that I've experienced myself. And I think the AI doesn't have lived experience. And that's where I can create something that is a lot more valuable. 

[33:52] Karl Abbott: 

Yeah. And that's editing. That's the editing process. You know, when it comes to putting content out there, whether it's on the internet, whether it's in print, in a book, whether, you know, whatever content you're writing for, you never write the first draft and hit publish on that. Yeah. You are always at a state of editing. So if you consider what the AI put out the first time as your first draft, which is a little different than how we've always done it, where you have to sit down with a blank piece of paper, but let's just for the sake of argument, say that's your first draft. You're still editing that down before it's what you want to actually put out there. And in that sense, that is not really any different than what writers and thinkers have been doing for centuries. Take the thought, shape it and shape it and shape it, and then go ask people for their thoughts on it and continue to shape it until finally you're ready to share that thing out. 

[34:54] Razy Shah: 

But what I'm seeing now is with all these agents. So when I speak to some of my friends who are way more advanced when it comes to the AI usage, they will show me, oh, look at this. I have a Hermes agent on Telegram. I talked to her and she's able to go write LinkedIn copy, create the visual to go with the copy and she posts it. So there's no, they're not even involved in the editing anymore. So they just have given it the context. They just let the AI, the agents run with creating the visual, creating the post and posting it after them. So there's like no human editing involved anymore. So I think that's where like the hardest thing to choose is still being involved, I would say, right? To be me. Like you can let the agents do everything. Like, okay, look, let my agent do everything. Let my agent do that. 

[35:39] Karl Abbott: 

Well, and I mean, I think you see it with vibe coding, right? Like people will let their agents just vibe code something for about six months. And anybody who's ever done vibe coding knows that for one thing, you start running out of context windows and you got to deal with your agent needing to understand the code yet again and again and again. But if you let a project go completely agent driven for six months and then something goes wrong with it and you've got to do something about it. Well, the only nobody really knows what the project's about at that point. And it's you and the agent trying to debug something that the agent forgot how it wrote, what it wrote, and you never understood what it wrote. I mean, if you take the human out and you take your understanding out, at some point you just end up in a spot where you've got product, whether it's content or actual code running on a server somewhere that you don't understand. 

[36:30] Razy Shah: 

Yeah. 

[36:30] Karl Abbott: 

And that's kind of scary. 

[36:32] Razy Shah: 

Yeah. Then you go back six months to six months or you go look to what it's done over six months. It's going to take a lot of time. 

[36:39] Karl Abbott: 

So it's like, do they even go read their LinkedIn feed to see what actually got posted? I mean, do they know it's not posting the wrong thing? 

[36:46] Razy Shah: 

I think the only thing we read is when someone flags it to them and say, hey, look, this sounds weird. 

[36:52] Karl Abbott: 

And then LinkedIn's like, your content got flagged as inappropriate because like, say, the agent just decided to go off the rails one day. 

[37:02] Razy Shah: 

Yeah. So yeah, maybe beyond the restrained part that I answered earlier, the hardest part to choose later on is like choosing to still have friction, you know, not fully go towards convenience where it's smooth and it's completely frictionless. You're not involved. I think like choosing a bit of friction is going to be harder, harder to keep choosing. 

[37:21] Karl Abbott: 

It is. But that's the moments where we learn. That's where we there's a lot of good to come from friction. I mean, I enjoy frictionless workflows as well, but you got to have your moments of friction. Exactly. And I know writing LinkedIn content is not everybody's favorite thing in the world to do, but I think you do have to have some things out there that you're like, no, I'm not going to let the AI touch this. I'm going to do this completely 100% human. Yeah. For me, that's my photography mostly. When it comes down to like editing and color profiling and some of the more technical aspects of digital photography, I don't have a problem letting AI like do some analysis or, you know, suggest some different changes here or there, but I'm not taking it out there in the field going, okay, we're here at this location at this particular time. What's the angle I need to do or anything like that? It's still a very human process when it's out in the field. 

[38:14] Razy Shah: 

I think that's really good advice. So you still need to do something that is fully, fully human. 

---