329 Understanding Your AI Preparedness - Melinda Byerley Jul 22 2026

Greg Bray: [00:00:00] Hello everybody, and welcome to today's episode of the Builder Marketing Podcast. I'm Greg Bray with Blue Tangerine.

Kevin Weitzel: And I'm Kevin Weitzel with OutHouse.

Greg Bray: And we are excited today to be joined by Melinda Byerley. Melinda is the founder and principal advisor at Fiddlehead. Welcome, Melinda. Thanks for joining us today.

Melinda Byerley: I'm glad to be here. Nice to see you guys again.

Greg Bray: Well, Melinda, let's start off by just getting that quick background. Tell us a little bit about yourself and the things you've been working [00:01:00] on.

Melinda Byerley: Sure. So, as I said, I'm the founder and principal advisor at Fiddlehead, and I have spent more than 25 years helping companies understand how data and marketing and performance and governance and executive decisions connect.

I have been in the Silicon Valley for the last 25 years, and my work now over the last three years has been focused on pre-investment work with AI. So, thinking about investment risk, the decision quality with AI, signal integrity. I'm doing things like helping leaders ask what needs to be true before AI can get embedded in the workflows and customer interactions, reporting, and obviously capital decisions.

Kevin Weitzel: That's a lot of groovy stuff and a big old mouthful. So, here's what we're going to do. Before we get started and all that, let's start off with something fun. Let's hear an interesting factoid about yourself that has nothing to do with work, family, or the home building industry at all.

Melinda Byerley: I was one of the first 100,000 people on Earth to go to Antarctica.

Kevin Weitzel: Oh, that is cool.

Greg Bray: Do you know which number in 0 [00:02:00] to 100,000?

Melinda Byerley: No, no. But at the time it was, you know, forgot how many years ago, 2008, they told us, and I looked it up, it was something that, like, less than 1% of humanity had ever done at that time. So, there you go, fun fact.

Kevin Weitzel: A big giant ship and then you get out a little dinghy, and then you go to shore, and you get to frolic about with all the penguins.

Melinda Byerley: So, I made the decision not to actually land. I do care about the environment, and I felt that that peninsula did not need one more person setting foot on it, but we did cruise by, and I did get to smell the penguin dun.

Kevin Weitzel: That I hear is a big problem there.

Melinda Byerley: But it was beautiful.

Kevin Weitzel: Yeah.

Greg Bray: So Melinda, you told us a little bit about Fiddlehead, but give us a little more detail about the kinds of services that you offer to your clients.

Melinda Byerley: So, right now we are doing a variety of things. But the primary way that clients start with us is through what we call a readiness discussion. Sometimes that's a working session because companies don't know what they want to know or what decision they have to be made, and sometimes it's a briefing. So, what do we need to know? We know we have something we want specific information on. [00:03:00] And then usually it's into some form of diagnostic because here's the business problem we want to solve, are our systems ready to do it? Do we have the right systems and processes in place to do it safely, or at least in a manageable way?

Greg Bray: So, AI, it's going bonkers, right? It's like, I don't know.

Melinda Byerley: Have you heard?

Greg Bray: It blew all our minds when generative AI kinda hit the scene. Obviously, there'd been some things before that that didn't get as much attention, but now it's even accelerating even faster, if that's even possible. It seems like every day there's new news, everything else, and I think there's some business leaders who are going, "Oh my gosh, we're being left behind. What do we do? We've got to hurry up, go fast, go fast, go fast." What are you seeing? Like, if they're just diving in, what's the big warning sign you just want to put? If you just had one thing that said, "Before you jump..." What should you do?

Melinda Byerley: Well, the first thing is to think about the price of the tokens. So, one thing that I don't hear very many people talking about is how subsidized [00:04:00] these models have been for so long. For those of us old enough to remember Uber and Lyft, right, there was a point where it was super cheap, and now we have surge pricing, and this is starting to happen.

For example, GitHub has now gone to a metered model for Copilot usage, and developers are reporting that they've run out of their allocation within hours for a month. And so, everybody's kind of frustrated with that, and I think we're reaching a point where if you're doing a lot of usage, so this is more for people that are really sort of investing in a heavy way. If you haven't thought through how to manage your prompts, how well they've been designed, how much token usage you're using, you're going to see your costs go astronomical. That's one piece of it.

The other side, you know, you said one, but there's two pieces of it. Because you might say, "Oh, I don't know much about this, and we don't use it that much, so why should I care?" I would say that yesterday it was announced, actually quietly, that Starbucks has abandoned their sort of internal AI project that they were using to optimize [00:05:00] inventory in the stores because it wasn't accurate.

What we say in the tech industry is that we tend to overestimate something in the short term and underestimate it in the long term. So, I like to reassure people that you will not be part of the permanent underclass if you don't adopt everything AI by next Tuesday. It's coming, and there are things that we can and should do to prepare, but I think that the FOMO, as the kids call it, that's been a little bit overhyped.

Greg Bray: Let's drill down into that cost concept just a little bit more. For those who may not totally grasp this token concept, can you just give us just a little bit more detail for the layperson about when you say how much tokens cost and how prompts connect with that and some of those concepts?

Melinda Byerley: Sure. So, a lot of us who have worked in the software business understand the idea of build once and then sell it many times. We call this scalability. The difference here is that if you look at how the, what they call the frontier models, like OpenAI, Claude, and ChatGPT, those founders [00:06:00] and those managers are talking about AI as a utility. So, this is the difference between buying something once or even buying per seat and paying per volume usage.

Depending on what your favorite sort of capital metaphor is, in business school we talked about whether or not to drill an oil well, was one of the sort of big questions in sort of valuation. How do you decide when it's time to actually drill the well on the land that you own? And you have to think about, well, what will the price of oil be in X months or X years, how much do we think we can get from the well, but we don't know how much.

The more sort of practical application or sort of practical metaphor in the home building industry is an adjustable rate mortgage. When you think about it, for the last couple of years, these tokens have been subsidized to a massive extent. Some people say, we don't know if this is true, but the sort of general rule of thumb is people think that that $100 a month Claude subscription is about $10,000 in compute cost. If you think about that, essentially, if you've bought all this stuff very cheaply and now you're dependent on it and the price suddenly [00:07:00] goes up, you own an adjustable rate mortgage

Greg Bray: It just hit balloon status.

Melinda Byerley: It is starting to balloon. We're starting to see, even the last couple of months, Anthropic has changed their pricing, or people felt like they were hitting their limits if they were really into this, they felt like they were hitting their limits sooner.

Kevin Weitzel: That's almost logical, and it's something that you would think would happen. When Silicon Valley is investing billions and billions of dollars daily into this, there are people that are going to want to see ROI on that investment, and it's not going to be at $10 a month for X, Y, or Z, because that's just the initial investment. That's not including the carry cost of all the computing that takes place, and the heating and the cooling, and the infrastructure and everything that goes into it. They're going to come to the piper for payment at some point.

Melinda Byerley: Well, and I would add to that, they're capacity constrained because of the very logical and understandable concerns about the placement of data centers. Some of the fear is misplaced, some of it is not. There's good and bad in that discussion, and we need to have it as a society. But [00:08:00] from a business standpoint compute capacity is constrained right now. On top of that, we have the launch of Mythos, you know, which is sort of the super AI, if you will.

And setting aside sort of the existential fears about humanity, the bigger issue is the security risk that can come from it. You know, Firefox finding thousands of security holes that didn't otherwise exist, or couldn't have been found by humans, I should say. And so, the frontier models appear to be vectoring towards the support of enterprise and business, because if you're capacity constrained, you want to sell that capacity to the people with the greatest ability and willingness to pay for it. And this is, again, we could go down that rabbit hole and talk about how the models have changed or the frontier models are sort of changing to support that use case.

Greg Bray: So, the short version is it's going to get more expensive barring some dramatic change in the current path that everybody seems to be on.

Melinda Byerley: I'd say that's generally true. It's kind of interesting, there's an offsetting force right now, which is that Google released an academic paper a few months ago. The short [00:09:00] summary is that it actually will allow companies to get more out of the GPUs that they've bought, so that'll be a counterbalancing pressure. But I think in the near term, 12 to 18 months, we'll see prices rise.

Greg Bray: All right, you got to plan on what does it cost? You can't just assume it's 20 bucks a head, kind of a thing per month, isn't probably going to do it. You need to be thinking about how do you then optimize how you use it accordingly because you can influence that cost based on how you use it and how you teach people and the kinds of things you're doing.

I just saw a simple thing about, you know, if I tell it to output something in text versus a Word doc versus a PowerPoint file, it can make a huge difference on just the raw cost of that. Just knowing that has changed some of the way I ask, even if I want a PowerPoint at the end, I wait as long as possible to make sure I've done all the homework and all the revisions and everything before I tell it to do that, as opposed to just iterating within [00:10:00] PowerPoint over and over again. Is it that simple to just train people, or does it get a lot deeper into that, how we use it to be more optimized?

Melinda Byerley: Wow, we can have a long conversation about that. This is my personal opinion. I don't think anybody in this space knows, and I don't trust anybody who say they know. I've met some of the people working on these models, and even they will tell you that nobody knows in some cases. We're all figuring it out together as a species. My personal opinion is that not every person is going to want to be the equivalent of a software developer.

Writers see the value in AI for many reasons, you know, I've spent eight to 10 hours a day for the last three years to get to this place. Companies have jobs to do, and they need people to do those jobs. So, not everyone on the team needs to become an expert at doing this. That's my personal opinion is I think we're moving into a space where let's show people what AI can do for them and help build it for them and inspire them and then let them play on their own so that they're starting to see the value at the office. And then I [00:11:00] think when people start to see the value, the curiosity becomes increased, and there's a desire.

But to your point, yes, how you prompt, how you structure your asks, your ability to create a repeatable system is what drives the cost. It's personally what I've been focused on with Fiddlehead's internal systems, which is building a process and an operating system, not just one-off build this, build this, build this.

Greg Bray: So, when does it move from the one-off into a system? How do you define system in that context?

Melinda Byerley: I bet we've all felt it, because when I talk to people, everyone nods their head. I think we've gone through this journey of, wow, I can make a recipe, to, wow, it can look at my calendar and do meeting notes, to, wait a minute, I have a project for a client and I can put all this stuff together and I can have context.

The next problem becomes, wait a minute, what happened last month? And what happened the month before? And how do I know which version it is? I had a conversation, but where was it? What was the idea? What was the version of [00:12:00] that? Where do I find the file? That's what I mean by operating system, which is I need to be able to know where is the thing that I built. I want to know the most recent version of it. For example, a paid search report for a client, it needs to know how they measure things, because it's not just knowing what Google Analytics can do, it can be how this client defines a conversion versus that client. And so what is the leadership context? Some people like to have the reports one way, and some like to have it another.

I think we've all run into that moment, that sort of barrier of projects where you get to, like, 80% of what you want, and then you go, "Now what?" And I think that's where I've started to work with tools like Codex and Claude Code to essentially, not encode, but what I call documents as code, using GitHub and those tools that developers have, but to think about ideas and processes and systems. But I don't think you have to go there to get value. I still think there's a lot of things you can do with it.

Greg Bray: You know, you triggered something for me, because just the other day I was in ChatGPT going, "I had this conversation a couple weeks ago," and I could not find it. The whole search [00:13:00] chat thing, I'm like, "For as good as you are, it should be a lot easier for you to know what I'm trying to find based on this search string."

Melinda Byerley: That's a very true thing.

Greg Bray: Kevin, have you run into that, where you can't find the one you've done in the past?

Kevin Weitzel: I can't find an email I wrote last week, so trust me, I'm right there. I feel 100% of the pain that you just described.

Melinda Byerley: I'll give you and your listeners two quick tips. One is be disciplined about the name of your chat. While you're in there, name it, and if you need to, put the date or put an order or something, anything. That'll help. The second is to open chats frequently, because as that context from a token optimization perspective, the longer that window gets, the more tokens it uses to give the next answer.

Kevin Weitzel: Hmm.

Melinda Byerley: So, you can do things like create a handoff prompt. At the end of that session, you can say, "Give me a handoff," open the next window, take that with you, and keep going, and that's one way to keep sort of distilling that context down.

Greg Bray: Oh, good insights. Let's pivot a little bit, Melinda, kind of to the executive [00:14:00] leadership concept here about how do we manage this in a company. I think there's been a lot of encourage people to experiment. Maybe they're doing it on their own in free accounts. Maybe they bought their own because nobody told them. Maybe the company's organized something. What are some of the pros and cons to letting people just kind of figure it out versus setting up corporate accounts and some of the things that can go wrong if someone's not paying attention?

Melinda Byerley: I think we've come a long way from last October in terms of people understanding what can go wrong from a governance perspective, in terms of leaking data, in terms of potential for things that can go wrong. And I think any conversation about this should have a discussion about what is the worst thing that can happen, and are we prepared? And I think that's true about all risk in general, right? If we're buying home insurance or if we're buying car insurance, you think about what can happen. Can I self-insure, and if not, what do I do to protect myself? So, I think that it's not just a tool problem. That's a long-standing belief I've [00:15:00] held, that tools do not solve business problems.

The business problem is really a leadership and governance and a decision quality problem. So, it starts in marketing. I have believed this for a while because we're the first to experiment, and we move very, very quickly. But ultimately, the leadership of the company owns the consequences from the marketing team. And so, while it can make everybody faster, it also has the danger if it's not used properly to make a weak signal look bigger than it is. And so we have to be careful because the machines are tuned to please us. We have to be thoughtful about how we interpret that.

So, just to make an example that's I think very specific, you can make this custom GPT that'll make your content consistent, but it doesn't guarantee that your inventory is right, that your incentives are right, that the fair housing considerations have been built in, or that your community claims are current and approved by your brand. And so, I think there's a fine line.

You don't want to slow the teams down unnecessarily, but it's more about being explicit about what is safe for us, and where should the human be involved? What is [00:16:00] non-negotiable? I could go into the risks, but all of those things, you could create an embarrassing situation where the wrong thing happens to the wrong person. The level of that depends on where you are. If you're a gigantic home builder, that could be a national exposure to you. If it's local and it's a small team, maybe the risk factor is low.

Greg Bray: Hey, everybody. This is Greg Bray from Blue Tangerine, and I am so excited to let you know that the registration is now open for the 2026 Builder Marketing Summit. We're gonna be in Dallas, Texas this year on September 23rd and 24th, and we are working on an amazing lineup of marketing, OSC, and leadership content for you.

Please check it out at buildermarketingsummit.com and get your registration in today. Remember, there's limited seats available, so don't miss out. Again, buildermarketingsummit.com. Can't wait to see you there.

There's got to be a lot of folks that aren't really considering the idea that if I don't tell people what to do, they don't always think about all the consequences [00:17:00] of things. You know, I had a experience recently where I downloaded a research report that was actually something I had to pay for, that someone had done. It was really interesting, and as I opened it up, they had this big terms and conditions page on the beginning about not emailing it to all your friends. But now it had the line in there that said, "You may also not upload this in bulk into an AI tool."

And I thought, "Okay, these people understand how to protect their stuff in the new world." Now granted, because you put it in there, I don't know that the AI tool's going to read that and go, "Sorry, rejected," which is probably where it needs to get, but I don't think they're there yet, right? But how often are companies allowing stuff to get uploaded into tools that they aren't really thinking about the implications of what happens to this data if it's out there?

Melinda Byerley: Well, it depends. So, one, if you don't tell people what to do, they're just going to do whatever they want. It's sort of like the culture discussion. If you don't define your company's culture, doesn't mean you don't have a [00:18:00] culture. It will just define itself. So, it is a leadership question. It's where do we want AI to play? It's not, in my opinion, just use it. Because basically we're asking people to light dollar bills on fire now. I mean, I've heard stories of gigantic companies telling their developers that they have a $300 a day Claude requirement. You must spend $300 a day. When has a business ever done that? We're going to measure you on how much money you spend. really?

So, it comes back to, okay, what are we trying to accomplish? What is the outcome we seek? What risk are we willing to tolerate to get there? Sometimes we are willing to make that risk in order to get the reward. The good news is that by all accounts, when one is on the appropriate plan, for example, if one is on an Enterprise or Teams plan, by default that data is not shared with the model. Even in the $20 a month plans, you can opt in. You have to go in and say, "I don't want my stuff to be shared," and that happens.

The other thing I would mention is that on these frontier [00:19:00] models, it is unlikely that a single piece of data will be there because your data is being pumped in along with everybody else's. So, you know, one should not be putting in HIPAA data. One should not be putting in FERPA data. One should not be putting in proprietary methodologies. But I think, the cat's out of the bag. I'm certainly being more thoughtful about what I put out there in terms of what we do and how we do it until I can understand more about how it may or may not be used by the agents or by other companies.

Kevin Weitzel: Okay, so while I'm paying to have my information be private and not shared out to the infrastructure of the world of AI, what happens when I stop paying?

Melinda Byerley: Great question.

Kevin Weitzel: Is that in perpetuity, or is that only while I'm not paying?

Melinda Byerley: I don't know the answer to that question, Kevin. I'd have to follow up with you. I don't know. But it's a really good question. I do think it's worth thinking about. By the way, as a side note, you can extract your content from the models right now, and I encourage people to do that. Both ChatGPT and Anthropic, [00:20:00] Claude allow you to export your chat history. Do this on a regular basis. It's a good idea. There's lots of stuff you can do with it later we could get into. It's just good hygiene to keep building your house on your own land.

And I do think, as a side note, this is going to go local over time. Maybe not at enterprise scale, but certainly I think the open source models are almost there. In a year or two, I think the future will be local and distributed, and that will help everybody because we'll be able to benefit from AI without everything being out there for everybody to see.

Kevin Weitzel: So, there's the layperson's belief that, oh, AI's going to take over my job. And then there's the group of marketing and AI savvy people that say, "No, it's just, your job may be taken by somebody that is leveraging AI." However, it's the very billionaires that invested all this money into this stuff that will practically and absolutely benefit from the elimination of some of the costs of the human factor. So, my long-worded question is, what kind of [00:21:00] reassurance do we have outside of self-preservation? So, let me give you a better example.

An airplane is flown better, more accurately, more succinctly by a computer than any human could ever match. A human pilot cannot match the ability for a computer to control the accuracy and trajectory of an airplane, and the landing and the taking off of said airplane. So, it is the pilots' union that keeps any of that from creeping into taking their job. They'll go on strike while they try to implement any of this stuff. So, what's to keep that from happening in any other industry that don't have self-preservation? Because I can't tell you a home builder, they wouldn't love to, "Man, I can just replace all my marketing people with this AI that's just going to do everything for me."

Melinda Byerley: We're all savvy businesspeople. Beware the easy answer. If it sounds too good to be true, it probably is. Look at what I said earlier about Starbucks quietly disabling their chatbot. A few months ago, Amazon announced a 90-day pause on AI [00:22:00] development. To be very, very blunt, which is my Midwestern way, I would ask anybody if you think your developers are better than Amazon's. Chances are for 99% of us, our developers are not better than those at Amazon. So, if Amazon's saying, "Wait a minute, something's off here," then I think that should give the rest of us pause. Again, it's coming.

There is a point where the more you work with this, one, you sometimes feel like you're in a bit of a rabbit hole, so you have to reach out and make sure we're calibrating here. But those of us who are spending a lot of time hands-on who are not developers, I don't stand to gain anything from these companies going public. I am somebody who is, like the rest of you, trying to find my way for my career. But having lived here in the Valley for so long, the more I use it, the less scared I am.

Now, I have concerns. I think a really important question is how do we help our young people understand what good is? What is quality marketing? And deciding what the thing is that will be built and what the quality will [00:23:00] be is the very human part of this. That is something that's hard to develop, and that critical thinking that says like, "No, this isn't good enough," or, "Wait a minute, this doesn't sound right," or, "I don't like this," or, "I want it to look like that," that is the part that can and should be remaining human.

I got involved three years ago when one of my esteemed colleagues in the Valley's very big investor said, "We need people that are on the humans' first side." This is my impassioned plea to anyone in marketing to really come to the table. You can put your head in the sand and say, "It's not coming," but humanity needs you. There's so many developers involved in the conversation, and they're coming to me going, "We don't know how to talk about the safety issues. We want help from marketers to help talk to our congressmen and talk to other people." So, we could have this conversation in six months and it might be very different, Kevin. It is literally a work in process. In my 25 years at this, I have never seen this.

I will say to you, though, that the last time I felt this way about a technology was when the iPhone was invented. My husband [00:24:00] will tell you that I dragged him into the Apple Store and I said, "This is gonna change everything." I couldn't tell you how. I couldn't have predicted that Uber and Lyft and Facebook and all the things that came out of it, but I knew it was going to change, and this is that. We are standing on the precipice of something that is going to change us. It's going to change us as a society.

I know that I wanted to be part of that for good. I felt in order to do that, I wanted to know as much about it as I possibly could. Not everybody's going to have the time or space to do that. I think that's the part where people have to understand that just assigning your team 20 more hours a week to learn AI while expecting them to do their job is unrealistic and unfair. There is a point as I work with this stuff that I start to realize, wait a minute, at some point it'll just be cheaper to hire a human to do this right now because it requires so much maintenance, so much governance, so much checking, so much risk that it's not worth it. And this is why I think you're starting to see the beginnings of that from some of these major corporations.

Greg Bray: I've seen the situation [00:25:00] already in our own processes where we'll get into something and it's like, wow, it's helping me, it's helping me, it's helping me, and then we hit that wall of it's not quite working right, it's not quite working right, and all of a sudden, nobody knows why because we don't understand how it got from where we started to where we are now, and where things went sideways.

Either we have to start over and do it ourselves because the thing's due, like, now, and we can't wait to figure this out, or we're completely dependent on the tool to fix it. And so we start now in this, like, iterative process of like, well, no, check this, check that, check this. It's like almost there but not quite. And that's where I feel like, okay, somebody needed to check it a little bit earlier to make sure we weren't totally off, or to understand how we got to step three and step four before we're now at step nine and there's a problem.

Melinda Byerley: Right. It's like thinking about where the wrong answer is expensive. The wrong answer is probably not abstract, right? It could be the wrong [00:26:00] incentive, the wrong availability, the wrong reference, the wrong claim. Like, what is it we're willing to put up with? I like to say that if those of you Star Trek fans, that all of us want tea, Earl Grey, hot. That is what we all think AI is.

Captain says, "Tea, Earl Grey, hot." But right now, you have to go in, and you have to tell the machine what is tea. What is a cup? It should be served in a hot cup. It should be served at this temperature. How long do you steep it? You have to tell it that it's not Coca-Cola, and it's not coffee, and we use milk, but not at the same time as lemon, et cetera. And if you don't do any of those things, you're going to get a hamburger.

And so, if you think about it, some people's brains are going to be really good at this. It turns out, I think in a lot of ways, and this is my sort of my personal opinion, is a lot of us that are neurodiverse are actually enjoying this a great deal because we're for whatever reason hardwired to think about things this way. Why is it like this? If your kids were like me and asked why a lot, welcome to AI, you're going to love it. Because there's a lot of like why does this work and how does this work and [00:27:00] thinking about that, and not everybody wants to think that way. And there are people who are wonderful at their jobs, at their core job, that shouldn't be forced to.

I have a friend who's a wonderful writer and editor, very senior in experience. I don't want her spending her day thinking like this. When I give output, I want her to say, "This is good or not good." That is what her skillset is. So, I think it's really thinking about what you're good at and what you want to do. And when you start to understand is the more you work with it, you start to see, wait a minute, it's not everything. It's not tea, Earl Grey, hot. We are a long way away from that. In some ways, it gives you more comfort. And I think that's what I would say to young people, too, is it's okay to be scared. It is okay to be concerned about the future of humanity and yet we need to get involved because those of us who care, if we don't get involved, then the people who don't care will run the show.

Greg Bray: That was a mic drop moment there, Melinda. That was awesome. So, what does get involved look like then, if we just give that just one more minute of thought?

Melinda Byerley: If you are one of those people that's still holding out, you are right to [00:28:00] be concerned. I am not here to minimize you, and I'm not here to say everything's going to be rosy and perfect. At the same time, I think it's harder to have a voice if you don't know what's going on. So, start the journey, and you are not too late. I asked this question three years ago. I said, "Aren't I too late?" So everybody that's involved feels that way. That's the first thing. Everybody feels they're behind. Nobody thinks they know anything. Just start.

Use it for recipes. Use it for your workout plan. Just start to see what's going on, and each step you take, you will become more informed. You will ask better questions of your leadership, of your team, of your colleagues, of the world, of your representatives in Congress. I think the more informed we are as a species about what this is, the better position we'll be to influence its future. I got involved in technology because I wanted to better humanity. I believe that's what science and technology should exist for. That's something I care about.

Kevin Weitzel: I use ChatLZT, which is a very unknown platform for a workout [00:29:00] plan, and it just told me just to loaf around the house in a recliner.

Melinda Byerley: Well, that's where the critical thinking comes in, Kevin.

Kevin Weitzel: Oh, okay.

Greg Bray: Well, Melinda, for those marketing leaders today that are like, "Okay, my team's playing with it. I need to go to the next step," what would you recommend as kind of the next thing to look at when getting beyond just the individuals using it to get their individual tasks done? How do we move forward or to that process state that you were talking about a few minutes ago?

Melinda Byerley: I think it's about establishing what we call, it's a fancy word, governance, but what it really means is a plan and a process. What is our business problem? What are we trying to accomplish? How will we get there? How will we know when we get there? What risks are we willing to take? That's going to involve some more change because you're going to have some people who are like, "Ah, I want to use this tool. I just want to use that tool." There has to be some coordination now, because there's a benefit.

If this person on your marketing team has found a great way to maintain consistency [00:30:00] or has built a skill in Claude that everybody else can use, you want to leverage that, and you also want to make that part of the intellectual property that your company owns. So, work with your attorney. That would be considered a work product. If someone is using your Claude instance to make a skill on behalf of the company, that is IP that should stay with you when they go.

That doesn't mean they won't take it anyway, sometimes they will, but at least they're not selling it or they shouldn't be selling it. It's something that you can retain and have future employees use. That's just one example. You want that knowledge to start being shared. We're seeing it happening informally, just like we're doing here. You have small pods of people collecting and being like, "What are you doing and how are you using it?" And I think that's where it can grow.

I would also say, backing all the way out of this, I strongly believe in what I call the narrow slice, which is find the smallest atomic unit you can, the most unsexy, not tea, Earl Grey, hot, but can you consistently edit something according to our brand? It seems like it should be the easiest thing to do, and I think those [00:31:00] of us who have started doing it realize it's harder than you think. And if you get that supposedly simple and the most deterministic, if you can, thing right, you learn so much about it that the next step becomes obvious, and then the next and so on.

So, it's almost like scale the vision back from this, you know, AI is going to do all the things, to what can AI do this week, and how can we do that, and how can we make the next step and the next step? It's perhaps less sexy, as it were, but I think it's a more sustainable path.

Greg Bray: Well, Melinda, this has been a great conversation. I wish we could keep going, the clock is telling us we need to pause here. But any last words or thoughts of advice that you want to leave before we finish up?

Melinda Byerley: I would say if you're using it on your own, start to think about your operating system, as it were. Start to think about building something methodical and repeatable. And if you're not sure how to do that, you can ask. You can ask Claude to tell you how to take a project and move it into Claude code. You can ask GPT and so on. And when you start to think that [00:32:00] way or to ask it how to ask, if that makes any sense, "Claude, I'm thinking about doing, how do I do it? Now give me a prompt that I can use to start," and then go to the next chat. That was the big aha for me in the last few months, was instead of asking it to do something, is to ask it, "Help me," how to ask it.

Greg Bray: Great extra layer. I agree. I think it makes a big difference on how you do that. Well, thank you so much, Melinda for being with us. If somebody wants to connect with you, what's the best way for them to reach out and get in touch?

Melinda Byerley: I'm on LinkedIn, of course, and you can always find me at fiddleheadhq.com. There's a button there if you're ready for an AI readiness session, you can reach out.

Greg Bray: Well, thank you everybody for listening today to the Builder Marketing Podcast. I'm Greg Bray with Blue Tangerine.

Kevin Weitzel: And I'm Kevin Weitzel with OutHouse. Thank you [00:33:00]