Beyond The Prompt - How to use AI in your company

The death of SAAS and raise of what marketing folks can do with AI already with Noah Brier

Episode Summary

In this conversation, Noah Brier, founder of Percolate and BrXnd.ai, discusses his journey from advertising to becoming a leader in AI-powered marketing solutions. He explores the balance between exploration and exploitation in business, highlighting AI’s transformative role in solving complex brand challenges. Noah underscores the importance of hands-on experimentation, the shift from traditional SaaS models to bespoke software solutions, and the need for companies to capture and leverage tacit knowledge. The discussion also covers practical applications of AI, insights from leading CMOs, and how technology is shaping the future of business. Noah explains how AI can democratize knowledge, streamline workflows, and turn unstructured data into actionable insights, positioning it as a critical tool for driving innovation.

Episode Notes

In this conversation, Noah Brier, founder of Percolate and BrXnd.ai, discusses his journey from advertising to becoming a leader in AI-powered marketing solutions. He explores the balance between exploration and exploitation in business, highlighting AI’s transformative role in solving complex brand challenges. Noah underscores the importance of hands-on experimentation, the shift from traditional SaaS models to bespoke software solutions, and the need for companies to capture and leverage tacit knowledge.
The discussion also covers practical applications of AI, insights from leading CMOs, and how technology is shaping the future of business. Noah explains how AI can democratize knowledge, streamline workflows, and turn unstructured data into actionable insights, positioning it as a critical tool for driving innovation.

Key Takeaways:

Noah Brier's blogpost on building plugins: https://newsletter.brxnd.ai/p/building-chatgpt-plugins-brxnd-dispatch

Noah on LinkedIn: https://www.linkedin.com/in/noahbrier/


00:00 - Introduction to Noah Brier
00:45 - Noah's Career Journey
01:25 - Exploring AI, New Ventures, and Transitioning from Exploration to Exploitation
03:28 - The Power of AI in Solving Problems
07:14 - The Practical Applications of AI
10:26 - Building AI-Driven Workflows
16:35 - Empowering Organizations with AI
22:20 - Hands-On with AI
31:36 - Making Tacit Knowledge Explicit
33:10 - Fun and Quirky AI Projects
35:30 - Using AI for Brand and Product Development
37:33 - Exploring New AI Tools and Techniques
39:59 - Challenges and Opportunities in AI
42:26 - The Future of SaaS and AI Integration
47:03 - Practical Applications and Prototyping
47:42 - Reflections and Future Directions
53:49 - Closing Thoughts and Takeaways

📜 Read the transcript for this episode: Transcript of The death of SAAS and raise of what marketing folks can do with AI already with Noah Brier |

Episode Transcription

[00:00:00] Noah Brier: The part that needs more hype is just, it does so much right now. I think that part of the challenge in the market is there's so many people talking about the future and what it might do. That people have kind of lost track of like how powerful it is just to In its current form.

[00:00:15] Noah Brier: Hi, I'm Noah Breyer. Uh, I guess I'm a reformed SAS founder spending about 15 years building software companies. And more recently the founder of a company called brand AI BRXND. AI, which is an organization at the intersection of marketing and AI, which is, uh, We put on events for marketers all around the world. And then most recently, although I haven't really told anyone yet, a new company called Alephic whose aim is to help CMOs answer their hardest questions using AI code and expertise from inside their organizations.

[00:00:49] Henrik Werdelin: Noah, thank you so much for joining. You and I obviously know each other for a long time. Maybe just explain kind of what you do because you are a man of many talents.

[00:01:01] Noah Brier: Yeah, well, I guess I am a little bit of a bunch of different things. I started my career as a writer I was a journalist for a little while and then I moved into advertising I was a creative director copywriter did a bunch of stuff there and started my first software company in 2011 Uh, it was a company called percolate. We sold that in 2019. We went through the whole process, raised lots of money, did the whole thing. And then, uh, about two years ago, I was like, I just need a break from everything. And, uh, just, Went back to writing code. I wanted no employees and no investors. I just , didn't want to talk to anybody. I just want to sit in a room and build things. Um, and it was a good time to start building things. Cause AI was sort of really just, you know, it's like GPT three APIs were available halfway through the year, , some of the image models started to come online. Um, Henrik and I worked on one of the very first AI things that I worked on together. Um, Yeah. And then I just been , essentially exploring over the last two years, um, actually lately I've been explaining my journey as , you know, the explore exploit trade off. Um, so, I've been kind of thinking of it as like, I went into explore mode for two years and then actually over the last sort of six weeks, I finally decided what I wanted to do. And so I'm sort of starting to make the transition out of pure explore mode. Um, because I was doing a bunch of fun stuff building things with people. Big brands. And I was like, I think this is kind of what I want to be doing.

[00:02:27] Jeremy Utley: And okay. Okay. Hang on. Hang on. There's so many things that already you just have to stop because first of all, I don't know if Henry told you that this would be catnip for me, but explore exploits, what do you think is going to happen here? Henrik? Uh, how do you know when to shift? Let's just start there, by the way, we'll talk about AI at some point, I'm sure, but as you have been an explorer and you know, it sounds like you're moving into exploit, what. Were you keying in on as an innovator or as an explorer that said, okay, now it's time.

[00:02:57] Noah Brier: I think really for me, the what's happened. So, my journey over the last two years is like, I was Building things. I was experimenting with things. Henrik and I built a funny dog generator, um, where we crossed all the dog breeds together. And I started a conference about marketing and AI, did all this stuff. And then a bunch of brands came to me and I was building things with the brands. And really what happened is sort of like around six months ago, I started to get calls from CMOs who were like, Hey, we have this problem that we need to solve. And actually like, it doesn't really have anything directly to do with AI, but like maybe now that AI exists, we could solve it. And I was like, Oh, well, that's pretty fun. And I started doing a bunch of those and I was like, yeah, this is awesome. I kind of just want to be doing this. I just want to work on really weird, hard problems that two years ago, You could have solved maybe with like, you know, six months or a year or two years and lots of people and lots of budget. And now we can do it in two weeks or four weeks.

[00:03:55] Jeremy Utley: I was going to say, don't say weeks. Okay. So what's like the canonical example of a thing that a CMO asked for

[00:04:01] Henrik Werdelin: before, before you remember that I'm just going to tease out a little bit, because honestly, no, you and I, we've spoken to this before, but like, I am obsessed with this in between time, which is basically this wonderful, vulnerable.

Exciting, curious, daunting time that entrepreneurs have when they are in between what they have just left and their new thing. And if you're a lawyer, then your next job is as a lawyer. So, you know, life is kind of easy in that way. When you're an entrepreneur, we tend never to want to do the thing that we just did. And so it's kind of weird. And everybody, because we're so linked to the narrative of our last startup, we kind of have to go out in many ways, redefined our self narrative. And so as you're coming out on the other side of the in between time for people who have have either started or going into that phase, takeaways on how to make that less stressful and more enjoyable?

[00:04:58] Noah Brier: Well, I think for me, a lot of it was like, Starting from a place of returning to the stuff I like doing the most, right? So it's like, you know, after many years of running, a relatively large company and spending a lot of time managing people and dealing with HR issues, one of the things I was desperate to do was just get back to work like building things and sort of feeling the creative energy that you get when you make something.

And so I think my main priority was to Open up more time for that. And let everything kind of flow from there. I'm a hands on learner. Um, I think one thing entrepreneurs have in common is that they're not very good at learning things they don't experience.

Um, like you, sort of, you've got to have that thing. And for me, you know, as somebody who also likes building and writing code, that's particularly true with technology. And so, I think that was the biggest piece is just kind of like finding that baseline. The thing I missed most in that time of, running Percolate was just, you know, eventually , you've built a company and you're hiring the people that hire the people that build the things, and I was just desperate to get back to, Being the one who got to build the things.

And so everything really flowed from there for me. And then it was very much just being in the explorer mindset. I didn't think about it in explore exploit terms until very recently. Um, now that I look back, it was like, I just sort of said yes to everything. And I kind of did lots of random stuff.

I put on a conference for no reason other than I was being interviewed for a article and I said I was going to have a conference. Like I, I didn't have a plan before that. And I actually, the guy who was interviewing me, I don't think he really believed me. And he was like, are you sure you want me to include that in the article?

And I was like, yeah, totally. Yeah. Cause then that will hold my feet to the fire. Yeah. So I was, it was very much like, sort of yes mode, but also, I don't know, this is the most magical, magical. Moment I've ever experienced in my life, right? I mean, I remember when I got my first computer, but like, I very vividly remember a few years ago, the first thing that blew my mind was taking a unstructured page and sending it to the GPT three API and with some Jason and getting a structured output back.

And I was just like, Wow, I'm like never going to do this any other way ever again in my entire life. And it sort of all just followed from there on the AI side. Like a lot of the work I've been doing and even putting on the conference was like with the explicit intent of, um, I think AI kind of like needs.

In some ways it needs less hype and in some ways it needs more hype. Like it needs less hype in that we don't need to promise these giant things. And I think like whatever, AGI may or may not come true in whatever timeframe and people can speculate on that, but like the part that needs more hype is just, it does so much right now.

I think that part of the challenge in the market is there's so many people talking about the future and what it might do. That people have kind of lost track of like how powerful it is just to In its current form. And that's sort of like, I feel like I've spent the last two years trying to convince people that like, you don't really need to look into the future on this one. You can just.

[00:07:58] Jeremy Utley: So one thing you're reminding me of that Ethan Mollick said to us when he was on the show was even if all development stopped, it's going to take the world 10 years to catch up with the current capabilities. So again, not thinking about the future the overhyped part, but thinking about the present and the under hyped part.

What. Are you seeing that you would say is currently under hyped? What are the specific capabilities or modes of interaction, use cases, et cetera, that you go, people really knew even just to do this, , their lives would be different.

[00:08:31] Noah Brier: , well, I have a great one. , I was at a dinner recently and I got seated next to the CIO of one of the biggest media companies in the world. And. He was sort of like, you know, a little bit maybe of an AI skeptic. And he said to me, like, how do you articulate the value of this stuff? And I was like, I don't know, like how much of your job. As CIO is about making data from one place, be able to work in data from with data from the other place. And he was like, ah, kind of all of it. And I was like, that's the thing AI does best, right? AI does the transformation of data from one format to another. You know, that was my Jason experience, right? Like I took unstructured data in the form of a webpage and turned it into structured Jason, but the same is true everywhere. And if you think about like. How much work happens inside organizations that is fundamentally like asking a person to turn unstructured data structured you know, Salesforce and CRM is about taking salespeople and asking them to turn their unstructured sales meetings into structured data, right?

Like now we record the meetings and literally I just, , built a thing for my new company where we take the meeting transcripts and , you know, Out comes a full readout of the meeting in the exact format we want with the sort of opportunities identified and everything else.

[00:09:48] Henrik Werdelin: I'll give you another example, which I think would be so much of your world and you probably already done it.

But like we've started a bark to have. Um, build bots that allow the salespeople to get interviewed by a bot so they can write a design brief in the way that designers really like it. And it just seemed to be like such an obvious thing, but like, obviously they have all the information. They just don't know how to articulate it in a way that the designers kind of enjoy.

[00:10:12] Noah Brier: Totally. And , you know, I mean, to me, sort of the core feature of AI is that it can be this fuzzy interface that sits between anything and anything else. And a lot of problems can be boiled down to needing a fuzzy interface that can sit between anything and anything else.

Like it just becomes a creative exercise of how you're going to get there. And sometimes you need more than one step and you know, we can go into kind of how you build and what you build and why you build. But to me, it's that fundamental thing. Like. If you stripped away everything else you can do, you know, I think there's so much focus on it, like writing things and whatever, like the thing I use most in whether I'm working myself or working for other folks the kind of core function is taking data that currently exists in one format, whether it's unstructured or structured and changing it into a different format so that you can use it in this totally new way that somebody didn't imagine.

[00:11:04] Jeremy Utley: It reminds me about, uh, one of the things that one of our partners and I have worked on is a bot Henry, he's a bot or made me think of this, but, uh, a pro sports organization realized that their HR department, because of many initiatives, which we don't have to name the HR departments, Processes are increasingly onerous on hiring managers and the rest of the organization.

And they have very kind of specific formats that they need stuff kind of output in or input into, depending on how you think about it. And we basically help them identify an opportunity to build a bot, which takes a hiring manager from an unstructured conversation into the specified kind of.

Characterize it the way you want the output that HR requires. And no surprise hiring managers like that a lot more than sitting with the word document or the PDF that they've got to edit. Right. But it's actually not hard. But the other thing I would say is It reminds me, Henrik, of the conversation that we had with Greg McKeown and Dave McRaney, which never may be published.

I don't know, but is this idea of English to English translation. I mean, Noah, what you're talking about is one format to another. At the most kind of conceptual level, there's also the Jeremy English to Noah English. And for example, if Noah English is not the way you understand that, if what you replace the word dialect, right.

Or replace the word personality, right. But , there's also an opportunity for the fuzzy interface to actually enable human beings to understand each other in a way that they previously don't.

[00:12:35] Noah Brier: Can I ask you guys a question on that one actually? Cause the HR one is very sort of top of mind. For me, not specifically, but generally like one of the things I really wonder is like, is it the best thing in those situations to build the thing that you're describing or to just recognize that it's really dumb that we have all these things in place that involve all this bureaucracy and this way, this sort of heavy duty requirements that, and try to fix those. I don't know, that's one of the ones I've been running into in my own work quite a bit lately.

[00:13:07] Henrik Werdelin: Intellectually, I think that if you think of everything as inputs and outputs, I think sometimes we have to make the input output in the process big because that is what everybody will feel comfortable with knowing that at one point we can skip that step.

But I think right now, like we We're basically all LLMs in one way, right? You know, like I'll give you a prompt as a question and you give me a response as a, as a vocal answer. And I, you know, in a kind of a weird way, the way I'm computing. The future of a lot of work is that we will be prompting humans and LMS, and then we would get responses and we would choose basically the LM or the human, depending on what the skills are, those specific, um, the thing that, uh, question for you on the, um, on the meeting stuff, we had a guy on the podcast recently, and he did a pretty cool thing that after, He basically then showed how he would normally do, uh, basically how he would, uh, feed the, the interview into the, into NLM.

And then he was extracting things like what are the key insights, what are some of the quotes I should keep and stuff like that. I sure just was talking about, you just did it. Are there anything specifically you're trying to extract out of meetings that could be useful?

[00:14:27] Noah Brier: No, I mean, well, I think that the broader point here, which is, again, in sort of my own experience and my own decision to kind of move in a little away from the sort of pure explorer is like, I have found that the kind of the real trick with all this stuff is that you combine the AI, you combine code, and then you combine this sort of unique expertise of the individual who has sort of a specific way that they want to work.

And so, you know, I have. We have our own ways. Uh, I'm my new company. I started with a guy you might know named James gross. Um, and, uh, James is coming back together.

[00:15:07] Henrik Werdelin: Jeremy, this is like the equivalent of the blues brothers, where it seems like Nora's like about to now drive a truck around and getting people back in the band.

[00:15:15] Noah Brier: So, you know, James from our percolate days had a very specific format. Every salesperson in the organization used to publish a thing called the field report after every sales call. And in that field report, it had a kind of fairly specific format. Everybody hated it. Right. Cause for the exact reasons we described earlier, like they don't like having to go in and write stuff up after the, had the meeting, um, but it was really helpful and it was a great way to kind of distribute knowledge and information throughout the organization.

And so. Um, you know, I think what we have done is taken a lot of his thinking about what matters and what you're listening for when you're having a call with somebody that is specific to him. So, you know, like, what are the opportunities? And specifically, what does an opportunity mean? And, , all of these types of things.

But I think, you know, again, in my Own journey. I think I started from a place of like, well, maybe there's a SAS company to start, right? Cause I've spent a lot of time starting SAS companies. And I think one of the conclusions I came to is I don't know what the SAS company is because I think so much of the value is in the specificity, right?

So, you know, our specific workflow and output that we're looking for, , is probably fairly generalizable to companies that look like ours. But, you know, the beauty is. To change the specifics to exactly the thing you want, right? I mean, that's not even weeks or days or even hours, right? That's like, three minutes in one of the prompts to change a couple of pieces, like it's not even, you know, like

[00:16:39] Jeremy Utley: there's a problem to your point. It's actually that's where the majority of the work happens is that last little bit.

[00:16:44] Noah Brier: Yeah. Um, and so, there's a lot of work in sort of breaking down the problems.

I think one of the other things I've come to recognize is that, like, um, I think there's a little bit of confusion in the market about the difference between AI models and AI software. Um, and , , we even see with the new open AI one model, right. It's sort of doing this chain of thought process that we have been doing, you know, anybody building things has been using sort of chain of thought in various forms for a while now. It was a paper published by Google in 2022. And I think a lot of what. I've built over the last few years has been not just chain of thought on an individual prompt level, but chain of thought blown out to workflow.

Right. So it's like you have chain of thought chain together multiple times. And that's how you get a valuable output. Um, and, the reality is that like most hard problems require that kind of workflow. Right? Like I was talking to a journalist recently and trying to explain this. And , um, I just said, like, sit down for 15 minutes with a pen and paper or whatever your computer and just write the article versus run your regular process where you like, do research, you interview people, you like, figured out, you talk to your editor, you go back, like, what's gonna have a better output? Obviously, the Large language models themselves are great at that single step, but when you tie them together and you use many steps, that's where the sort of real value is, and you can build that expertise, not just into the individual prompt, but into the way you sort of architect the whole system, right?

And so , I've come to find somewhat surprisingly that like my experience in building software systems, um, Turns out to be like really useful in this AI thing. Cause you know, one part of it is how do you prompt the system? And some of that just comes from like experience. But another part of it is like, how do you model the data? How do you model the workflow? How do you think about breaking things up? Um, and that's like kind of software development one on one, right? Like talk to any engineer.

[00:18:45] Henrik Werdelin: Noah, if I can go back to the basically the experience one of the things that I often get asked is obviously same questions as you were being asked by the CIO is like, what can I use it for now? And I think that's a really like important kind of like point and I obviously use it for everything. And so have gotten incredibly more efficient. One of the things that has been complicated as I've been thinking about it is, When you try to atomize what I'm doing is that I have a little bit of like experience in a specific domain.

I know a little bit about AI, so I understand which foundational models I should use and what kind of APIs that are out there. I know a little bit about technology, so when I need to kind of tie these things I can, you know, look at JSON or Python and then I enjoy talking to people so I understand how to wrap all this kind of like into a chat or communication kind of methodology that kind of is relevant to this space perfing.

None of these things are very complicated. You know, like if you're sick one, but if I was thinking like, who do I send to somebody, if somebody is asking something, like, who should I hire to help me come up with like 10 bots so that I can be 40 percent more efficient, you're kind of going, Oh, he can't really be an engineer or classic one. , it's not a business person, right? It's not the AI expert. So what is the steps that you suggest to people to help them through this journey?

[00:20:13] Noah Brier: I'm not sure yet. Um, I don't know. It's like I do have a bunch of those things. Right. And, like I, I write code, I've built companies. , I sort of generally think in products. Right. And I do think, that you're probably, massively under selling your ability to think in products. Because I think that that again, like, I think even bot sort of undersells it like what we're building is kind of like applications in which we're outsourcing sort of intelligence and decision making.

In various steps, right? And so your ability to sort of break down problems in the way that you naturally do when you build products, it turns out to be like very helpful for this, right? Cause you're like, Hey I need to figure out, well, but to do that, well, I have to break it into three problems, not just one problem.

And then actually like, this is the best way to solve this one. This is the best way to solve that one. Um, I think if we. Step all the way back

[00:21:04] Noah Brier: , uh, I had this sort of moment , a year and a half ago now almost where I got access to build a chat GPT plugin for the first time.

And, my job over the last two years has been do whatever I want. And so, like, I canceled all my meetings morning after I got the email that said I had gotten access to chat, GPT plugin API. And I was just like, this is what I'm going to do today. I'm going to sit and like, try to build a plugin and see what it's like. And so, you know, I've done a lot of software integration and then I went to read the API docs and, I expected to see API docs that looked fairly familiar. And when I opened the chat, GPT plugin, API docs basically it said You put a manifest file in the root directory of your application and in that file you describe how you want data sent to you and you describe how data is going to be sent back to us and we'll deal with the rest and like that is fundamentally the opposite of how everything else I have ever built is built because in every other case.

Whether it's Google or meta or whoever, whatever API Salesforce, you're interacting with, they write the API spec. They tell you how to send data. They tell you how data is going to get sent back. As long as you do both in exactly the way they describe, everything is going to work fine. And so, you know, I had this moment where I was like, Oh, , this is actually counterintuitive, right? I have, um, uh, a lifetime of intuition for how to use computers and how to build things. This runs counter. To that intuition in the like truest sense. And so I think, to answer your question of like, where do you start and why is it, honestly, I think people just have to get their hands on it.

And I think part of the challenge with answering the question, Henrik, about what is this good for? What do you use it for? Is , some of it is just that so much of it. Is so trivial, but when you put it all together, it's amazing. Right. So, you know, I can talk about this kind of big application I just built for, you know, pulling down the transcripts and doing all this stuff.

But like the reality is, on my day to day basis my keyboard shortcut for the chat, GBT Mac app spends a lot of time, like, And it's like, Hey, can you fix the bullet formatting on that? Just like the stupid stuff that took you 90 seconds. Or, when I'm writing code, it's like, okay, I could do that.

That would take me three minutes. Like I might as well just have the. System do it like yesterday, I've been using cursor now which is the new, you know, whatever. It's not so new, but they've added support the ID that's added support for son at 3. 5. And, uh, you know, the big thing I was doing was when you're building an application, there's a bunch of pages that are basically the same. They just have slightly different data pulling into them. And you try to sort of abstract as much of that as you can into kind of components or whatever. But fundamentally, you still have to build a bunch of pages that have very little difference.

And so I was just going in and eventually I was like, wait, I could just prompt this thing and be like, Hey, can you look at this page and this page and just make a page for companies that works just like . Those two, I just want the list. And it's like that is not hard work. It's three minutes of my time that, but now it's, It compounds.

It's 10 seconds, right? You're just like, oh yeah, that's right. Like that's

[00:24:06] Jeremy Utley: okay. One thing I want to do, Noah is come back to your comment about just people got to put their hands on it. And , that's actually a recurring theme on this show. You talked to Jenny Nicholson who I know, you know, right.

She talks about leaving her ad agency and having that, little house on the Prairie moment. I don't know if you've listened to our conversation with Russ Summers. It wasn't until he was laid off from his previous job that he goes, he had the space. So here's my question. Very few people are afforded either entrepreneurial, you know, free range or a layoff.

Do you think about for friends of yours, maybe who aren't entrepreneurs, who aren't, you know, largely autonomous self directed agents, but who are in the context of an organizational structure, for example, how do folks practically create, whether it's mechanisms, accountability, et cetera, to actually get hands on the device, so to speak?

[00:24:59] Noah Brier: Well, I think in organizations, it's a pretty funny moment right now where they're pretty hungry for people who are just going to go. I mean, like, yes, you have to be a little bit self directed and you have to, have be a self starter who's going to go do that. But I think, you know, when I look around at the kind of companies that we work with increasingly what I'm finding is we're working with folks who are not like the person with that job. They're the person who sort of just, found a way to get going with it and just figured out ways to integrate it into their work. And, normally it started outside of their work.

And, it's funny cause I look back, like when we started percolate, it was sort of in the social media, uh, explosion and, uh, kind of much the same thing happened inside enterprises, right? Like, at first it was cMOs saying, Hey, I hear about this social media thing. We've got to do something.

They tried to make it happen top down. And then eventually they realized like, Oh, actually there's like probably the best person to deal with this. Is the person who knows the most about it. And it sort of became a bottoms up thing. And the funny thing is now like 10 years later, many of those people are CMOs.

Um, right. They've moved to that point, so , I'm not sure if it's a cop out, but it's like, the answer is you just have to like, just do it. , but again the challenge, and I totally get it. And I have a lot of empathy for folks. It's like, you ask somebody what should I do with it?

I feel like the thing everybody goes and does is they ask it for something stupid and it gives them something stupid.

[00:26:23] Jeremy Utley: Or they Google their own name.

[00:26:24] Noah Brier: Yeah, whatever it is. And it's like, At the first conference I did I found this German word, which I'm not positive is a real word, but I've been running with it.

It's a finger zen fuel, which is like fingertip feeling. That is a real, that is a real word. No, you just made that up. You just made that up. That's not a real word.

[00:26:40] Henrik Werdelin: It's a, it's a real word because , the Danes have adopted it. Really? Yeah.

[00:26:45] Noah Brier: Um, yeah. So it's like, you know, you need fingertip feeling, right?

That's how you sort of learn. You know what it feels like. And this is one of those things where it's like, uh, I think it's just so hard. So you just have to go try it and find ways to integrate it.

[00:26:59] Jeremy Utley: I actually love that. I totally love it. And I know Henry's going to move our conversation along.

So I just want to give you the gift. I don't know if you can see this post it, but this is a gift to you. Okay. Thank you, Noah, because you asked the question. Um, about the HR practice, like, what do you do, I don't know if you remember, but like the fuzzy interface between what's required or whatever.

And I realized as I was listening to you that you actually answered your own question in describing the field report. And here's the phrase I thought was really powerful. Hated, but helpful, right? And there are processes that people hate and they hate probably because they aren't naturally inclined to do it, but it doesn't make it wrong.

Right? And your co founder, it sounds like identified a hugely helpful practice that sure. Other people may not naturally do, therefore they hate it, but it's so helpful. It's worth overcoming that activation energy to me that the answer to your question lies in the balance of those two things.

[00:27:56] Henrik Werdelin: The other one that I've heard recently is what is it that you do at work that you don't get paid to do?

And that most people have like something that they just end up always being the one that does it. And that also seemed to be kind of like an interesting place to go and hunt for some of these things.

[00:28:13] Noah Brier: Yeah, and you just keep finding new ones, right? , one of my favorites lately has been when I'm like working on some new problem or research I do a lot of research.

I was like researching some new area I realized I could load up a GPT with 20 papers and I've got a specific prompt and then I kind of think with that GPT So it's like I was doing a bunch of explore, explore it reading. And so I like loaded all these explore, explore, or actually the other one recently was tacit knowledge.

And that's the other one I wanted to sort of comment on from your comment, Jeremy is , another way to kind of frame it is like, there's a lot of things inside the organization where management is trying to drive. The like making tacit knowledge explicit, right? They're trying to sort of like help everybody to understand the thing that lives in your head because we know so many people learn so much through experience.

And there's real value as management in helping others. And I think I offers this like amazing, amazing opportunity to kind of turn the tacit explicit. And I think you know, that's a piece where , we can't even comprehend yet what that's going to mean for companies.

Right. Like I was just listening to, Toby look, okay. , the Shopify CEO on a podcast. And, he was basically saying like the exciting thing about running a company right now is like, to imagine how terrible the way we're running companies is going to look in 10 years.

[00:29:37] Jeremy Utley: That's great. I mean, even a simple GPT is effectively a codification of tacit knowledge, right? And that's one thing that I liked about Ginny. In our conversation with her, we were just asking, when do you know you need to build a GPT? And she said, when I want to share knowledge with someone else, right?

That's so simple, but it's such a good way to think about it. , it's a way of externalizing your thought process and effectively scaling your own tacit knowledge.

[00:30:01] Noah Brier: Exactly.

[00:30:02] Jeremy Utley: I wonder, , so this is a little bit of a tangent and if it doesn't trigger anything, no worries, but are there ways to identify the tacit knowledge That's worthy of a GPT, you know what I mean? Like what are their heuristics or kind of rules of thumb for which of my areas of tacit knowledge should I be codifying?

[00:30:19] Noah Brier: I have this two by two for what it's worth. Cause I think like, although I don't get this question in exactly these terms, I get a sort of version of it, which is like. The version that you get asked from enterprises, right.

It's like, where do we start? Which in my mind is sort of a version of that question. It's like, , how do we qualify what kinds of problems? And so my two by two is kind of, um, the Y axis is impact on the organization and the X axis is the complexity of the problem, right? Like how many steps does it take to solve it?

,so in that case, like low impact, low complexity, that's just like, Give everybody access to chat, GPT or Claude or whatever, Gemini, one of these models and let them figure out their own stuff. That's a lot of the, um, you know, change the bullets or even some of the do the job that is not part of my job technically.

And just sort of like help me solve that problem. I think the top left, which is the low complexity, high impact, that's the place where to me, GPT is going to have the biggest impact and a good place to look is. Where do you have problems that there are a sort of small group of people who are really good at them and it'd be great if everybody could be just that good.

And I think. In that way, it also kind of reframes a lot of the conversation around AI. Cause obviously so much of it is about what's going to happen to the workforce and, you know, we can talk about that if you guys want to talk about it, but I think that reframing what the potential of AI as distributing best practices, sort of like raising all boats is really interesting and really true. I think that's exactly what Jenny is saying, right? It's like in those moments where you want to take the tacit knowledge that exists in your brain and make it explicit and available to other people. That's what it does. Now, you know, again, GPDs are best at solving a specific kind of problem, right?

They're best at solving those single step problems because you're dealing with the model almost independently. And then, when you get into the kind of multi step high impact is when I think you think about. Building something greater, whether you call it a bot or a agent, or, you know, it's like when I'm describing the transcript thing, , that runs through 16 steps to get to the final output. And it wouldn't be anywhere near. The level of fidelity without running through all those steps.

[00:32:32] Jeremy Utley: Practically speaking there in that case, do you chain GPTs together? How do you call the various steps of the, or is there someone manually actually?

[00:32:40] Noah Brier: No I'm just writing code. So I build it, whether it's like build it in code.

I find air table to be a great platform for building. Um, I really love it. They've added some nice AI functionality. You can write code in there. And so that's a way to sort of like build those multi step processes without building all the sort of scaffolding and code around it. But I would say over the last sort of year and a half, I found a stack that works for me and I've made it really easy for myself to just kind of stand up and deploy apps. And this transcript thing probably was five hours of work from start to finish.

[00:33:16] Henrik Werdelin: , um, the, the project that you and I worked on together, which was fun was, uh, Jeremy was, . We were kind of riffing around the fact that there are Wikipedias for dog breeds, but not for mixed dog breeds and so Norik helped create like a mixed dog breed and so there's 336 official dog breeds and so you suddenly wants like a Wikipedia of all of them.

It's like 336 and 336 which is what? What

[00:33:46] Jeremy Utley: factorial do you get a factorial in there? You've got to.

[00:33:49] Henrik Werdelin: Um, and so, no bill is really fun site where we basically had GPT write the description of what these breeds might kind of like, well, it might be some of the properties and then render an image of what they might look like, and it was remarkably well in the way that it performed, but Noah

you have other, like , you're very good at making these kinds of small apps all the time. What are some of the. fun little quirky apps that you built recently.

[00:34:17] Noah Brier: Yeah, well, right after we did that dog one, I did a brand collab one that was sort of in a similar vein. It's a brand brxnd. ai. Um, but you could take any two brands and you choose the product and it'll make the collab. And what it goes through, it runs through this process where it actually, I took a bunch of collab announcements and I used those to fine tune a model where it writes the collab announcement first and then from the collab announcement it actually generates the image.

Um, to me, the theme there that is really fun is like making things that could exist but don't exist. , in mass, like that's really interesting. Actually, I was recently just having fun. Um, I haven't launched this particular experiment, but I built a e commerce site, like a fake e commerce site, but it all generates on the fly.

It's a little bit inspired by web sim. If you've played with web sim, but it's all e comm. So you like type in a product and it like builds the search result page. So I was doing that. Playing with one of those super fast, uh, inference API APIs. And, and so like literally builds it all on the fly. And then if you decide to click on one of the products, it builds the entire product detail page on the fly.

And the whole, the whole idea is like the prompt for , um, , all the products is like, be real, just like, 15 percent off from reality, right? So like, like it should seem almost real, but there's sort of something fundamentally removed, um,

[00:35:44] Henrik Werdelin: One thing that you also learned when you did that, which I thought was very inspiring was you pointed out that if GPT hasn't understood the essence of your brand, then you haven't communicated very well. And so based on, on that inside, I now use it for a lot of things. I use it for our own brands to see. All this stuff we're putting out is basically the way that GPT kind of computed that. Is that what we wanted? But I also think that you can do that with kind of everything else. Like I have a new book coming out and basically put the whole thing in and saying, basically, what is this book about?

And then, I mean, like directionally, you want it to be what the takeaway should be.

[00:36:25] Noah Brier: It's, and I mean, that's one of the big pieces of advice actually I have for folks especially at brands who are just getting started is like focus on Using this as a reaction machine instead of using it as a generation machine.

I think part of where people get so overwhelmed is like trying to get it to write things, but it's like, it's amazing at that, right? It's amazing as a reader and for anything where what you need is like a fairly regular point of view on the thing. That's, Awesome. Right.

[00:36:53] Jeremy Utley: Well, one thing, one thing I've noticed that you have to do there just for people who actually take that seriously. And I totally agree. You've got to tell it to be like a Russian Olympic coach, right? Be brutally honest, pulled no punches because otherwise it's like you get an 85 like B plus there's a couple of things. Right. But I go, I know this isn't B plus work. I know it's not right. But, and so you actually have to, because it's a helpful assistant, right? You've got to force it to be brutally honest with you.

[00:37:20] Noah Brier: Yeah, , I actually had a pretty funny one. I have a, a Claude project for doing my editing. Um, cause I find it does a better job in editing. So it's got a bunch of examples of my work and some specific instructions and also a copy of strunk and white elements of style cause that's my preferred style guide.

, but the other day I put something in and it basically just insulted me. It came back and said, this is not up to Noah Breyer standards.

[00:37:44] Jeremy Utley: Wow. Yes. Yes. You know, another thing that you just triggered, I don't know why we talked to Steven Johnson, you know, who's one of my heroes, a writer who's been at Google building notebook LM. And when you mentioned loading all the explore, exploit, or tacit knowledge papers and having a conversation, I don't know if you've done this yet, but a super cool thing you can do is load it into notebook LM, and then it will generate a podcast on the fly that is actually, I would suggest is a really cool extrapolation synthesis of all those documents.

And for me, what I like is instead of now sitting, looking at my stinking screen, which I'm so tired of. I can effectively take the synthesis on the road and go for a walk and have a thoughtful, uh, comprehensive view or not comprehensive, but at least a thoughtful overview of the stuff that I load in.

[00:38:34] Noah Brier: Yeah. I have not played too much with Notebook LM only because I have a policy for myself that, um, you should avoid, uh, Google projects that are in labs because they're probably just going to kill it.

[00:38:46] Jeremy Utley: Yeah, yeah, you don't want to fall in love too quickly. This character is going to die. This character

[00:38:50] Noah Brier: Yeah, I know, but it's like, Stephen Johnson is also one of my favorites and I have been very, very inspired by him both personally and From his writing and also , he famously wrote a piece about the way he used Devon think in like 2008 and, um, you know, since then I have been thinking about how I have, been fairly heavy note user and, you know, I guess it's now come to be called personal knowledge management.

But, um, yeah, I need to dig in more. I, uh, I have this bias against, uh,

[00:39:19] Jeremy Utley: No, your bias is founded for sure. I mean, hi, wave at Google wave. Um, but it's also, I would say you are going to fall in love. So that's warning you and you, it will rip your heart out of your chest if they take it away.

But I was in a session yesterday with, , 50 executives. On the fly. We're generating a press release about a new kind of AI initiative for this company. Then we load in that in its quarterly earnings into notebook LM. And while we're talking, we have notebook LM generate a podcast. And then we played it.

Noah, I have never seen people's faces look like that. I've been teaching for 15 years. I've taught. In person, , tens of thousands of people and virtually a lot more. I have literally never seen people's faces like that. You had people, they were holding their faces in their hands because it's just like, there's no way this is happening right now.

It's crazy.

[00:40:13] Henrik Werdelin: So let's go to what CMO is asking you, uh, for right now. And what you are

[00:40:19] Jeremy Utley: finally going back to my canonical question from the beginning by the way Noah, we're lucky to get there because most times we never get to the first question that we started with This is a success.

[00:40:31] Noah Brier: Uh, yeah, what are cmo's asking? So, um, I think obviously a lot of them are just asking How do we get started? Where do we get started? What should we be thinking about? I think increasingly, what really kind of I found most interesting was I was getting questions like, hey, we need to understand.

We want to look at the entire competitive space. We want to understand everything that they're saying. We want to understand who they're talking to, how they're talking to them. And we want to be able to sort of both understand how we rank stack up against that, but also where the white space is. So like who is talking to whichaudience And, how heavily flooded is that particular audience versus another one?

And that was actually one of the moments where I was like, Oh, this is what I want to be doing. It's literally got a call where it was like, Hey, we've been thinking about this for a long time. And uh, we're really interested in figuring out a way to solve it. And Maybe there's some way you could do it.

And by the way, we need it in three weeks for the CEO. And I was like, uh, okay, sure. Yeah, let's try. And, you know, we scraped like 20, 000 articles and classified them and scored them and built these matrices. And, the sort of magic of it is not just that you build the whole thing, but because it's code, you can quickly flip from being backward looking forward looking right where you can move it from just being a thing that does analysis of what happened in the past to now putting in new stuff and analyze what might happen.

I think that general idea, though, like I actually were just about to start a new project with a very large company. Taking a sort of global brand strategy and helping them to localize it, right? Because it's like a classic sort of Marketing problem is we have this strategy it got developed in the u. s Or whatever largest market is and now all these other markets need to execute on it but like they don't have access to that the tacit knowledge of the Individual in the sort of home market who developed it and We don't want to decide for them what they want Some level of autonomy in that local market, but how can we help guide them and sort of like expand that?

And so, we're just starting a project on that. I think that general idea of like, I think my big aha, honestly, over the last six months has been, I was looking for like the software solution the SAS answer, like what is the SAS product to be built in this? And I think I came to the conclusion that actually like this is net new software, right?

This is sort of software or software never existed before. Because it couldn't write. So it's that strategy problem or even that content problem, like that was, uh, you know, I've solved sort of versions of that content problem in the past with a big team and three or four months and, like a lot of budget.

Um, And now, essentially like my big team is, however many tokens I used on, uh, across the models. And

[00:43:16] Henrik Werdelin: just to see that, I understand. I mean, what you're articulating is that this idea of having generic software where we as users have to adapt ourself to the common denominator, because that's how you scale a SAS software.

Now we're building bespoke software as it were per company, per department, per individual.

[00:43:40] Noah Brier: I think, I mean, you know, I'm still thinking through this, but I do think like it's a little back to the future. I think that, what has happened and I think part of why the whole software world is a little bit, unsure of sort of what AI means is because when you've got these amazing models that are available to you for, you know, like fractions of a penny per call, right?

The value of the specific code kind of goes down, right? Like you've, the intelligence is what matters the most and the kind of specific code you write, which used to be the kind of most valuable foundational element of a software product, I think kind of becomes less valuable and it becomes more about sort of how are you incorporating the individual expertise inside that organization whether it's around the strategy or whatever else and sort of applying that because like.

That's where, I, and I think there's, there's quite a bit of evidence that this is true, right? If you look at, if you look at some of the studies on just prompting techniques, and sort of few shot versus chain of thought. There was a paper from Microsoft in December about med prompt and essentially, you know, it was this sort of like very complicated prompting technique that very few people will ever use.

But the takeaway from the paper and there's a great chart is that, Uh, a good enough prompt can outperform a fine tuned model, right? Um, and so to me,

[00:44:58] Jeremy Utley: just to a frontier model, a good enough prompt, a good

[00:45:02] Noah Brier: enough prompt to a frontier model can outperform that same frontier model, but fine tuned, right? And, , my takeaway from that stuff is like the more sort of. You know, specialized you get in the way you build these systems, the more specialized your prompts get, the more specialized you are in the way you build the workflows around them, which is like, just how do you break up the prompts?

How do you break up the problem into pieces? The more I sometimes call them private tokens. So it's like, you know, if you sort of imagine that these things were just trained on all of the public information, naturally they come back with, median answers to questions, right? And so how do you get them to deliver above median answers?

Well, obviously you give it a bunch of private tokens, right? And the private tokens are basically the expertise that lives inside the heads of the individuals inside these companies. And so I don't know. I mean, the challenge going back to your question, Henrik is like, I think, well, I know that building SAS is fundamentally.

Counter to that because the whole thing you have to do when you build sass as somebody who built sass for over a decade is you generalize problems, right? You've got to make it so everybody can use it. And, I just think like now we live in a very different world. And that's without even accounting for the fact that, you know, That we're in this explore exploit thing, right?

Like we are not in exploit yet, right? We are definitely still an explorer. And I think part of it is like all these companies want to come out like Salesforce just had dream force. And Mark Benioff is on stage saying agent force is the answer to every question that you never knew you had. And it's like, I don't, you don't know how that could possibly be true in a world where we don't even know, like we're exploring by definition when you're in an explorer phase, it's like, you don't even know the questions yet.

Right. Right. Um, and so how could they know the answers to the questions we don't know yet? And so, I don't know when I combine those things, I just see that you can build incredible value by, Taking the, implicit knowledge and expertise from individuals and combining it with some expertise about code and software development and AI.

And you can just sort of very quickly roll up kind of amazing solutions to problems that can perform at world class level. So I did another project with a very large brand where we In 10 hours prototyped a copywriting solution. They had already seen multiple versions. And none of them were at the level that they required.

And in 10 hours by working with one of their copywriters we were able to prototype a version that was able to deliver copy at the required level of the organization. Wow. And it's like, but not cause I'm great at writing copy because like, I was working with somebody who's great at writing

[00:47:46] Jeremy Utley: copy with the human, with this asset knowledge.

Okay. So you kind of brought us full circle back to where we started around, explore, exploit. And my question for you is, so you think about the founder of the future or the entrepreneur of the future, you're moving from the exploratory. You deliberately went into an explore mode after exploiting and exiting, et cetera.

Now, as you go back into exploit, I know right at the beginning, one thing you said was, What was attractive to you about going into explore mode is you didn't have to have any employees. You didn't have to have any investors. There's all this stuff that you didn't like, right? How does your working mastery of and with generative AI affect how you're going to move into exploit mode this time around?

[00:48:29] Noah Brier: Um, I'm not sure yet. I mean, I'm just at the beginning of that new journey. so it's probably hard to say where the sort of lines are between My mastery of AI and just my prior experiences and, wanting to correct and sometimes overcorrect for, for different things.

But, , one piece is just like. A lot of the work is clearly going to be outsourced to various systems. I just got this system done this morning for doing the transcript so it, it runs it through the whole process. It kicks it out to Slack. Um, it even has a vector database sitting behind it.

So it can show you related conversations That happened. It sucks out all the questions that got asked in the conversation and that it uses a separate vector database for finding similar questions and then creating canonical answers as new answers come on board. And some of it is just like, we're going to experiment with a lot of that stuff and see sort of how much we can do.

Um, it's all, I think it's cool. I don't know. , I literally just. Um,

[00:49:32] Jeremy Utley: but what about like, are you looking for something different with employees? Are you looking for a different kind of cap table? Like, are there more pragmatic,

[00:49:39] Noah Brier: no investors which I don't know that that has anything to do with generative AI, but it's just I'm happily investor free, but, , our first hire is very senior engineer.

I think it's very different than what I would have approached, Naturally, I think my biggest open question, which kind of relates to the question you had, Henrik, about sort of like the skill set is like, so I decided that the first person I wanted to hire was one of the best engineers I ever worked with who is not deep in AI.

And my thought is he's a much better engineer than I am, and so he can help me on the engineering side and speed up all the ways that I work. And I know a lot more about A. I. than he does, and so I can help him on the A. I. side. We sort of meet at this software development point, which turns out to be kind of hugely important like what I keep finding actually is that.

A big part of being successful in building something with a company is just figuring out the data model, right? This is kind of classic software development one on one. It's kind of like a strategic process. What's the problem? I was working on a proposal generator recently with somebody and, they sent me a whole bunch of their proposals and we were going through it and they were like, oh, we have this part of the proposal on this part of the proposal. And I was like, , well, I don't really get how these are, Different like, and they're like, oh, well sometimes, and it's like, it turned out that they're not really different.

Right. They just call it different things sometimes. And , when you're actually trying to figure out how you're going to put that in a database, like sort of data modeling, one on one is like, okay, well, that's in the same table. And you're going to sort of combine those two ideas. And so, you know, that's something that I think is a place where we kind of meet, but , it's still a fairly Open question to me of who are going to be the best kind of people to work on this?

And , I guess, partially from my prior experience, I'm betting on working with someone who's extremely talented, who I like working with, it seems like a solid place to start and then, you know, a good way to kind of experiment and learn on, on the kinds of skills that it's going to take.

Take to succeed. I certainly think that just we're going to need a lot less people. , I can already tell you that, like, The things that I've been able to do in the last two, I mean, you know, I, I keep telling people, I feel like I have a superpower, right? I think this is what a lot of people are writing code and have gotten into AI feel like , it's kind of the most amazing thing.

Right. , I feel like I just outsource every, you know, I'm like, okay, well, I know how to do this, but there's no,

[00:51:59] Henrik Werdelin: I don't even think you have to write that as good. I mean, I. , as you know, don't write any code or really, but I understand enough now to ask Claude to write me code and then I know how to copy paste it either into cursor or into replit.

And then I understand enough to press the play button and then basically say, I just got this error message. That's

[00:52:20] Jeremy Utley: the green one, right? That's the green.

[00:52:22] Henrik Werdelin: Yeah. And then sometimes , Claude will, will answer like, Oh, you just have to fix like line 58 and go like, no, no, no, no, you just have to print out the whole main.

py again. Yeah.

[00:52:33] Jeremy Utley: What's up with that? Like fixing, like, I'm going to go in and correct one line here. Come on.

[00:52:38] Henrik Werdelin: Like just copy paste the whole thing and assuming that. And so

[00:52:42] Noah Brier: I will say, actually, it's funny you say that. Like recently. I feel like the models have gotten that message and somewhere in the system prompt, they are now sort of giving you the full context all the time.

And I've actually been struggling with some GPDs and quad projects where I'm like, no, I actually just don't want you to keep repeating the entire thing every time. Can you just give me like the changes? Because like, I, like I get it. You've got the whole thing, but like, I just, I wanted to know that you changed three things in 700 words. Just tell me the three. And, uh, I've been like sometimes struggling a little bit too.

[00:53:17] Jeremy Utley: But it's, sometimes it doesn't is the thing. No, I mean, that's the thing.

[00:53:22] Henrik Werdelin: Yeah, that's the worst thing when you're in voice mode, because it's still kind of seem to be prompted in text mode.

. Nora, thank you so much for doing

[00:53:29] Jeremy Utley: that. Drop that mic. . That was very awesome. Thank you.

[00:53:33] Henrik Werdelin: Really, really appreciate as always Nora and excited to hear about the new thing. I guess that's a require a separate call so I can, can get the deeds.

[00:53:41] Noah Brier: Yeah, we should have a, let's have a jam session soon. I got to show you all the weird shit I've been building too. I'm into that. Talk to you soon. Thanks guys.

[00:53:49] Jeremy Utley: Um, wow. Okay. So that was really exceptional. There's a lot there.

[00:53:55] Henrik Werdelin: He's such an interesting guy. You always been such a thing. Um, He used to be kind of like a brand strategist and it's always , so fascinating what he was kind of like. His writing is really interesting. He writes like an amazing blog post also. Um, why don't we just start from the start? What was just some of your big takeaway, Mr. Utley?

[00:54:18] Jeremy Utley: You know, there's a bunch of stuff here I'm just going to rattle through and then you kind of just dive in wherever you're interested. One question that I thought was super cool was, uh, when's the last time you had a cancel all meetings moment. He talked about that, you know, getting access to a new feature. And he said, you know, I've just got to get my hands dirty. I think that exhibits a level of enthusiasm and investment. I think it's a good question for everybody to ask. Another point that I thought was really cool was the idea when I, we were asking about.

Jenny and Russ and he having time to experiment and he just said, don't wait until you have the job. Just explore and become the kind of person who deserves the job. I thought that was really cool. Like not waiting until you have permission, so to speak, I think especially for young folks, there may be a sense that, Oh, it's kind of outside of my scope.

No, not if you bring it inside your scope. I thought that was really cool. And then of course I love his comment about, uh, Gen AI is not just a generation machine. It's actually a really exceptional reaction machine. And I think that's a cool way to, flip the paradigm for some folks when they're dipping their toes into Gen AI.

You know, one thing I've had folks do, for example, is say, think of an idea that you really like, ask Chad GPT to critique your idea. Ask Chad GPT to poke holes in your thinking. And all of a sudden you go, Oh, It's actually really good at understanding. So I think that's a cool kind of, uh, way to turn the tables on AI that, that really helps people see a different side of the capabilities.

What about you? What stood out?

[00:55:51] Henrik Werdelin: I mean, I really liked just this kind of very obvious idea of you can do so much with it now, like we, on this podcast, you know, like newsletters, it's always like, what's going to happen in five years and with the new model and artificial intelligence and all this stuff, but I think the reality is, as Ethan Mollick also said, like, you could just start, um, You know, get going on now.

And I think it's somewhat fascinating that it is really, it is a little bit complicated to get going beyond like the first, beyond the prompt, beyond the kind of the first kind of like initial attempt you have, because it is about trying to understand how do you take your chain of thought and how do you kind of atomize that.

And then start to kind of like figuring out where can I plug AI in that. And so I thought that was interesting. Um, the whole thing about SAS going away, which obviously it's fascinating, , in the venture industry that this has been the hottest kind of like investment opportunity for us many years, because it was such an incredible thing.

Right. You had people often pay a subscription revenue for something that literally didn't cost anything to replicate. It just you make one version of the software and then like you sold too many people. And so the whole idea and I think Shamath have , talked about this too, that any SAS software increasing now will be bespoke software build on top of AI.

In terms of organizations, but also like for individuals. And so maybe that prompted the question, if somebody were to write the perfect software for you, what would that software be? And maybe there's something there. Um, I would say finally this is not something we spoke too much about, but he brought it up and he wrote this piece, uh, I think a year ago where he talked about building the first. Basically plug into chat2BT and he wrote about this thing that's called a manifest, which, and I'll give you a little bit of explanation why I think it's such a big thing. When you have a website, you have something called often a robot. txt file on your website. And what it does is that it tells Google.

Kind of crawler when it comes by your website, what it should expect, what it wants you to think of and all this different stuff. So it's a little bit of like, Hey, Google, here's the cheat sheet for my whole website. And so what the manifest file is, is a file that you can put on your website to basically do the same thing for your website.

Um, when AI start to interface with your website, here is how you should expect data to come from me. And then what chat to PT does is basically say, okay, I got it now. Now just give me data, whatever you in the, whatever form you want. And then, and that started me kind of this whole rabbit hole of. What is the future going to look like when we have agents, like when we get the new iPhone and we start to talk to a Siri that actually understands and are able to do something, what a Siri then talk to, do they just talk to websites?

And obviously there's somebody who builds companies. What am I then expected to have built in the other end? And so, um, what I, and some of my colleagues and specific Nicholas, as you know, I've kind of thought a lot about recently is like, That might mean that we have to almost build the receiving agent on the other side, or we have to think about our companies less.

So as just an app or a website, it's increasingly a piece of intelligence that can talk to the intelligence that we had use of just ask. And so this whole agent to agent conversation for me. Kind of really got started when Noah wrote that, blog post about manifest files. And so, um,

[00:59:28] Jeremy Utley: let's link that in the show notes so that folks can dig into it.

[00:59:31] Henrik Werdelin: That's a great, so

[00:59:32] Jeremy Utley: long way to

[00:59:33] Henrik Werdelin: take away, but, uh, that was the takeaway.

[00:59:36] Jeremy Utley: No, no, the other thing, there's two other things that I thought are just kind of cool shorthand tips for folks. If you're looking for opportunities to get started, one is what's the hated, but helpful. I thought that's a great kind of diagnostic tool.

The thing that is so useful that you want people to do it, even though they hate it. Wow. That's right in the crosshairs of building. And then the other one was around the GPTs for tacit knowledge. And I liked his two by two of impact and complexity. And he said that the upper left is where there's the biggest, Opportunity for GPTs problems where it'd be great if a lot of people knew how to do what only a few people currently know how to do and codifying that knowledge where it's high impact, but low complexity. I thought that that's pretty helpful two by two. If you're thinking about where do I get started building GPTs? So hated, but helpful. The two by two of a high impact, low complexity. You heard it here first folks.

[01:00:37] Henrik Werdelin: And I think with that. That was another episode of Beyond the Prom. We should get one of the Suno to make like a jingle for us.

It was some kind of cool. We

[01:00:48] Jeremy Utley: are going beyond the prom. You should make the

[01:00:52] Henrik Werdelin: AI do that.

[01:00:53] Jeremy Utley: Actually, uh, you know what that reminds you of is there's an old episode of the office where, uh, Jim gets in the car with Andy and, uh, they're sitting there silently for a minute. And Jim goes, you have any music? And Andy goes, I thought you'd never ask. Give me the beat boys. And Jim's like, I meant like a CD or a CD.

[01:01:16] Henrik Werdelin: That's awesome, dude. And with that, that's a wrap. Thank you again for people listening. And as always, we'd love with other people get introduced to this podcast and we'd love to get feedback. So either share it, like it, subscribe, all that stuff. Or if you have any ideas of who we should talk to or things that you'd like us to yeah, like, a subject matter for us to address, then just send us, connect on LinkedIn and, uh, let us know.

[01:01:42] Jeremy Utley: We're eager to hear. Peace.