Latest / Serious People Podcast / How to Build a Business with a Team of Agents | Scott Patterson (Core Fiber Founder & CEO)
Transcript
- [00:00:00] Scott Patterson: When I pulled up my daily briefing on Monday, it already was giving me, saying, "Hey, like, this is the perceived impact of the price change so far on, you know, on your D2C customers. Like we're seeing an increase in conversion," or whatever that might be. [00:00:13] Noah Levin: Wait, so you didn't, you didn't ask it to give you a second analysis? [00:00:15] Scott Patterson: No, I didn't, I didn't. I didn't- honestly, I, I didn't even think about it at the time. I said, "I, we, I wanna make this price change and get it done." And it came back to me without my input to say, "Hey, I'm assessing the performance here. I'm not gonna come back with like a firm, you know, call until, you know, a week or two weeks of data, but, you know, here's the early read." [00:00:30] Noah Levin: Welcome to Serious People. I'm Noah Levin. Today, I'm talking with Scott Patterson. Scott is a principal here at Serious People, and he's also the founder of Core Fiber, a business he built as a solopreneur and is running today with a ton of help from, you guessed it, his agents. This is a really practical episode full of things you can try today. We get into Scott's daily briefing, how he manages his company context, and how he used an agent to do a project he'd been putting off because it touched so many different parts of the business. I think you'll be leaving this episode motivated to try something new. Here's my conversation with Scott. Welcome to Serious People. [00:01:10] Scott Patterson: Excited to be here. [00:01:11] Noah Levin: I don't have to welcome you 'cause Scott, you are a principal at Serious People and also a close, close friend and a great guy and the founder and CEO of Core Fiber, which is a lot of what we're gonna talk about today. Scott, you have a background in CPG. You worked on Pet Honesty as a brand. I think you were the number three employee there. What led you to wanna start your own CPG, and then why Core Fiber? [00:01:34] Scott Patterson: Yeah, so my experience in Pet Honesty taught me a, a few different things, and one of the big ones was that when you build a brand that solves a real problem, that can be in a space that consumers really love to interact with, you know, in the case of pets, people love their pets, you can build something that's lasting. Uh, and so I knew I wanted to solve a real health problem. Self-proclaimed biohacker, so figured it was probably gonna be in the human, uh, supplement space, and spent a lot of time sort of figuring out where in the human space there was a big gap. Looked to the fiber category as, okay, it's this legacy category dominated by these large brands that have been around a long time. 95% of Americans are fiber deficient, so there's a major problem to solve. Why hasn't anyone really solved it yet? So looked at that and said, "Okay, there's a massive opportunity here to bring a modern consumer brand into this space with a better product." The writing was on the wall for the category, and, uh, yeah, haven't looked back since. [00:02:21] Noah Levin: Bring us up to speed on what is Core Fiber, the product, what's the business looking like right now, and, and tell us a little bit about the team you have around you. [00:02:29] Scott Patterson: So Core Fiber is a prebiotic fiber blend. Unlike most fiber supplements out there, which just use either psyllium husk or one other sort of staple fiber, we use five different types of fibers. And so that's sort of what makes us different. We are exclusively e-commerce today, so we sell on Amazon, we sell on TikTok Shop, and then we sell on our own website. And so in the... You know, when you're scaling an e-commerce brand, there is ton you can do in terms of your organic presence, your paid performance. There's ton on the back end, making sure you have plenty of inventory in stock, making sure that you have the right amount of forecasting so you're capital efficient. You're not just, you know, burning and lighting cash on fire. As everybody knows, the ability of AI to do all of those different things in different ways has just grown exponentially in the last six months alone. And so when we set out to build this business, and I say we, sort of, I, I do have a co-founder in this business, we knew that we wanted it to be a large company with a very small footprint in terms of headcount because of the advances in AI. We do spend a lot of time thinking about where we can get leverage across the business. So you know, I have agents for pretty much all of those functions that I, I mentioned before and then some, whether that's, you know, supply chain and operations, whether that's product development and R&D, across our various sort of marketplace or e-commerce functions, really sort of have a, have a, an agent across every facet of the business. [00:03:42] Noah Levin: I'm really excited to go into, like, the specific use cases that you have brought in to show us, because I think seeing how agents work in a very, like, real environment with actual business units flowing through the machine is a lot more useful than seeing sort of abstract case studies put on by tech people for tech people. We're gonna get into that in a second. Before we go into how you're actually running Core Fiber behind the scenes, let's just, like, paint the picture of who you are as a guy and your life and, and, and what your portfolio of time looks like right now, starting with the fact that you have- a wife and three kids all under the age of three? [00:04:15] Scott Patterson: Yes. [00:04:16] Noah Levin: So what time are you allocating to this business during the day, and what else do you have going on priorities-wise? [00:04:21] Scott Patterson: Yeah, I mean, that was a, a big reason for wanting to sort of lean into AI even more is as someone who does have three under three, is it a bit crazy? And you, you never know what your schedule's gonna be like. And so in addition to wanting to get leverage and, and have more time back in my day, I also needed something that could provide me a level of consistency and sort of maintenance in the business when I might not be able to be as consistent with my time in the business as, as I would like. [00:04:42] Noah Levin: The picture I wanna make sure people get of you in your life is that you're a very well-rounded guy. You get in a workout. You've got a family that you love very much and spend a lot of time with. You told me once that you couldn't meet for breakfast on a Friday because every Friday you have breakfast time with one of your three kids individually, which I think is just the best thing in the world. These agents are unlocking something very human for you, which I think is fantastic. So let's go back to Core Fiber now as a business. As a business, what are the goals that you need to hit that you think about every day or that you look at on a week-over-week snapshot, where as you design your agents, you're, like, trying to get progress toward this metric or this goal? [00:05:19] Scott Patterson: So I think if, if I was gonna summarize it into sort of, like, three high-level goals, it's obviously growth, which is, comes in the form of revenue growth, profitability. I'll also lump sort of cash flow in the profitability piece or, or minimizing burn, and then brand awareness, right? And so if I think about how we grow this business, it's growing our brand and awareness to ultimately grow our top-line revenue and doing so profitably. And if we can do that in a way that's also cash flow positive, like, we could build this thing forever. [00:05:43] Noah Levin: Let's start talking about your control center. When you log in in the morning, what are you actually looking at all day? [00:05:49] Scott Patterson: If I look back over the evolution of the last few months, I am spending more and more time on a daily basis inside of, you know, for me it's ChatGPT or Codex. And so I am trying to keep as much of my work there as possible. And so I would say I spend probably 50% of my day at this point in time in the ChatGPT interface. [00:06:06] Noah Levin: It's not always obvious what it means to, like, spend your day inside of ChatGPT, especially if you're used to using it as sort of like a Google Plus- [00:06:14] Scott Patterson: Yeah. [00:06:14] Noah Levin: ... uh, answering questions for you. [00:06:15] Scott Patterson: Yeah. [00:06:15] Noah Levin: What are the types of things that you're working on with ChatGPT? [00:06:19] Scott Patterson: Yeah. So the way I think about it is I've segmented my workflow out into a few different work streams. So Core Fiber is obviously the, the largest work stream for me. That's where I spend most of my time during the week. But then there are also other projects I, I work on as well or other work streams. So Serious People is a great example. You and I have been working together to, you know, deliver AI solutions for clients. That would be its own work stream. And I have a personal work stream where I, you know, have various, uh, projects I'm working on for myself, whether it's myself or my family, what have you. So that's how I think about it is like, okay, overall, where do I spend my time in the day? And then there's one that sits on top of that, and these are the sort of like primary, uh, sessions, if you will, or agents that drive the day-to-day. And so you can see up here I have, you know, an executive assistant one, I have a chief of staff one. I have a, an ongoing sort of data brain research project I'm doing. Those then inform all of these other work streams and sort of guide them. [00:07:06] Noah Levin: So I think what's not maybe obvious to people who have not been in these tools for a while or have not been in them at all is that chats or sessions on the left-hand side can live forever, right? Like the chief of staff doesn't have to be rebooted at any point. It's just a persistent chief of staff that you interact with every day. [00:07:20] Scott Patterson: Exactly. [00:07:21] Noah Levin: The other, uh, thing, and well, I think we'll probably look at this in a second, is that they're not MECE, right? They're not mutually exclusive. And that's okay because these sessions are actually talking to each other, right? [00:07:30] Scott Patterson: Yeah. Exactly. And some of these sessions, for example, the chief of staff, as I'm working through things in that session, I will either have it directly reference these other sessions and go update them and make sure it's sort of like managing them, or it will do it on its own because it knows sort of what persists in each one. [00:07:45] Noah Levin: So you're in Codex all day, and you're running these persistent tasks, doing the things that matter in your work streams or on top of them through chief of staff. How does the context for the work get stored? Where's the memory of like, what are you working on? What's the status today? What's the performance metrics of your business? Where does all that live? [00:08:03] Scott Patterson: For me, the company brain, so to speak, lives in Notion. So that's where the context for everything is. Those are the SOPs. That's where all the skills are. And I should be very clear, like I'm not going into Notion and maintaining it. Like this is all getting maintained via ChatGPT. And then separately, I have a, think about it as almost like a project management system. I use Obsidian, but it's just a, it's a, you know, app version of what you could get with ClickUp or Monday or any of these PM tools that has the actual tasks that need to be done on a daily or weekly basis. Again, similar to Notion, I almost never open Obsidian. It's not meant to be sort of a, an interface that I'm actually looking at. It's all something that ChatGPT, or in this case Codex, is using and referencing. And so it's determining, "Okay, hey, what do we need to work on? What are our due dates? What's the context for this task? What's the last thing that you did on this? What did you say you were gonna do on this? What are other action items that need to be populated, or what are dependencies?" So that, it, it essentially all happens in the background on those two platforms. [00:08:56] Noah Levin: Do you mind pulling up Notion so we can look at it while we talk about it? Do you wanna give us like a tour of the types of things that are stored in Notion, and again, stored by your agent, not by you? [00:09:03] Scott Patterson: Yeah, exactly. So at a, at a high level here, the things that sort of I'm, I'm storing in Notion here, and again, like I mentioned this before, I've got this split up into what I call my work stream. So we're looking at Core Fiber right now, but this sort of permeates in those other work streams I mentioned. So in here, you know, it's got information about what is the source of truth for various things. Google Sheets is where I have my live systems, my sort of quote-unquote "data warehouse," if you will. Mission Control, that process through, through Obsidian that I mentioned. That is sort of where tasks live, their status, their priorities, priorities, et cetera. And then within here, you know, you can see I have things about the brand voice of Core Fiber, what is our company strategy, you know, our, our go-to-market strategy. I've got a whole glossary of people I interact with, decisions that get made. So if I've created a coupon on Amazon or if I've changed the price somewhere or if there's a new product I wanna work on, like all of that- all those decisions get permeated within Notion. And then I've also mentioned before, I have SOPs. So these SOPs are really skills. These are ChatGPT skills. So this is everything that I am using, or I should say ChatGPT is using to manage all the projects, work streams, everything that, that I'm doing on a daily basis. [00:10:08] Noah Levin: This episode is sponsored by Valet, the founding sponsor of Serious People. One thing that comes up in this conversation is how much useful work you can get out of AI once it understands your business. Sometimes the result is a dashboard, an analysis, or a little internal tool. Then you need a way for your team to actually use it. Valet lets you publish AI-created artifacts to a stable URL. New sites are private by default, and people in your organization can access them after signing in. You can try it at valet.dev. Thanks to Valet for supporting the show. Time and time again, you see that if you don't give the agent a source of truth to play with and to keep in the background for all of its tasks, it will lose focus, it will not maintain context correctly, and sometimes it will reinvent the wheel again and again. Okay, so you've got Notion as your sort of second brain of all of your, uh, context. You said you're using Obsidian as almost like a project management tool the way you'd use ClickUp. [00:11:03] Scott Patterson: Yeah. So this is a, you know, this is a, a free version of a project management system, if you will. This was something that Codex built on its own to manage the various tasks that I, you know, are inevitably coming up on a daily basis. So for instance, if I'm working on a project, let's say on Amazon, I need to update my Amazon listing, and maybe that entails updating the main images, and maybe that entails updating pricing or bullet points or what have you. In addition to updating the sort of Amazon listing as the parent, there are a bunch of smaller tasks that have to happen there. And so when I sit down and... Or when, when I sort of sit down with Codex and I know that's something I'm going to work on, it's already got a pre-populated list of tasks that it's extracted either from conversations I've had directly with Codex, from calls I've had maybe with my Amazon agency, from emails or Slack messages with the agency or my graphic designers or what have you. So all of those tasks get accumulated in here via daily sweeps that happen several times a day to pull all that context in. And those tasks then become the sort of foundation for either how I spend my day or how Codex is spending its time. It knows that, hey, if there's a task due on this date, like, and it's got a sort of an approval to already work on that task, like it's working against it. [00:12:08] Noah Levin: So we talked about your context in Notion. We talked about your context in Obsidian. There's one other piece of context that we haven't looked at yet that I think you smiled when we were talking about it earlier. You said your data warehouse, but then you looked a little bit sheepish about it. Can we look at your data warehouse? [00:12:23] Scott Patterson: Yeah, absolutely. I refer to it as a data warehouse, as a non-technical user. So essentially what I have is a, I call it a two-step process. It's probably a lot more steps than that, but it's a, it's a two-sheet process, if you will. What I'm showing right now is this is the, the sort of raw and normalized data that I have. This is where, uh, Codex is pulling from, you know, whether it's Shopify or Amazon or Meta Ads or Google Ads or, you know, in any of our sort of source data platforms. It's pulling all that information in here, and it's dropping it into this worksheet, and there's, you know, tabs for sort of every single different function. In fact, there's usually multiple tabs for every sort of source data that I have. And then where this is actually going is it then gets compiled into sort of what I call the, the Scott-facing. I'm calling it my operational dashboard. This is where I actually look at things on a trend basis, whether it's a daily basis or weekly basis. So I can see everything, whether it's my full business P&L. I can see individual channels. And so this is where I look at the individual metrics. Um, now keep in mind, there are probably two or three tabs in this sheet that I'm actually looking at on a weekly basis myself. Maybe it's the, our company P&L, our weekly P&L, monthly P&L, LTV models or things like that. But the reality is most of this is actually getting pulled in from Codex. Like I mentioned, I'm in Codex most of my time. And so in, let's say in the morning, I get a briefing that's put in front of me from Codex from my chief of staff agent, and it's summarizing what's happening into this sheet. It is summarizing the things that I care most about. And it knows what I care most about based on training and, and what I've been telling it over time. Let's say for the last month or so, we've been doing a lot of really, you know, big growth efforts on Amazon. It's gonna focus a lot of that conversation around Amazon. So I'll get the full financial picture of the business, and then it'll focus on, hey, this past week on Amazon, we saw that traffic was actually up 30% due to, you know, these various campaigns that we were running or we increased ad spend by 50% or what have you. And so it'll give me the sort of highlights. But all of that information, whether it's financial data, whether it's other metric data, traffic I mentioned, maybe it's, you know, conversions, units, whatever, those are all getting pulled in from here. [00:14:21] Noah Levin: So I think what's interesting about the setup and super smart about the way it's working is that you're doing some of it via a proactive push into the data warehouse, and then some of it via a reactive pull out of the data warehouse. So rather than saying every day at 8:00 AM when your chief of staff wakes up to do its job, it should go all the way back to Meta and Google and Shopify and start pulling data directly from those APIs. You have that sort of pre-cached in your data warehouse. So really, it's like all teed up and ready for an agent to do math on. [00:14:53] Scott Patterson: Exactly. I mean, if I, if I was doing, you know, 10 times the, the volume I was doing and I was in a lot more channels, like Google Sheets would probably not scale to that point, and that, that, you know, eventually I, I'm aware that I will need to move to a different tool and system. But it totally works for what I need now. [00:15:07] Noah Levin: So you just showed your data warehouse in Google Sheets. You and I are working on a client engagement right now where it's an agency client, so they have multiple brands. Not quite the same as your business, but, you know, pattern matches. And they're paying us money to build them an actual data warehouse in, you know, something like BigQuery with data engineering and pipelines and whatnot. Why is it worth it to them to have a real data warehouse where you're okay not having one of those? [00:15:30] Scott Patterson: Yeah. So I'm a solopreneur, right? So this is my business. You know, the, the buck stops with me. And so my level of risk tolerance for this business is reasonably so, probably gonna be a lot higher than what an agency's risk tolerance is for its clients. And so they cannot be as cavalier with all of their client information and client data as, as I can, 'cause they're, they're, at the end of the day, like, it's not their data. And so when you have an agency, or, and this is, this is also true for a large organization that has a bunch of team members and, you know, a, a lot more data and infrastructure, you start to have sort of more risk controls that need to be in place. And so when you have that, you need a much more thoughtful and sort of adult data structure than what I'm operating with. [00:16:10] Noah Levin: Spot on. I, I think that there's such a big difference. There's almost an uncanny valley between building these agents for yourself and building them for an enterprise. Right? And there's lots of stops along the way on that scale. Data is one where there often is a right answer to a question that's asked, and getting an agent to get that right answer 100% of the time is very, very hard. So we've talked through your context layer, we've talked through the, uh, the ChatGPT harness that you use day in, day out to do your work. Why don't we look at some of your use cases? [00:16:40] Scott Patterson: So this is my chief of staff, and I have this as a, uh, sort of a demo environment, which is why you see this up here, demo. So essentially what happens on a daily basis, I get a daily brief put in front of me within this chief of staff session. So what it essentially looks like is if, if I were to prompt it and say, you know, "What's my daily brief?" or daily brief or what have you, you know, it's gonna spend time pulling together all of this contextual information around not just the performance data that we were showing from the data warehouse, but the things, the decisions that exist in Notion, um, which comes from context it pulled from email and it pulled from Slack, and it pulled from my call notes. And so it's giving me an update on, okay, what are the things that we need to be focused on today across these various work streams? So in this case, you know, Core Fiber is what you're seeing here. And then what are the things to really sort of focus on in terms of next steps? [00:17:26] Noah Levin: Could you give an example of what a recent recommended action was and how you then take that and start executing on it? Do you, like, keep chatting with it in the session, or do you take it somewhere else? [00:17:35] Scott Patterson: One example of recommended next action that I had recently is we're going through the process of a formula update, and so there's, there's a lot of sort of downstream impacts to when you update your formula. It affects, you know, your price and- [00:17:45] Noah Levin: By formula, you're not talking about an Excel formula. You're talking about the- [00:17:47] Scott Patterson: Actual- [00:17:48] Noah Levin: The ingredients in the- [00:17:49] Scott Patterson: Actual Core Fiber ingredients formula. Uh, yes. We have an ongoing back-and-forth conversation with our manufacturer, R&D team, our account managers, things like that. And it was sort of... The, the ball was in my court in terms of getting back to them with updates from, we have a scientific advisory board that we run sort of formula changes by, and the, the ball has sort of been sitting in, in my court to follow up with them and really sort of take that to the next step. And the reason that that might not sound that pressing, but the reason that it's important is because we're trying to time this formula update before we place another order for our products, and we need to make sure we place an order soon so we don't run out of stock in our sales channels. And so that's something where, you know, it will recommend it as a next action, and then it will continue to push it up as a priority the longer I sort of let it sit. [00:18:31] Noah Levin: Okay, so you've got your chief of staff brief and you've looked through it for the day. [00:18:34] Scott Patterson: Yeah. [00:18:34] Noah Levin: Maybe now's a good time to talk about the pricing example that you mentioned. You had this, uh, task that you were putting off for a while. You wanna tell us about it? [00:18:41] Scott Patterson: Yeah. So let's say, let's say I sat down, you know, it's, it's 8:30 in the morning, I've run my daily brief. One of the items in there, it says, "Hey, something we've been talking about for a while is this pricing update. Like, we need to make progress on it. It's currently on hold." So previously, if I needed to make a pricing update, and, and this is actually what happened here, is I would, you know, I m- would probably drag my feet because there's so much that it's involved. You know, we have to go in and, and not just make changes on our website and go into Shopify and make the change. Like, that sounds pretty easy. But there's image assets that need to be updated across Shopify. There's, you know, images in, in our emails. There are implications for our subscription customers. And so historically, there's a laundry list of items that have to be done in order to make a price change. And what I finally sat down and, and did is, you know, I said like, "Okay, I'm gonna sit down and I'm going to put this into a Codex session, and I'm just gonna have it put together a plan for making a price change on our website, and let's see what it does." And so I sat down and, you know, I, I- gave it a talk track, which is why this, this message is, is fairly long. [00:19:41] Noah Levin: When you say gave it a talk track, you mean you, you dictated? [00:19:43] Scott Patterson: Yeah, I dictated to it. So I s- I sat down for a few minutes and I dictated to Codex and said, "Hey, I wanna make this price change on our website. I know there's all these downstream implications." Some of them I, I pointed out, like, "Hey, subscription prices and things like that." But what I want you, Codex, to do is come up with the full list of what are all the implications for making this price change, what can you actually do yourself versus what needs input from me or, or, or a human, and then, like, build, build that plan. That was sort of what I asked it. And what it came back with is it ultimately said, "Okay, hey, this is a coordinated effort. This is the current price. This is what you wanna change it to. These are the various required updates that need to happen across all these different systems, across emails, across our product display pages, across our subscription. Where's that language showing up on the website? And then what are some decisions that need to be made before we say go?" And so it, it gave me that as an output, and then I said, "Okay, great," like, "Here's my answer to those decisions. Now go and actually do this." And within the span of, I don't actually know how, how long it took, uh, f- f- four minutes or so, um, it made the actual updates. On our website, for example, this is our, our homepage. This button which says save 30%, this percentage off was a different number. This button which comes from our email provider, like this 30% off here, this was a different number. It adjusted all of the areas that it needs to change across the website, including not just the price that I told it. This was sort of the anchor price that I told it, "Hey, we need to go from $49.99 to $39.99." It updated that. It updated the subscription pricing. It updated the buttons on the website. It updated, uh, this sort of main image right here. There was a banner that sort of was highlighting a 50% off sale. It removed the banner. And so it made all of these various changes across the site, made the changes where it could to the emails, everywhere where there was a sort of touch point. And then it also spit out a list for me of, "Hey, these are all the things that I couldn't change," maybe graphics within emails or things like that that need a human. And so it then took those other items and created a task list for me in Mission Control, "Hey, these are the things that you need to update to really take this price change from 85% complete to 100% completion." [00:21:44] Noah Levin: I think this is such a great example because it's an orchestration of lots of really nitpicky tasks that are easy to get wrong or execute poorly or miss, but across a strategic-level decision. The changes have to be coordinated. They have to execute at the same time and properly. I w- I just love that you delegated this strategic project to it, and it delegated back to you sort of low-level task work, turned a high-judgment thing into a low-judgment thing for you. [00:22:11] Scott Patterson: And the other benefit of this is if, if I compare this to, let's say I had a junior employee and I said, "Hey, you're in charge of the price update. Like, go do it." Aside from the fact that obviously this is a lot faster, the other thing that it's doing that something somebody like that might not do is now, you know, when I, when I pulled up my daily briefing, we made this pricing change last week. When I pulled up my daily briefing on Monday, it already was giving me, saying, "Hey, like, this is the perceived impact of the price change so far on, you know, on your D2C customers. Like, we're seeing an increase in conversion," or whatever that might be. [00:22:41] Noah Levin: So you didn't, you didn't ask it to give you those analysis. [00:22:42] Scott Patterson: No, I didn't. I didn't, I didn't... Honestly, I, I didn't even think about it at the time. I said, "I w- I wanna make this price change and get it done," and it came back to me without my input to say, "Hey, I'm assessing the performance here. I'm not gonna come back with, like, a firm, you know, call until, you know, a week or two weeks of data, but, you know, here's the early read." [00:22:58] Noah Levin: That's amazing. [00:22:58] Scott Patterson: That was, that was pretty cool. [00:22:59] Noah Levin: I love that. [00:23:00] Scott Patterson: Yeah. [00:23:01] Noah Levin: So talked about the pricing change. You had this example that we were talking through earlier that I thought would be fun to, I'm not sure if it's a good visual, but we can talk through it, about this formulation change you were looking at, and you were basically using your agents as, like a, like a PhD-level R&D team. [00:23:17] Scott Patterson: Yeah. [00:23:18] Noah Levin: What was the situation that you were actually working through with them, and how were you using them? [00:23:22] Scott Patterson: One of the things that was important to me when, when I built Core Fiber is, you know, I, I always wanna be evolving our product formula to be better and better, and we base that off of customer feedback. We base that off of what we see out in the market from other product that are doing well and what customers are saying about other products. And so when you have an ever-evolving product formula, whether you're changing the dosage of ingredients, you're changing the ingredients themselves, in theory, you could be changing the various marketing claims you're making. Change to the product formula affects everything from price to how you can position your product to the availability of ingredients and therefore sort of what are your lead times for being able to get those products or, or to get those ingredients and make your product. And so there's all these downstream things that can be affected either positively or negatively. And without AI supporting, there are a lot of different stakeholders that need to be involved in these conversations, both within your business and without your business. The thing that I was sort of really trying to optimize for the most was, A, I wanna have a product that works. B, it needs to be within a price range that makes sense for us. And C, it needs to afford us to make the, the health benefit claims about our product in a way that is accurate. What is now our R&D agent became this sort of all-encompassing tool that allowed us to collate various research, uh, out there about various ingredients that we were considering. It allowed us to sort of compare ingredient A versus ingredient B versus C in terms of the health benefits we could claim and the dosages needed, and how that would impact the fiber that we are able to claim on our product, and do that while also having a conversation with our, our manufacturer R&D team, with our scientific advisory board. And so it was able to sort of keep the wheels on the bus, uh, as the bus is, you know, barreling downhill towards this big change. And ultimately, the, the outcome was we were able to make a product formula update a lot faster, ended up being less expensive for us in the long run, and made, made the product better. [00:25:10] Noah Levin: You basically treated this agent session as your R&D department and gave them a strategic goal with a hypothesis to go and chase. [00:25:18] Scott Patterson: Exactly. Exactly. [00:25:19] Noah Levin: So I think it's obviously a lot of people listening are not gonna be in the nutritional fiber supplement business, but might have some R&D equivalent in their organization. And so I think to just generalize this, you started a new session, but in Codex, which is already operating on top of all this context about your business, and you told it what? [00:25:39] Scott Patterson: I told it that I wanna make a formula change. I gave it the guardrails. So for me, the guardrails were whatever the final formula we come up with, it has to equate to a six grams of dietary fiber per serving. And I told it that, like, I don't want a fluctuation in our, our cost more than 10% up or down. Well, actually just up. Down is good. Uh, so I, I gave it those guardrails and, and the, I guess the other guardrail was I told it like, "These are the marketing claims we want to be able to make about our product." So when it's going checking research, it knows like, "Okay, what can I substantiate and what can't I substantiate?" So tho- those were the guardrails I gave it, and then it went, did a deep research project, came back to us, and, and the, the outcome I told it I wanted was three different options to choose from, three different formula options that I could then take back to our R&D team and our advisory board. [00:26:32] Noah Levin: The prompt that you were describing, I think what's interesting about it is that it's, it's, it's a very pre-computed prompt, right? Like, it took years of you working in CPG and developing expertise around supplements and formulations to know how to prompt it in that way. [00:26:45] Scott Patterson: Yeah. [00:26:45] Noah Levin: When people are thinking about how to leverage agents in combination with their expertise, a lot of times that manifests as just a really good prompt. [00:26:54] Scott Patterson: Yeah. [00:26:54] Noah Levin: Right? [00:26:54] Scott Patterson: Yeah. [00:26:54] Noah Levin: All right. So we've talked about some of your, like, core demand side, like pricing changes. We've talked about the R&D department. Let's talk about back office for a second. [00:27:02] Scott Patterson: Yeah. [00:27:02] Noah Levin: So how are you using these agents to administer the business, to be your G&A, basically? [00:27:07] Scott Patterson: I've got a very busy life as, as most of us do. And so one of the things that I knew was gonna be really important was this business needs to have, like, a stable cash flow, right? I'm not building a business that I'm trying to go out and do a Series A and Series B and whatnot. Like, th- things could change, but right now I need to have a stable cash flow that I can run this business in perpetuity. And so one of the most basic things I needed was visibility into our burn rate and what was, you know, bringing that up or down and where there was opportunity to sort of be more efficient. And so, you know, I, I work with s- I still do work with a bookkeeper, um, who helps with a lot of sort of like, sort of day-to-day finance, SaaS paying invoices, you know, accounts, accounts payable, et cetera. But I've started, you know, and, and a lot of this was from our conversations together. I, you know, I have a CFO agent, if you will, that is starting to, or that, that has been keeping sort of better visibility into the, the finances of the business, because if you're looking at just a P&L, like that doesn't really give you deep insight into your burn and sort of the capital efficiency of it. And so I have an agent that is now my sort of CFO partner in this that is doing a lot of that, and it's made my bookkeeper more efficient. It's definitely made me more efficient. It's made me aware of where we were either wasting money or where there's opportunity to sort of pour more cash into things. Um, and so that has been a, a huge stress reliever for me, uh, and, and frankly a big time saver too. [00:28:23] Noah Levin: We were talking about my CFO agent, I think a few weeks ago. I realized that I had all these atomic tasks, like one that was chasing invoices, one that was paying bills, and that I kinda wanted someone to oversee all that. I think my prompt was something along the lines of like, "Be my CFO. Do things that a CFO would do, like be very smart." And I, I think I actually said, "Be very smart. And by the way, take ownership of these other sessions that are running." And I dialed up the model and I walked away, and I came back a few hours later and it had hired, uh- No, it had created sessions under it that were alongside my now named accounts payable and accounts receivable sessions. I, I now had a controller. [00:29:06] Scott Patterson: Yeah. [00:29:06] Noah Levin: Seems reasonable. [00:29:07] Scott Patterson: Yeah. [00:29:08] Noah Levin: And I had a f- a minimum cash flow that I needed to maintain- [00:29:10] Scott Patterson: Yeah [00:29:11] Noah Levin: ... and it had an opinion on where we were relative to that. I don't spend a lot of time interacting with my CFO, but I do get my chief of staff prompting me saying, "Hey, your CFO is worried about your margin." Uh- "It th- it, it thinks you underpriced that product." [00:29:22] Scott Patterson: Yeah. [00:29:23] Noah Levin: And it's kind of running the department outside of my line of sight. I can go and probe, uh, at any level anytime I want, and I do periodically. But it's performing functions that are useful but not directly observable, which is sort of a hard thing to wrap my brain around a little bit. [00:29:41] Scott Patterson: Yeah. [00:29:42] Noah Levin: Talk about your, uh, your accountant and how your accounting processes, like your month-end close works. [00:29:47] Scott Patterson: We do a month-end close, um, basically by the sort of second, second week of the following month. And historically it's, you know, it's been me and, and my bookkeeper sitting down and, and sort of taking a look at everything, figuring out where things look off or what expenses are missing or, you know, what have you. And we, we st- I want to be clear, we still do that. Uh, but the change has now been the inputs that go into informing that month-end close, whether it's evaluating transactions in QuickBooks, which is sort of our, our financial record. We use Ramp as well for paying vendors and things like that. It's doing most of the manual interaction, and it's logging those things along the way. And then the other benefit is it's cross-referencing against the, that operational dashboard, my sort of data warehouse, if you will, to make sure that what I am tracking and using to report on real-time performance is in line with the actual transactions that are happening. [00:30:36] Noah Levin: So we have one more, I think, use case we wanted to talk about, which is new product launches. Talk us through how you're doing new product launches now. [00:30:43] Scott Patterson: Yeah. So, you know, again, for us, we're a CPG business. It's a physical new product. I think this would apply to a lot of other businesses, be it SaaS companies or what have you. And so historically, launching a new product in a CPG has been some, some level of art and science. You're looking at market data. You're also sort of figuring out, like, where do I think consumers would be interested in this, or where do I think the existing product suite doesn't really sort of fit what consumers are looking for? And, you know, it, it would be a very, very lengthy process. Now, at least for me, that process is almost exclusively done via AI. So if I have an idea for a new product I want to launch, or if I know I want to launch a new product, uh, that I want to be complementary to our existing set, like that will be another sort of sit-down dictation prompt. Maybe spend a few minutes doing that and then gets handled sort of end to end from not just figuring out, okay, where's the white space? How would you price this? How would you position this? How does this sort of fit in with your existing product suite? Or, you know, does it require sort of a, a, a net new sort of way of thinking about your, your brand to, okay, does your existing manufacturing set have the capability of making this? If not, here are the manufacturers that can make it. Here's what it would look at from a gross margin standpoint based on, you know, the assumed cost of this product. And so it is a really great starting point for figuring out, okay, what is viable and what is not viable for me. And it saves so much time now. [00:31:56] Noah Levin: Is it doing things that you couldn't do or things that you just don't have time to do? [00:32:01] Scott Patterson: It's, I would say it's, it's the latter. Um, there's probably not things in there that I couldn't do. But what I would say is, because again, it has this context of my business and it, it sort of knows what to look for, it is bringing to light things that I might not have thought about, at least immediately in the product process and, or product launch process and development process. And it might be bringing them, you know, I might have brought them to light later on only to then change my strategy or thought process around what is actually viable. And now it's doing it all upfront in the planning. [00:32:29] Noah Levin: We've looked at a lot of use cases. If I'm watching this podcast, I have an actual operating business. It's not a fiber supplement business, although it might match to some parts of your business. What's your advice for someone who's either not using any of these tools, or probably the most common for people listening is gonna be a casual user or maybe intermediate user of either Codex or Claude. Like, what should they do today to just get moving in this direction? [00:32:53] Scott Patterson: First and foremost, start with a real problem, right? Like, the reality is, if you're trying to build a business, there's a problem that is preventing you from doing it or there's something that's slowing you down. So start with a real problem. And then the other component of that is start, right? Like, I think you said this earlier, it can be way too easy to get in your own head and think, "Okay, well, I, I don't know how to prompt this correctly," or, "I don't know how to tee this up." And the reality is, like, the, the technology has just gotten so, so good, and it'll get better as you go. So it's about getting at-bats and just starting to make incremental progress on, on a thing. Like, if you go from zero to one, like, that's still progress, right? You don't have to go from zero to 100. And so that's what I would say is figure out whatever a problem is that's in the way, and just sit down and start dictating to it and see what you can accomplish in five minutes. [00:33:34] Noah Levin: My experience is that when you bring in a new problem and you say, "I'm committed to taking this thing that I would normally do by clicking around in five web browser windows and writing a document in Google Docs and sending an email," that doing it all through Codex or through Claude is a bit of a schlep. [00:33:50] Scott Patterson: Yeah. [00:33:50] Noah Levin: Um, 'cause you're sort of brute forcing each step, right? You're saying, "Let me teach you how to get into my email." [00:33:54] Scott Patterson: Mm-hmm. [00:33:55] Noah Levin: "No, use this account, not that account." [00:33:57] Scott Patterson: Yeah. [00:33:57] Noah Levin: "Write in this voice and not that voice." [00:33:58] Scott Patterson: Yeah. [00:33:59] Noah Levin: Um, and here's why. It's a lot like training a new employee. [00:34:02] Scott Patterson: Right. [00:34:03] Noah Levin: And I think that the subjective experience I've had is that when you're just starting out, it doesn't feel like it's worth it because you're not getting any leverage. In fact, you might be getting negative leverage by doing it the hard way this first time. And so really the, the value comes when you start putting workflows through it that you know pretty soon you're gonna have to do a second time, because then that second time it all clicks. [00:34:23] Scott Patterson: Exactly. You have been a big proponent of this for me as well, is, you know, if, if you're unsure where to start, sit down and do a screen recording of the thing that you... I- if it... There's a thing that you're doing that you don't wanna do anymore or you wanna do it better, just sit down, record a screen of you doing it, and explain what you're doing, why you're doing it, and then just share it and say, "Hey, help me do this better. Help me make this process better." Again, incremental improvements are still improvements. [00:34:43] Noah Levin: And you can just use your computer for a week and then go into Codex and say, "What can you automate for me?" And it'll say, "Well, you've spent most of your week on TikToks and maybe stop with that." Um, the other thing I think that might feel like a little bit unapproachable if you're just starting out is you've got this, like, really robust context layer, and you've got this sort of operating system going on of a chief of staff and the different tasks. How do you start that from scratch? Like, how, how do you step into it? [00:35:10] Scott Patterson: The way that it got built is I just sat down and again, at the beginning, and that's probably the way a lot of these things get built, is just sitting down and, and doing a long dictation. I think it might have been a 30-minute dictation or 20 minutes, and I just sat down and said, "Hey, if you're truly starting from square one, this is who I am. This is my business. This is what I'm doing. This is what I want to accomplish. I'm trying to build this, you know, workflow or this operating system," or, you know, it doesn't, doesn't need to be technical language, but I'm trying to build a system that makes me more efficient, that gives me time back in my day. This is where I spend my time today. Like, help me figure out where to start. I want to build a plan that solves this. And so you're, you're essentially just i- in some ways griping to it in, in a lot of ways and pointing out, "These are my pain points, these are my problems." And if there's anything any of us can speak to, it's the things that are, are frustrating us. And then just hit enter and see where it goes, right? And it'll come back with a plan. It might not be the perfect plan, and it doesn't need to be the perfect plan, but it's something to get you started and to get you going. [00:36:03] Noah Levin: I think we should end on, like, a, a cautionary note, which is that you and I, I think, are, like, wildly pilled on this stuff and are getting, going deeper and deeper in, right? And there, there comes a point where you are doing productivity not for the sake of the business or the sake of work-life balance, but for productivity's own sake. [00:36:20] Scott Patterson: Yeah. [00:36:21] Noah Levin: For you and I, who also consult on AI for a living, we can kind of pretend like it's work- [00:36:26] Scott Patterson: Yeah. [00:36:26] Noah Levin: ... just to, like, futz around with our own stuff. But there is such a thing as rebuilding your to-do list too many times. [00:36:33] Scott Patterson: Yeah. [00:36:33] Noah Levin: Or optimizing your optimizing function, and there's, it just turtles all the way down. I think that the, uh, the, like, friends don't let friends aspect of this episode is that if you get so excited about this stuff that you find yourself going deeper and deeper and spending your weekends away from your loved ones just adding new recurring scripts to your Mac mini, stop. [00:36:54] Scott Patterson: Yeah. [00:36:54] Noah Levin: And touch grass. [00:36:55] Scott Patterson: Yeah. [00:36:55] Noah Levin: And talk to your kids again. And really being less cheeky about it, like, stay rooted in the actual business that you're trying to run. [00:37:01] Scott Patterson: Exactly. [00:37:02] Noah Levin: I think that's one of the things that's nice about physical goods and human services is that at the end of the day, it resolves to, like, stuff and people and conversations. And so you kind of know if it helps the business if you ship more units. [00:37:14] Scott Patterson: Exactly. Having been in roles where I actually had people that I- that worked for me, I would sort of say, say the same thing that I used to say when, you know, when I had a team, which was, you know, always have a specific concrete goal in mind and a, a quantifiable goal, whether that's widgets you're trying to ship, whether that's impressions, whether that's followers, app downloads, what- whatever your thing is, like, have a number in mind, and then at the end of a week or day or whatever your time function is, ask, "What progress did I make on that actual goal?" And you should be able to have said something to that effect. [00:37:43] Noah Levin: Yeah. Well, I think one of our metrics, both of us, is hours spent with kids. Uh, and so I think- [00:37:49] Scott Patterson: Time back, yeah. [00:37:50] Noah Levin: Yeah. So that's a good place to end, and let's go get some time back. [00:37:53] Scott Patterson: Let's do it. [00:37:53] Noah Levin: Scott, thank you for coming on the Serious People podcast. [00:37:55] Scott Patterson: Thanks for having me. This was fun. [00:37:57] Noah Levin: Thanks for listening to Serious People. If you want more conversations like this in your inbox every week, subscribe to Serious People on Substack at seriouspeopleai.substack.com. You'll get the episodes and practical notes in your inbox along with something to read and something to try. And if you're trying to make AI useful inside your own business, Serious People can help you find the right work, build the system, and put it into production with your team. Find us at seriouspeople.ai.