Breakeven Brothers

Cool, alright, welcome everybody to Episode 21 of the Breakeven Brothers podcast. How's it going, Brad?

It's going great. From the last podcast episode, I thought I would just start off today with a banger, starting the auction at $200. So if you're on the YouTube stream, take a look at this beautiful Breakeven Brothers cup that I created. It even has the date and my name on it. So yeah, we'll start it at $200, and let's chat offline in the YouTube comments. We'll start the bidding war. My guess is $500,000. It's one of a kind, folks, take a look. So yeah, I mean, the red is just stunning. That's beautiful. Yeah, yeah, it was created not too long ago, so it's very fresh and unused. Nice.

But on a more serious note, I've been doing good. I've been playing some Oblivion. So we've been chatting about this. But recently, one of our favorite childhood games, Elder Scrolls: Oblivion, had a remaster come out. And I've played a lot of it. So that's been really fun. Nice. Yeah, I'm going to look right now. I think I've played a little bit of it. I want to play more. I think I've played three hours in total. That's nothing. That's nothing. I think I'm level 10. And yeah, it's a great game. I even listen to the soundtrack when I work because it's just, it's good vibes. Yeah. Yeah. Very cool.

Yeah. On my front, it's been pretty busy with just regular work, but then also just trying to do a bunch of cool new stuff. So I've been contributing a bunch to my private GitHub repo. Just trying a couple of different things. I can't remember if we talked about it last time on the podcast, but I was able to get like a demo MCP server. Did I talk about this on the podcast, or did I talk about it privately?

You did. Okay. No, you did. I won't go too much into it, but just refined that some more. And what I've been really working on a little bit this week was trying to really perfect the prompting of a user's query into an agent action. Nice. Not the main topic that we're going to talk about today, but it's been kind of pulling my hair out a little bit, to be honest. Because, you know, trying to use the LLM where it's good and then trying to use code where it's not. But sometimes those lines aren't super black and white, and so it's like, which is going to be better? Is it better to try and like code all the different scenarios that someone could say and try and get to that outcome, or is it better to let the LLM interpret it? Kind of getting mixed results with both. So yeah, just the joys of tinkering, you know. Nice.

So yeah, I was at Deep Dish Swift last weekend, and it was in Chicago. First time visiting for me. I think it's their third or fourth year. So it's mostly an iOS, Swift-focused conference. And they had an indie day. They had two regularly scheduled days, just normal conference talks. And one thing I noticed the whole time is people kind of hinted at AI. In each and every talk, maybe not every talk, but in most talks, it was, "Hey, this is like a new language feature in iOS," or "This is a new Apple feature." And either AI can or can't do this. And I think the trend from what I've seen at that conference is people used to be kind of hesitant about using AI, especially with Apple products, due to documentation not being super accessible. The training data that these LLMs are trained on isn't really that Apple-focused. But I think people are really excited about using the tools and creating kind of Apple-specific workflows to make using AI IDEs or other AI products and tools a lot easier and build some really cool Apple-specific stuff. So I was excited because I feel like I've obviously been a proponent of pushing for AI adoption. And I think at the conference, it's a perfect time to kind of have that take on a larger scale. Or at this conference, you know, there are lots of people from various companies representing, you know, Uber. What other? Oh, you know, Zillow. Saw some Zillow folks. You know, various companies are on board, and hearing people on stage talk about AI brings it front and center. And they're bringing it with the approach of, "Hey, you should use it now because it's good, and it's only going to get better." And we've been beating that drum since day one. And so it was really, really cool to go out there, learn some new iOS stuff, meet a ton of cool people I've met on Twitter. And then get this feeling that AI is here and people are getting exposed to it in a better scenario because I feel like lots of the times people want to use it but just need like a push. And I think at the conference, people were saying, "Hey, you should try it," like repeatedly. And every talk was, "I'm getting stuff done both as an indie developer and like working at big tech by using AI tools. This is where it's great, and this is how you should use it." So yeah, great time at Deep Dish, but I think the AI theme was extremely prevalent, and I imagine across all these coding conferences coming soon, it's going to be kind of the underlying theme of, "Okay, they showed me how to do this, but can AI do it, or how well can AI do it?" And that's kind of, you know, an interesting take now with AI being so front and center.

Yeah, yeah, for sure. Well, and one thing, it's not the main thing that we're going to talk about in this podcast, but this is actually news to you because it's, it's pretty fresh, but um, I started a newsletter that's for accountants to learn how to apply like technology, not just AI, but like technology to their workflows to help automate or to help kind of figure out when to use LLMs versus not. So it's called Augmentic Accounting.

That's a great name. Hats off!

Yeah, right, because it's going to augment, you know, your accounting abilities, right? So, but basically the idea, and I think what's great about like things like Deep Dish, and I think engineering does this really well, and I think accounting has some room to catch up, hence the newsletter, is I feel like, you know, there's going to be such a big sea change with these tools that's coming. Like, it's so fast. Like you and I have talked so much in this podcast about how quickly things are shipping from OpenAI, from Google, like just everybody is just constantly putting more and more stuff out there. And then even outside of like the main foundational models, you know, Stripe is doing a lot of really cool things. Zapier, or Zapier, whatever the pronunciation is, they're doing lots of cool things with like MCP. And so there's just so much that's going on that it can be pretty overwhelming. But like accountants and people that are in the accounting industry, especially those that work at like enterprises, I think smaller, like when you have your own firm, I think there's a bit more, you're just used to learning and used to kind of getting into new tools and trying things out. But I think for folks that are working in an enterprise, working at a company, you kind of have to go out and seek out those opportunities to learn and to, you know, get your hands dirty in a new tool or something. And so I think the idea with Augmentic Accounting was like, let's have a place that like accountants can like learn, you know, what these tools are, like how to use them, kind of walk through like automating workflows and try to share as much knowledge as we can. It's not, you know, obviously accountants do, we deal with a lot of sensitive data. So like we have to always be careful, but like the idea being that like we can share like how we're doing something. I can't give you the exact details, but I can share with you how I automated a prepaid asset amortization. I can kind of just walk you through the thought process. But engineering does a great job because you guys are always going to conferences, always kind of sharing like this is what's going on in the space. And so, yeah, there's just, if you're proactive about getting those skills and equipping yourself with the skills that can be relevant for the next year to three to four years, then you're going to be in such a league of your own compared to those that, you know, didn't do that proactive outreach or they kind of just, you know, stayed in their role and didn't, didn't bump into the walls at all. And that's where I think you're at more risk of like, you know, losing some value in the workplace if you just keep those skills and don't ever grow them, you know, to where the market is going, you know?

Yeah, I would say it's coming. If you can use it earlier rather than later, again, you'll be much more well-equipped to have an easier onboarding curve than almost getting it when everyone else has already gotten it and you having to play catch up a little bit once it becomes a lot more mainstream. What I've seen is that I think initially AI tools, workflows, IDEs were very much kind of brought in from the employees' perspective and showcased to, you know, leadership. Now it seems like all these companies are adopting it kind of top-down versus bottom-up. And with that approach, people are expecting, at least leaders are expecting productivity gains across the board. But if it's the first time you're using a tool like ChatGPT or Gemini for Google Workspace, you're going to feel very much like, "How do I even operate this? How do I become successful with it?" And so you can ease that burden by trying it now. And again, yes, it's hard to find the time, but hopefully there are like AI programs at your company, or if not, after work, just trying to chat and interact with these things, I think is probably your best bet to get some experience. For the newsletter, I wanted to ask, what platform are you using to kind of operate that now? There's a whole category of products in that space.

Yeah, I'll tell you, I tried Substack, but I think I had, this is embarrassing for being like a tech podcast to have this kind of issue, but I had some, I had two Substack accounts and they were like conflicting with each other. It's a long story, I won't get into it, but I just said, "F it," and I just went to ConvertKit, or I think it's just called Kit.com now. So, landing page, you know, that has an email intake, does all the back-end of like, "Confirm your email," and then sends you to a thank-you page. So, yeah, pretty, pretty straightforward. You know, depending on when this gets released, there will have already been an announcement on my part about it. It hasn't been announced yet, but yeah, super excited about it. And I think, you know, there's so much, going back to what you're saying about like getting in front of these tools now and like just getting through that earlier is always better. I remember there's just so much frustration as like a non-technical person, like just hooking up, you know, the chat or OpenAI, like API, like just getting the API key and like all that stuff that like, you know, honestly, engineers kind of just are used to. It can be super annoying, frankly, for non-technical people to get like error messages when they're just trying to generate a little poem, you know, and stuff like that. So my thought was that like I could help just kind of facilitate as people want to learn. Like I've been exactly in those shoes. Like I don't have a computer science background. I don't have an AI background, but I do have some knowledge that probably is relevant to them trying to kind of get going in their journey. I mean, I've been in their shoes if they're brand new to learning. This part of when I was building the newsletter, I was looking at some of my old Udemy courses. And, you know, it was kind of a walk down memory lane of like, "You bought this in 2017 and 2018," you know. And so, you know, hoping to kind of like help people that really want to get on-ramped to like understanding how to use these and understanding programming a bit more, like not just AI, but also talking about like, you know, how do you interact with APIs? Like what's JSON? Like people don't, you know, people think it's a name, you know, so like explaining what JSON is and like how that all works. So not just AI, but specifically like, because I think the future of accounting is going to be people who can build. And I think there are two prongs. I think it's people who can build out like processes or automations to be able to handle the volume of work that's going to get thrown towards you and be efficient. And then I think the second one is build relationships, which obviously my newsletter is not about, you know, it's not a psychology newsletter. But like if you can be good at those two things, if you can build out, you know, whatever you need to build out to kind of get the data where you need to or get the journal entries where you need to, and you can, you know, be someone that people want to work with and be helpful to the organization or to the team that you're on, you're going to be okay. But I think if you're not doing those things, that's where it's going to be like, you know, yeah.

I've seen it time and time again, people mention, as we get more AI just intertwined in our lives from work to personal life, being a human is going to be much more valuable. When I think about that, it sounds kind of dumb on the surface, but I think it will really come around. And part of that is talking with people, building relationships. When I was in Chicago at the Deep Dish Swift conference, there were people I talked to on Twitter or on X, and I was trying to find them at the conference, and people, they've talked to me, but I don't necessarily remember them. So there's a ton of personal kind of connection building, networking skills that really take a lot of effort and, you know, skill to pull off. You know, there are people that are popular at the conference that aren't popular, like just meeting people, interacting with them, asking them the right questions where, you know, being a conference attendee, I don't know, it takes a lot of effort to mingle with people, especially like entering large groups of people. And so I was really happy with how much socializing I did. And I think always starting out in these conferences, I always have this like fear of leaving the conference without talking to enough people. So coming in, I'm like, I want to talk to five or 10 people that I had met online, or at least I know the face or name, and happy to report back that I think this conference was a giant success. Tons of people I talked to over the past year, especially when I went indie on VoxByte, people I had talked to, I kind of met in person, said hello, had great conversations. So really excited. I think yeah, in the age of AI, being a human is definitely important. Understanding and building relationships with coworkers, peers, companies, you know, everybody within your organization, extremely important, I think, to get things done. Yes, we'll have more AI, but we're not going to remove people completely.

And on that note, Duolingo recently put out kind of an AI manifesto. So it's not the first time we've seen this. Shopify released one, I don't know, maybe a few months ago, their founder, Tobi, mentioned AI is here and it's coming. Shopify did, or sorry, Duolingo did more or less the same thing. And I have their notes kind of pulled up on my other monitor. And a few, I don't know, maybe exciting headlines or scary headlines, depending on how you look at it. It's mentioning they'll gradually stop using contractors to do work that AI can handle. AI will be used in their hiring process, and it'll be used in their performance reviews. And headcount will be given to a team if they cannot automate more of their work. Yeah, I mean, pretty interesting. On that, I mean, if you think about performance reviews and people and AI, you think, "Oh, that sounds complicated." The other bits of AI and hiring, also a very complicated space. AI for contracting work, well, not that complicated. To me, that sounds pretty straightforward. But either way, I think when people read these, there's a very visceral reaction of "I hate it" or "I love it." And yes, that might attract more people, and yes, it might have current folks running for the hills. So what do you think about it?

Yeah, I mean, that's a huge part of that post and then the one from Shopify were kind of huge parts of why I wanted to do the Augmentic Accounting newsletter because I think the right people should look at that as an opportunity to like, "Okay, this is coming. If I get these skills, then I'll be ready to kind of take this challenge on," and the challenge of companies expecting you to produce more value with these resources. I mean, they really do. Duolingo's proof of it, and Shopify's proof of it. And like, you know, just based on my own experience, like companies are constantly plugging away at like, "How do we, how do we get this going? Like, how can we get some benefit out of this?" And they're going to find it. Like, they will find use cases, and that'll become more and more of the expectation. So if you're in the camp where you go, "I hate this. This is scary. This is job security threatening," like I would say like pause and figure out how you can tap into that and get the skillset you need to still be valuable. It can be an uncomfortable conversation to have sometimes with yourself, but it's one that you should have certainly because that's just the way things go. That's, that's just how business is done. You need to make sure that you're always adding value. And so being proactive about your skillset, looking around the corner, you know, a year, three years down the road, making sure that your skills are valuable. Beyond, once it hits AGI, I can't tell you what will happen. No one can, right? But for now, it doesn't know how to do accounting at all. So we're good, but it needs to help us. It needs to help do what we do, make us better, make us more efficient.

I like the way you frame it. If you are scared, I think that is an opportunity to look into it. And I think the way you look into it is how can you use the tools to make yourself more productive and then drive more value for your company or whatever you're working on. And I think looking at the comments of that LinkedIn post that Duolingo had, the top comment: "Farewell to my 500-day streak on Duolingo. I'm going to go use other learning tools," by an unnamed LinkedIn member, and tons of reactions on that one. And then the one right below that: "Fantastic update!" So as you can tell, the reactions are very mixed. Um, half the people are saying, "Boo," half the people are saying, "This is great." At the end of the day, I think most of it makes sense from where I'm standing. However, there are a little bit of controversial takes in terms of hiring and performance reviews. Anyone can clearly see that going astray. I assume they're going to test things and figure it out, but quite interesting nonetheless. And on that note, how has your streak been going? I'm just going to put it out there. I am on day 15, I think, of Mandarin. I switched from Spanish. I was just getting bored of Spanish, to be honest. So I did Mandarin, and it's actually been pretty nice, pretty hard. It is hard. Yeah, it's been good.

Yeah. Yeah. I'm sad to report that my streak has, has passed. I think I stopped doing it maybe like a month ago, and I was like, "Oh, I need to get back to it." And at the Deep Dish conference, they had the Duo, kind of like the bird mascot, at the conference. It was really cool. But yeah, this is kind of my self-reminder to get back to it. But yeah, cool.

Well, before we get into our main topic, because we want to talk today about Notebook LM and kind of just share some ideas around that, share what it is and how it all works, there have been two, I guess, particular news items happening since we last caught up that are going to be good to kind of try out or talk about here. I think the first one is, and I don't know much about it other than the headline, I haven't really dug into it yet, but curious if you have more. If not, we can sail past it. But the sale of Windsurf to OpenAI for like $3 billion. I know that was talked about for a little bit, and I think it's official now, right?

I don't know if it's actually official. I think, yeah, I think it's still rumored. However, I have seen tweets between the Windsurf CEO and Sam Altman, CEO of OpenAI. I don't know where that lands. I know there was a giant kerfuffle of, "Should they buy Cursor? Should they buy Windsurf? Did they put in an offer for Cursor a long time ago? Why not Cursor over Windsurf?" This whole kind of trying to understand where OpenAI is thinking in terms of spending their valuable dollars. But I think it's been heavily rumored and will be announced soon. Other than that, I have a small update that I have used Windsurf recently. It's great. I find it very hard to compare to Cursor. I know people go back and forth between, "I picked up Windsurf and never looked back." I have not had that experience, and I'm not sure why that's the case. Either these people don't understand AI at large and they're prompting differently with these tools. I feel like I approach it where I prompt pretty much regularly between the two tools. And just to be clear, I'm not opening up a window in each and asking the exact same thing. What I'm doing is just working with them for a few days. And I've found the Windsurf UI looks maybe a little bit better and more polished, but the performance and the editing experience and the inline edits across both products feel roughly the same. So I have spent my $20 on a Windsurf subscription for the podcast to kind of report back. So I think I have maybe, I don't know, 15 more days. I have reactivated my Claude subscription for the chat app as well because I was doing that MCP work. So we're kind of at a problem where something's got to give. I have, you know, ChatGPT, Claude, Cursor, Windsurf. I think that's all I have now. I almost signed up for Notebook LM when I ran out of credits there. So we, we need to slow down just a little bit. But I think Windsurf is great. Hopefully, it gets acquired and that falls through. And then I can see OpenAI just kind of shielding maybe their models and putting extra sauce in Windsurf, which would be really cool.

Yeah, it's cool. Yeah, for some reason I thought it was final, but I guess not. But maybe soon to be. Yeah, I would say, I don't know why you haven't used Cursor as MCP and just ditched Claude Desktop completely. I've heard about it. People have mentioned it. Like, why do you? That's what I use. Cursor for taxes and other MCP things.

Yeah. I don't know. I just never, it never felt right to me. Like, are you opening up the Cursor UI and just have no text editor there? You're just using the full chat pane? Like, can you pull that chat pane out and have like a first-class?

I drag it over. So it's like larger. Okay. So like I use it. Yeah. Like I used it. I built like the one I talked about. I built like a demo, and I recorded a demo for people that I work with. And I just dragged the whole thing over. And I was like, "For those that don't know, this is Cursor. Like pretend this is like an agent," you know? But it's pretty simple. It took me a long time to get the like setup right. But now that I've got it set up, like it's easy to add them now. Like it's not anything challenging anymore. At first, it was like a whole mind, mind bend, you know.

Yeah, I probably should use Cursor for MCP integration work, but more or less I've been building it and then opening Claude. And then every time Claude launches, it tries to connect to my Personal Capital MCP server and that whole mess. So, yeah, I should, but I haven't. Yeah.

The other article I was going to mention or see if we want to talk about a little bit here is the whole Apple App Store trial and the changes for that. Do you want to talk about that on this one, or do you want to save that for another episode? We can keep the people waiting.

We can talk about it. I mean, I think we're in early stages right now. So the Epic v. Apple trial, long debated, I think many years trial, where Epic was suing Apple for anti-competitive App Store measures. And I haven't read the full trial. I've been kind of on the sidelines waiting to see what happens. And recently a judge struck down and said Apple is being anti-competitive. And due to that, they're going to get fined and they need to change their policies. And so what I've cared about the most is, being an app developer, indie developer, iOS developer, what can I do now that Apple's walled garden might change? And I think where we're standing today, the largest change is that Apple previously wouldn't allow you to go outside of the app to purchase things. Now you can do that. Again, you're not getting Apple's like frameworks and UI to have in-app purchases by your choice. So you're leaving your app, going to a website, checking out with Stripe or your credit card, then coming back to the app. That will avoid any of the Apple fees. And that can be 15% if you're a small business or 30% if you're a larger business. And so immediately after this news happened, all these in-app subscription companies, even larger companies like Shopify, have come out swinging at Apple and saying, "We just pushed a new update. Our pricing is much more transparent. You can go pay off the app. You can manage your subscription outside of the app," where previously Apple would reject you and tell you, "Don't do that," and slap you on the wrist. Now that's an option. With that comes a whole bunch of do's and don'ts and seeing where that finally ends up. So I think today it's really hard to say what officially came out of it because this happened maybe two weeks ago. And the largest thing that changed was the App Store guidelines maybe a day or two after it got updated. And when Apple updates their App Store guidelines, they email all developers. So I saw the news on Twitter and thought, "Oh, cool, like what's going to happen?" And then maybe two days later, I got this long update that they've updated their App Store guidelines, like terms of service document. I didn't take a look at it because I knew other people would. And over the past few days recently, everyone's been giving their own take on if you're a small business, do that. But we still might want to use Apple's in-app purchases because people are confident about it. So it's kind of this hodgepodge of information that there's opportunity, how that opportunity will end up in terms of saving money, getting more dollars into your hands as a developer, but being safe and smart within Apple's ecosystem, still yet to be determined, but we'll probably have an update, you know, in a month or so. Cool. Cool. Good stuff.

Alright. So switching gears to the main, the big kahuna that we want to chat about today, Notebook LM. So we've already kind of previewed it a little bit. I'll just kind of kick it off and kind of share how, like I, frankly, I think I've talked about it a little bit in the last episode, we kind of previewed this episode, that I stumbled on it on LinkedIn. Someone had used it to like cite or like do the podcast of like a court case to make it more interesting. And so I took that and saw that the company I work for has, you know, we're Google Workspace. So we have Notebook LM Plus. And so just did that with a financial kind of closed document that we had that kind of talks about the financial results of the period that we were closing and had a ton of fun with that because basically you just, it's really simple to add a source. You can do Google, Google Slides, Google Doc, I think PDFs, copied text, and website URLs and YouTube videos, I think are the main buckets there. Missing Google Sheets notably, but we'll get more on that later. But basically, you can give it the slides as like a source, and then it can generate a summary for you, an audio, which has been by far my favorite feature. But there are lots of other features that it can do. Like you can chat with the source. You can do like what's called a mind map. So have you been able to use it too and kind of played around with those features and found one that you like the most?

Yeah, I had a similar experience. I don't know where I found it, but I think Spotify had integrated with it for their year-end Wrapped. And so maybe I think it was like this, and they gave you a podcast using Notebook LM discussing your listening habits for that year. And that's when I first heard it and thought, "Oh, this is really cool." And I think after that, I was asking some questions about how to market, split my marketing expenses. And the workflow I had done was I went to ChatGPT, I used their deep research toggle, which would kind of print out 10,000 words on a topic after doing web research. Then I would take that content and plug it into Notebook LM because I didn't want to spend time reading 10,000 words. I don't even know how fast I read, but that just sounds like a lot. And when I scroll the page, I'm like, "Oh, I'm not into it." So I want to make that more digestible and fun. I think the word that I picked up on is fun. Like I had a ton of fun listening to the outputs of Notebook LM. So yeah, like Ben mentioned, you can pass in sources, whether that's files, text, etc. Once you have that step done, you can ask it to generate, you know, this audio podcast. I've never done the mind map or chat. I've literally only put in input as text, asked it to generate the podcast, and listened to that. And so I think the first one I did was split my expenses, marketing, from the output of a deep research query, and I thought, "Oh, this is awesome." I think it was 15 minutes back and forth between a male and female speaker, making things a lot more fun. Like I was driving up to San Francisco and I put it on as a podcast and I loved it. You know, maybe I would have never looked at the deep research response because sometimes you bookmark things for later, you do all this or that, but it just felt a lot more approachable and fun. And then I didn't use it for a while. Then I picked it up recently going out to Chicago for Deep Dish. I had made three queries to deep research because I knew it worked and then pre-prepped all those recordings on Notebook LM. And I downloaded the podcast that it generated for each of those three queries. And then on the plane, listened to each of those. I think it was between 12 minutes to like 20 minutes, roughly. And I loved it. It was a lot of fun. I was completely plugged in. It's content that I wanted an answer to. But I felt like it was not super approachable, like it's very technical jargon. And I thought the speaker approach of a podcast felt so fun and exciting. And maybe you retain as much, but it makes it easier to kind of start that conversation or exploration into that area. But that, that's kind of been my experience.

Yeah, no, for sure. I think one of the benefits to it is, I mean, I use it the exact same as you, where I just took that audio file and like downloaded it and then like walked my dog and listened to it. And it lets you digest information without having to like sit there and read it. Like I've tried to read things while walking my dog, and it's horrible because I'm just, she's a good walker, but you know, just, it's just, it's not the same. Like it's just easier to listen as you're walking. Yeah. I had a couple of different use cases beyond just that kind of financial close deck. So that was one, um, where, and it generated the audio, and it was again fun and engaging, and it has two speakers that kind of go back and forth. And it's kind of funny because sometimes they'll kind of throw in like little jokes or like, you know, like statements, like they kind of like roasted, you know, something that I had put in there once, and I was like, "Oh, interesting." So you don't really have a lot of control over like how it's set up. Like you can't say, "Give me one speaker," you know, you can't say, you know, "Make both the speakers female or both of them male." Like I kind of, you kind of have to live with what you get on that, but it still does a really good job. So I did the financial close. And then the other one I did was like some, some Python code. Like I had a script that like did a bunch of steps. It was kind of actually, we talked about LangGraph last episode, I think. So I put that into the, like as a source in one of my notebooks, and that's where I did the mind map thing. And it kind of told you like each function was like a node, and then you'd click on that, and it would kind of like expand and show you more details of that node. So that was really cool. And then I think the other things that it can do is it could generate like templated kind of FAQs based on the document that you give it. So I think I had given it like a, like some kind of like legal contract, I think, and it generated like an FAQ. And like, I think another one was a briefing doc. And so it just basically spits out like a formatted, you know, version of like an FAQ or a briefing doc that's like already pre-built, and it kind of like, you could distribute that if you wanted to instead of having to make one via Google Doc or whatever. So super cool, um, and again, a couple of different use cases for it.

What's really interesting about it, no, no, wow, okay, maybe it was because I did it all in one day. I think the limit was three, and I almost paid for it. I was like ready because I think the AI tools that I paid for, it's either I buy it upfront just to give it a go, or I'm using it, then I run into the limit, and I thought, "And this is the time." I had already paid for Claude and Windsurf. So I thought, "Do I want to pay?" If I hadn't paid for those, I would have been quick to open my wallet and pay for it. But I was like on the plane and everything. So I, I think they have like a $10 unlimited plan, maybe I saw somewhere.

I think I might have that just because I'm on a Google Workspace plan, okay, both like on my job and then just on my kind of, um, you know, like the podcast type stuff. So I think it just comes with that. I don't know, I should know that, but I have access and haven't run into any rate limiting on in either environment. But what's also really cool about it is that you can like share your notebooks with people, and I think that's actually a really strong use case for it, is you can share the notebook, and you can share basically the same view that you see it. So you can see the sources and like generate the audio and stuff like that. Or you can just share it as a chat, basically like a chat app. So you can basically like grant users, you know, access to it, and if you give them like chat-only access, when they click on the notebook, they'll just open up, it'll look like a, you know, a ChatGPT or a Gemini window, and they can just chat with the sources, but they can't see the sources, or they can't generate the AI-generated audio, and they can't generate the FAQ docs, stuff like that. So I think it's super useful for like, you know, knowledge sharing. Like if you have like standard operating procedures or like kind of process documentation that, you know, is just a static file that people, you know, update once a year, instead of having that just be like a Word doc or Google Doc, you could throw that into Notebook LM. And then when new people come on, or you need to onboard new people, give them like chat access to that notebook. And then they can kind of ask, you know, like, "How do I do process X, Y, and Z?" and it can kind of lay out the steps. So I thought that was a really, really strong use case for it. Because that's what, you know, I've used it for already, you know, immediately. It was, it was trying to onboard someone onto a new process, put the relevant documentation in there, and then just shared that link, and it works like a charm because it just lets people, you know, be more efficient too, because then you're not taking the reviewers or the person who's kind of training you, you know, they still probably walk you through it the first time, but then subsequently, you can kind of ask questions and not take up their time too. So really good for knowledge sharing, I think. And I think that's probably an area that I'm going to continue to kind of use it more because it just, it's so easy to kind of add context to it. You know, it's very, it's very approachable, I think, for most people.

Yeah, I wonder what the largest use case is in the workplace for audio, because the two times I had used it, I was in the realm of digesting audio, as weird as that may sound. When I'm working at my desk or, you know, other areas in the workplace, not all the time am I listening to content. Like either I'll have no music, I'll have background music, like focus music, or I'll just have like very low general music. But sometimes I listen to podcasts at work, depending on what I'm working on. I wonder how Notebook LM could be, I mean, this just kind of came to me, but wonder which areas it would make the most sense to get critical information into, you know, your head through audio. It almost makes you think if the company, maybe a weird idea, if the company could have like an internal podcast, and this podcast was automatically sourcing emails or something that was from your inbox. And then every week you get like, I don't know, a 10-minute, 15-minute podcast talking about important things. And maybe on your commute into work, either you drive, you take the train, you walk, you bike, something. And if you work from home, you could do that while you make your coffee in the morning, whatever. Some avenue to give you, you know, boring or uninteresting information packaged in a way that's fun and approachable because that's my biggest draw to Notebook LM, is I want to learn about things, but it feels almost too high effort to go read a long doc about it, to go read all these blog posts. I want correct data fed into a fun and engaging system. Like I'm not able to talk back to the podcast speakers, of course, but like you mentioned, there are jokes, there's humor, there's grit. You know, it makes it a lot more entertaining. So I wonder in the workplace how far things could go to make it a little bit easier to plug in Notebook LM. Because, again, the times I've used it, I've been on the plane or like walking and, you know, during a normal workday, I'm not on the plane every day, and I'm not, you know, walking in an avenue that I'd be easily able to, you know, kind of plug in this audio content. So I do wonder how companies can be efficient and smart of one, packaging boring information and making it accessible, and two, plugging into like an employee's life to make it, you know, fit within their workflow. So, yeah, yeah.

It's an interesting question because it's like, well, what is the real, what's the real value of Notebook LM? I think a lot of it is the simplicity because I think you can, you can do, you know, the podcast like with, you know, I'm sure you can do it with like OpenAI's voice agents. You know, I'm sure there's a way to do it, but Notebook LM makes it so easy. But as far as right now, I don't think there's, there's no API access. I think it's just purely a front-end interface. Yeah. Yeah. So, you know, I think certainly there's an appetite for that, I think. And I totally relate to what you're saying of like being able to like listen while you're doing other stuff. It's funny when I was kind of demoing it to people I worked with, I said, "Oh, it's great because you can like listen to like work stuff like while you're like walking the dog or like doing whatever." The way it came off was like you can't ever escape work now because it's just always in your ear, you know, just kind of funny. But, but it's true. Sometimes you want to, you know, step away from your desk and but you can still kind of be engaged in other ways. But, yeah, it's an interesting question for sure. I mean, I think there's no technical reason why, you know, instead of having an all-hands meeting where it takes up everyone's hour, you know, they couldn't put the materials, put it into Notebook LM, and then just generate the podcast and distribute that out instead. I think that's like kind of meeting replacement for like distribution of information is certainly a good candidate for it. It's probably still times and places where it's appropriate to like have a meeting because maybe you need to interject or something, but like if you're just kind of distributing information like a one-way street, like an all-hands or, you know, certain, certain meetings where it just tends to be you speaking out into the universe and people aren't like interjecting or kind of you're not expecting any kind of follow-up, then it's, it's a great candidate to just spit out the audio file and just say, "Here you go, like let us know if you have questions," kind of thing, you know, because maybe that saves everyone some time. People can kind of listen at their own, at their own time, and if they have questions, they'll reach out. If they don't, then you know, everyone's hunky-dory, you know.

Yeah, it'd be cool to see kind of A/B testing between a large email blast with like maybe too much content, you know, you get those emails at work that, you know, need like a TLDR, and then you send that same thing but like Notebook LM. And again, I think the hard part is getting people to click on it and listen to it. So that is a bit of friction, but the same thing with a long email, people are just going to either never open it or just scan it. And I wonder if there's a question in there, how many people would kind of pick up on it, and it would then drive, you know, maybe the question is differently represented in the Notebook LM versus the email, you know, doing like a report based on the past week, you know, "What do you think about this?" And maybe there are two different answers, and you could source where they got that info from. But either way, I think Notebook LM is pretty darn cool. I think I'm trying to figure out ways I can plug it in more. I think the deep research to podcasts is an easy win for me. At work, I'll have to think about it a bit more, but very exciting. I think, yeah, once the API opens up, like, you know, Spotify had access to it, it seems like, to build out their stuff. But for the general public, I know Google is very busy trying to win the AI model war on top of everything else. So I imagine there's a lot of demand in that space for them.

Well, and one thing too, what I really like about it is kind of managed context and like managed sources, I guess is another way to put it, because it's not going to hallucinate things that you don't give it, which I find, especially, you know, we talked about the importance of, you know, needing to be exact in certain fields, like in accounting, like it's all dollars and cents. A lot of times there are, you know, contracts that underpin what we're booking, right? And so like, we don't have room for like AI to hallucinate stuff that doesn't exist. And so what Notebook LM does a great job with, and why I think it's got a lot of great potential in certain use cases like that is like it's only going to give you results on the sources you give it. Whereas if I go to, you know, ChatGPT, it's a bit more open-ended. Like you're just saying, "Hey," you know, like there's, there's enterprise RAG that's out there where like it'll search across your whole company, you know, index essentially, if you ask it a question. And sometimes you get, you know, sources that you don't actually want. Like maybe it's talking about the thing that you asked about, but maybe from a different time period, or maybe a different view. Maybe a more of a, you know, a marketing view than it really is like an accounting view. And so like being able to manage the sources and manage the context, I think is super underrated or underappreciated because sometimes, and I think as it gets easier and easier to get information from everywhere, like you're going to need to find a way to kind of filter out where you want to get your information from because, you know, we all kind of have our favorite sources or places that we know, "Hey, I'm going to go to this place if I'm looking for, you know, a certain answer," you know. And that's something that I think Notebook LM does well. So for folks that are, you know, I guess maybe a call to action or a challenge for people that are listening, you know, when you're using Notebook LM, like let us know what you think. Like what are you able to use it for, and like what do you think about like kind of the sources and that managed context? Because again, in my experience, it hasn't hallucinated anything past what, you know, I give it, and that kind of feels like a nice kind of controlled environment almost. So be curious what listeners think. And when, you know, let us know, you know, if you run into any issues or if you, you know, if it works great, you know, we would definitely want to hear it.

Yeah, hats off to Google too. I think their products and context management have been fantastic ever since day one. Google's AI Studio, kind of like the ChatGPT, but developer version, handled large file uploads, integrating with Drive. I guess I'm not surprised since they have Google Drive, but you'd be surprised that all these other AI companies try to play catch up on just file management. And that's the big thing with AI and prompting, is you want to give the AI as much context as you can in the right way. And again, I feel like Notebook LM makes that easy. And surprisingly so, other competitors in the space have a tough time collecting your files, organizing them, kind of doing the RAG approach. Like I think Claude had a limit of like 25 megabytes. Other providers are roughly the same. I could go to AI Studio, drop in 500 megabytes of a video file or an audio file. It uploads it with ease because of Drive and then integrates with Google's smart models. Like, hats off to them. I think they're killing it. It's a fantastic way to just almost have like your own RAG system with that. Then on top of that, you have a podcast, and who doesn't love that? So, yeah.

One last thing I'll add on that too is, um, as I mentioned earlier, it does not support adding Google Sheets as a source, which is really interesting. I'm sure there are like a lot of challenges to that. I'm sure otherwise they would, but it just kind of goes to show, I think spreadsheets are just complete, you know, data vomit a lot of times. And it probably makes it really hard to like put that into context and have AI understand it. But I do know that Google Sheets has recently released like an AI formula in Google Sheets. I haven't used it personally. So I know it's out there and haven't demoed it at all. But I think you basically type like "=AI" and then in parentheses, you kind of give it a prompt. So like, "Can you look at this data and tell me who the top five customers were?" And so I'm going to go back and look into that because it was a relatively new release. And I think from what I've heard, people are finding some success with it. But yeah, no integration with Notebook LM on that just yet. The one other alternative, if you're not into Google Sheets and you're more of an Excel diehard, is Pandas, the very famous data frame Python library that is used by a lot of accounting and finance folks that are, you know, getting into programming like myself. There's a new library. I don't know if it's actually affiliated with the people that made Pandas, but I think it's called PandasAI. And it basically lets you take in a data frame and then, you know, supposedly AI is going to like understand and contextualize the data frame. Again, haven't used it. I've seen it out there, but something to kind of, you know, keep in mind too, that there's, spreadsheets, I think are a challenge for the AI to understand. And even just like uploading my own like CSV files in like ChatGPT sometimes doesn't get it 100% right. But there is progress. Google Sheets has the AI formula, PandasAI, if you're down to kind of battle that one. And I'm sure eventually we will get to a place where you can add Google Sheets to Notebook LM and so on.

Yeah, I'm sure they're listening. And yeah, again, fighting the big fight. I think Google I/O is in two weeks, speaking of Google. They released Gemini 1.5 Pro Beta, I think is what they call it. Again, this is just like an upgrade on their already outstanding model. I think it has better coding and in specific front-end coding. So if you ask it to one-shot something, which means you ask it to do one thing, it'll do all the code for it and have good UI design, which as many of you may know, building apps is very tied to design. People can hate it or love it based on the way it looks. And if AI can make that better, that's amazing. So shout out to Google for pushing forward on that.

And then one other small topic I wanted to bring up, I know we're probably going to wrap up soon, but OpenAI had a bit of a weird situation these past two weeks. So if you aren't familiar, GPT-4o had a bit of a personality update, which was coined by Sam and the OpenAI team. After this personality update, the assistant or ChatGPT generally became very, let me find the word, sycophantic. And I got to ask Google to define this word. What I believe it means is extremely helpful, but almost to a degrading extent where it's trying to help you, but almost like always agreeing with you. And interestingly enough, there is an ex-OpenAI researcher somewhere on the prompt engineering side who had mentioned that early on in his time in OpenAI, there were lots of arguments about specific words used in these system prompts. So, again, when you're talking to ChatGPT, OpenAI has developed this probably 10-paragraph description of how the AI agent should act and respond to you. One of these disagreements came about using the word "polite" in a prompt. The argument was, should we use polite? Should we use helpful? How does that impact how the model behaves? Because a lot of these models are trained on this kind of reward scenario where we're looking for a specific output, the closer it gets to that output, the better. And I don't know if this is actually the case. I think this was an example of what can happen. But it kind of summarizes that OpenAI had more or less a system prompt update. Maybe they changed the word from polite to helpful. That small change might have made the AI assistant extremely on your side no matter what. To give you an example, you could ask things like, you know, "If I died and saved a thousand people, like, would that be worth it?" You know, some kind of like existential, bad news for you, but good news for others, greater good situation. The chat model with this update would say, "Oh, no, like, you know, you should stay alive," et cetera, like kind of like lean in your direction, no matter what, excluding all outside factors. Then Sam and OpenAI kind of went online and said, "Hey, the model sucks right now. I'm sorry, like, we're going to roll this back." And they posted a whole postmortem on what happened with more details. I won't go into it, but I think at the end of the day, it shows that these companies are shipping faster, where they are continuously. Google is fighting with Claude, is fighting with OpenAI. People are shipping fast without checking things as much. Two, small changes can really add up. In this AI world, yes, people are experts, but it is a black box to some degree. These changes, while they might be small and insignificant, can really add up to that chat behavior and personality. So very interesting. I'll link it in the show notes, but it was OpenAI trying to make something more helpful, but ultimately ending up to be something that was not helpful in the way that they had not expected, and rolling that back and learning about it. So pretty interesting. I didn't really chat with it much during this period, but I saw screenshots of kind of ridiculous examples of it siding with you no matter what. And I think as I've used better models, I realized that better ones actually challenge you and make it better, not just say, "Yeah, that's a great idea," no matter what.

Yeah. Yeah. The classic yes-man. You don't always want a yes-man. That's for sure. Cool.

OK, so let's wrap it up here with bookmarks. I had a quick one, actually. And Brad, this is actually news. OpenAI is buying Windsurf. I just, I guess I had bookmarked it this morning at 5 a.m. So from Bloomberg, OpenAI reaches agreement to buy Windsurf for $3 billion. So my bookmark was actually, and I had sent this to you I think right before we got on here today, but it's from Hacker News. For people that, you know, aren't familiar, it's like a forum like Reddit that basically talks about indie hacking and building businesses. And it's a post, I'll share the tweet, it's a tweet by Sam Hogan, but it's a screenshot of Kodium, who makes Windsurf, the company that makes Windsurf, kind of saying, "Hey, we released Windsurf, it's a generative AI IDE, check it out, let us know what you think." And then included in the screenshot is a response from someone that basically is just like dumping on it, saying, "You know, what's the difference with this and Copilot? It did worse on, when I did my testing on it," you know, kind of just like, you know, being a little bit challenging to them. And maybe there were good questions at that time, but I just think it's a great thing to show like just sometimes just just move forward and have some conviction in what you're doing. Clearly, it didn't stop them at Windsurf, so kudos to them. Um, there are always going to be people out there that are doubting you or naysaying you. So just move with conviction. You know, if you believe in something, go for it and don't let people on the Internet sway you. So you might turn into a $3 billion payday for you one day. You never know.

Yeah. I really wonder how many people they have. I know they're not. They're definitely small. And we've talked about the $1 billion AI acquisition with one person. So building a business using AI tools and being $1 billion. I don't know how many they are, but I know it's small, and that's super exciting. So congratulations to them. Cool.

Mine, I'll also go through it quickly, is: "Will software engineering jobs be the first to be displaced by AI?" And this has been a long-running thread as AI tools, AI IDEs, everything is kind of geared around coding because it's an environment in which LLMs do well. It's text in, text out, there's compilation, there are error checks. It's a pretty good kind of playground for AI. And I linked an article that talks about various scenarios that AI will affect software engineers. And the one that I found interesting was kind of this discrepancy between, given better tools, will we see fewer engineers, or given better tools, will we see a lot more software being created and therefore a higher demand? I can't remember the exact paradox or name for this kind of phenomenon, but it exists. And my two cents is, sometimes I get like pretty impressed by the models. I'm like, "Damn, like this will make me work a lot better." But then other times I think like new grads coming into software engineering, a little bit harder market, people are expecting you to be good at AI tooling and have the background. So honestly, I don't know how to place it. If I were to bet right now, I'd probably bet on the side of better AI tooling, a lot more software being created. It's still having demand, but there are a lot of doomsayers saying that software engineering will be a lost art in the next few years. And if we've seen anything recently, a lot of the CEOs from companies like Meta, and those behind models like Claude, are saying, you know, 80% of code written by companies is now AI, and that's only going to continue. But I remain hopeful. I like doing what I do. And I think piecing systems together, AI has not done a great job at that just yet. So we're safe. But an interesting article nonetheless, kind of walking down all these paths of, given AI increasing in kind of output and complexity and handling larger tasks, what will that mean for software engineers? So if you're a software engineer listening, take a look at the article, pretty interesting. And yeah, let me know what you think, where you land on kind of that spectrum of fewer jobs, more jobs, or maybe somewhere in between.

Yeah, that's an interesting line. Um, you know, 80% of code will be written by AI because I feel like, you know, that's still, like you said, the juice a lot of times is like hooking these things up and troubleshooting when things go wrong. So, um, you know, maybe at face value, one might go, "Oh, wow, that's, that's pretty dramatic or pretty scary," depending on how you look at it. But, um, I think similar to accounting, you know, there's a ways to go. I'm not saying it's impossible. Maybe we'll be there in the long run, in fact, I'm sure we will. But like, I think we got some time, you know. So yeah, like you said, just let's get used to the tools, get familiar with them, and make yourself more productive in the in the short to medium term and, you know, see what happens from there.

Yeah, yeah, it's definitely a catchy title, but will it actually play out in certain ways? I don't know, but you know, I'll be using it. So whatever happens, you know, I'm riding with the tide.

Yeah, absolutely. Cool. Awesome. Well, good stuff, Brad. We'll end it there, and yeah, talk to you later.

Sounds good. See ya.

See ya.

Thank you for listening to the Breakeven Brothers podcast. If you enjoyed the episode, please leave us a five-star review on Spotify, Apple Podcasts, or wherever else you may be listening from. Also, be sure to subscribe to our show and YouTube channel so you never miss an episode. Thanks. Take care. All views and opinions by Bradley and Bennett are solely their own and unaffiliated with any external parties.