AI Can Handle More PPC Account Work. But Who's Responsible for the Decisions It Makes?

Episode Description

Fred Vallaeys, CEO & Co-Founder of Optmyzr, and Aaron Levy, Optmyzr’s evangelist, are joined by Kirk Williams, owner of ZATO, a PPC agency that works mostly with eight-figure e-commerce brands. Kirk is one of the show’s most frequent guests and the author of Stop the Scale: Building a Digital Agency You Actually Like.

The question behind this episode is a practical one. AI can now take on more of the work inside a PPC account, from analysis to recommendations to parts of campaign management. When it makes the wrong call, the agency is still the one answering to the client.

Kirk, Fred, and Aaron work through what that means day to day: client emails written by AI, the cost of tokens, whether faster work is better work, and who is accountable when an agent makes a change.

Here’s what was discussed in this chat:

• How DTC brand operators are building their own AI systems, and what that changes for agencies
• A real example of an AI-written client email that read as rude, and how a quick call fixed it
• Why Fred tells Optmyzr staff to know who their audience is before sending an AI-written message
• Why brand and sounding like yourself matter when everyone has access to the same AI
• Why Kirk stays close to new AI releases without chasing every one of them
• Why speed is the wrong way to measure AI, and what to measure instead
• How a change-request workflow gives AI-made PPC changes a human owner and an audit trail
• Who is liable when an agent makes a mistake, and why Kirk is cautious about rolling out agents
• When to build your own PPC software and when to buy it from people who know the work

Why DTC Brands Are Building Their Own AI Systems

Kirk starts with something he’s seeing in his own client list. Plenty of DTC brand operators are building with AI, and a lot of it goes well past dashboards.

Some are building infrastructure for the business itself, like forecasting stock. Kirk said it sometimes feels more advanced than what most marketers are doing with agentic media buying.

Fred’s first thought was that this should be good news for agencies. If a brand has solid data and forecasts, the team can aim at what the brand actually needs, instead of working from a ROAS someone picked at random based on what the last agency did.

Kirk’s worry is the flip side. A forecast built with an AI’s help can be wrong, and a client who trusts it completely is a different kind of client to work with. Where AI has clearly helped his own team is the technical digging.

“It’s just a lot easier for us to hop into, especially some of the non-Shopify-connected platforms, where we can hop into BigCommerce, where it’s super clunky, it’s annoying. We can actually hop in and start to get data. Here’s actual item sales, map that to Merchant Center item IDs, map that to ads revenue, try to identify what’s actually driving sales, and yada yada. That used to be way more complex, if even possible, without SQL knowledge and all this stuff, right?” Kirk said.

That, in his view, is AI doing exactly what it should. Where it gets messy is in the relationship between the agency and the brand.

When AI Writes the Client Email

One email is the story Kirk keeps coming back to. The client is, in his words, “such a great, great person,” so a message that came across as terse, almost rude, and accusational didn’t match the person at all.

“And we received kind of one of these emails of, here’s what I need from you guys. And it was terse. It was almost rude. It was accusational. It was all this stuff. And so I just was like, oh, hey, do you mind if we hop on a call? And we hopped on the call, and I was just like, you know, this seems to be what it’s saying. Is this what it’s saying? He’s like, oh, no. You know, we got to a place immediately. I think what happened is he didn’t even read that email. But that impacted us,” Kirk said.

Kirk’s read is that people who spend all day inside AI can start leaning on it without noticing. AI tends to push as much work as possible onto the recipient, and he described getting a list of 25 tasks when a normal person would have seen that ten of them were their own to do.

Fred’s rule for Optmyzr staff is to know who your audience is. A long answer full of reasoning is fine to forward to a colleague who has an agent reading it, but a customer shouldn’t get that data dump.

Aaron named a problem most people have felt but can’t quite explain. Tonal shifts are one of the most frustrating things, he said: two weeks of warm emails, then one that is clearly from a bot.

“And the LLM writes like a computer thinks, not like how a human writes. So there will always be stranger sentence structures. There will be a setup, a pitch, a knockdown. They always will follow the same sentence structure if someone’s using something basic. And that’s just not how the human brain works,” Aaron said.

Why Brand and Voice Matter When Everyone Has the Same AI

Aaron asked whether brand matters more than ever, and Fred answered yes and no. No, because LLMs mirror whatever you tell them and will happily call any brand perfect.

Yes, because being identifiable and standing out matters more when anyone can ask an AI to write in anyone’s tone. Fred’s bar is that your AI should at least learn your tone and your style.

He described a tool called PeakReach that Optmyzr built to help staff share the company’s blog content. Most people don’t repost because they don’t have time, so the tool drafts options for them to pick from.

“But what we did is we had the AI monitor how people would take those little blurbs and tweak them before they posted them. And within a couple of weeks, like that closed-loop methodology, people actually started saying, like, oh yeah, it’s writing really well, it’s starting to sound like me,” Fred said.

Kirk uses AI in a similar back and forth. It helps most when a thread gets long, like one with 60 emails across an analytics team, an attribution team, and the client.

He asks it to catch him up and draft a reply, then keeps going until what goes out is still him.

Where AI Still Falls Short Today

Kirk has little patience for how some people talk about AI online. When the conversation turns to what AI can’t do, he hears the blame land on the user.

“Sometimes I think full-on AI proponents who just can never seem to have any criticisms for AI, there’s almost this level of it’s just endlessly the user’s fault. It’s kind of like, well, AI just needs the information, or it needs the prompting, or it needs you to train it better, or whatever,” Kirk said.

His point isn’t that AI is useless. Some limits may be natural, and an agency still has to run its business with today’s tools while building toward what comes next.

Fred’s version of the limit is cost. He expects AI to handle the tone problem eventually, but the back and forth needed to train it burns tokens, and many people are on plans with a weekly allotment.

“I’m actually scared now, like, what happens if on Wednesday my tokens run out? What am I going to do on Thursday, Friday? I literally have been working so deeply into AI that starting a PowerPoint on my own, it’s like, well, I know how to do it, but really…” Fred said.

So he’s now doing the math on whether something is worth perfecting or good enough, so the tokens are there for the next job.

How Much an Agency Should Experiment With AI

Aaron’s worry is speed. Something new gets invented every day, and he used ChatGPT ads as the example, saying they have to be a nightmare to manage because the product changes every week.

He framed the challenge as two questions: how do you catch all of it, and how do you make sure it doesn’t change tomorrow? Kirk’s answer is a philosophy more than a process.

“I always try to stay up with what’s changing and also somewhat stay on the edge of innovation, while also lagging behind enough where I’m not just, exactly what you said, constantly ignoring what I need to do now for the latest and greatest thing,” Kirk said.

He also pointed to research from RAND that he plans to use in his Hero Conf talk. RAND’s report says that by some estimates more than 80 percent of AI projects fail, twice the rate of failure for information technology projects that do not involve AI.

For Kirk, the number is a reminder, not a reason to stop building.

“And that’s okay, right? Tokens are cheap, there’s kind of this excitement, we’re all building, but there is some reality in this transitional phase that, I think the danger is that you neglect the real elements of business and what needs to make your business successful and what are we doing,” Kirk said.

Judging AI by Business Results, Not Speed

Fred picked up that thread and made it blunt. People are losing jobs over this, he said, and the reason is how they describe what AI did for them.

“A lot of people are losing their jobs because they’re not able to explain to their boss how they are using AI to benefit the business, right? And it’s so much about, well, we wrote 10 blog posts instead of one because we used AI, or we manage an account 10 times faster, right? But nobody cares about faster. Everybody cares about what’s the financial output, what’s it do to the actual business,” Fred said.

Optmyzr has started telling its staff to be goal oriented instead of task oriented. Work has historically been task driven, he explained, and now that AI can put the widgets together, people should focus on the goal.

That’s also where he sees AI struggle. It’s good at processing findings for a data study, but he finds it misses the flow that helps a person follow one point into the next.

Kirk agreed that marketers also have to allow some R&D, then raised a risk he thinks people underrate. Search engines could start penalizing AI content that isn’t close to human quality.

“Well, what happens when there are ways of identifying that more and more easily, and all of a sudden all of these sites get completely tanked in rankings, you know, six, nine months down the road, and someone had really invested in building all that out, right?” Kirk said.

Fred’s take was that this makes marketing more important, not less. It has become easy to build things, but it’s still hard to get a new product into the market.

What Humans Should Still Own

Fred asked what Kirk hopes AI will do for him, since his book argues against scaling for the sake of scaling. Kirk clarified that Stop the Scale isn’t an argument against growth or against any particular size.

His concern is aggressive scaling as if growth is the end goal. If he can keep delivering quality service and keep his health, he’ll hire more people.

“Increasingly, for me, ZATO is going to be humans talking to other humans with the aid of fantastic tools. I consistently see AI as a really, really fantastic, incredible tool, but I still do see it as a tool,” Kirk said.

He has even pondered a Stop the Scale type MCP, for consultants who want to do well with AI and still live their lives. The theme he keeps landing on is that AI supports him and his employees, and doesn’t replace them.

Fred pushed on that, noting that people like Jensen Huang and Sam Altman say we’re already in the age of AGI. Kirk’s reply was that these are people who are financially motivated to say it.

Fred granted the motive but said Altman himself treats AGI as a marketing term while still believing we’re there. Kirk’s deeper objection is about how humans decide things.

“So much of how we make decisions is based on things like pain, fear, hope, you know, all that sort of thing, right? And so, again, just because you can have a self-learning cognitive thing, I still don’t think that’s going to still be quite the exact same, which to me would be a replacement, until it also is able to experience things that influence how we think, right?” Kirk said.

Fred landed in a similar place. He doesn’t think a machine will ever truly be like a human, though it will be a good representation, and he isn’t sure it needs to be.

Can You Trust AI With PPC Decisions?

Kirk’s reason for not fully trusting Claude with account decisions is simple. There’s nobody to hold responsible.

“Part of the reason I cannot trust Claude is because I cannot hold Claude in any way responsible, and that does impact our decision. So, of course, Claude is going to endlessly tell me how confident it is of something until I push back the third time, and then it’s like, my bad, I was completely wrong, right?” Kirk said.

He contrasted that with a human employee, who knows the success of the job, and of their family, depends on the quality of the work. For him, as an owner, that difference changes the decision.

Fred’s answer is to copy a system software teams already use. No coder at a major company can publish their own code. There’s a change request, someone else reviews it, and it’s checked against other changes before it goes live.

Optmyzr applies the same idea to PPC changes made through its MCP or an agent, so every change goes in for human verification.

“What’s nice is it was a human-in-the-loop mechanism, but as a nice side benefit, it provides the audit trail, because now it’s like, okay, well, it was Jimmy who looked at it, and the AI explained to Jimmy that these were the reasons it said we should raise the bid. Jimmy looked at it, Jimmy maybe pushed back, and then eventually Jimmy agreed to it. So there’s a human in the loop, there’s a human accountable, and there’s an audit trail,” Fred said.

The audit trail pays off when a client asks, six months later, why so many bid changes were made. Without structure, Fred said, finding the real story in your chat history is a matter of good luck.

Who Is Responsible When AI Gets It Wrong?

Aaron turned the conversation to who builds these systems and who gets held accountable. He’s heard a few approaches from agencies, and not all of them end well.

“I think we’ve all heard the horror stories of the agencies that do a 20% layoff and then say, okay, cool, go find AI savings. We’ve heard stories of people finding AI savings and then growing headcount, growing client count,” Aaron said.

He also described a holding company that put a whole team in charge of building something, then let them go and handed it to contractors. His question was who maintains it after that.

Kirk said his team is busy doing what it normally does, so he hasn’t pushed a build-everything approach. The project he’s most focused on is a ZATO brain on the admin side, with SOPs, client information, and communication, to keep an eye on goals.

What holds him back is risk, and he was candid about it.

“And on the other side, what I just described, there’s actual risk. There’s actual liability. If I push out an agent that does something stupid, Anthropic is not going to be sued by my client, to just speak very transparently. I am, right?” Kirk said.

Fred pointed out how that differs from Google’s own automations, where a mistake can come with a refund.

“If you use Google’s own automations and they mess up, you can go to Google and it was money into their pocket. So they will generally make good and refund whatever clicks shouldn’t have happened,” Fred said.

Kirk said his team had just received a $13,000 credit from Google for one of their clients. Fred’s reply: Anthropic won’t do that.

Build Your Own PPC Software or Buy It From People Who Know the Work

Aaron brought his own cautionary tale. He has tried to vibe code, and it went poorly, because he doesn’t know where buttons should go or which framework to use in the back end.

Not everyone is a coder, he said, and not having to write code anymore doesn’t remove the need to know how software should be built.

Kirk said this is part of why he likes to pause and monitor. As things mature, solutions get made, and he mentioned people like Alfred Simon and Mike Rhodes, who are building structures and databases led by people who know what they’re doing. In his view, you can buy that and build on top of a significant foundation.

Fred’s case for buying is that an account has many layers, and the AI has to be handed the right thing to look at. He used an example from home: three school emails about the same event, from the principal, the teacher, and the PE teacher, with one saying 8:30 and another saying 9:00 a.m.

“Your AI will generally stop at the first thing it finds and not see there’s a conflict and figure it out. And that’s sort of the same problem you deal with in PPC, right? There’s so many layers to the onion that you have to make sure you structure it and you give the AI the exact right thing to look at. What is the actual state of the account in this moment? And then the AI can do its creative thing and can go to town,” Fred said.

Fred also made the case that the UI is the easy part. Aaron’s button can go wherever Aaron needs it, as long as the database underneath is sound and isn’t missing anything.

Why Kirk Still Trusts Experts Over AI Alone

Kirk closed with a small, familiar example. Twice recently he has received clearly AI-generated schedules where the email said Monday and the actual invite was for Tuesday.

It sounds minor, but with a calendar the cost is missing the event entirely, which defeats the reason for using AI. That leads to the argument he hears most: that humans make mistakes too.

“There’s just some level where it’s actually not that difficult to fire that employee and find someone who, for the most part, you actually can trust. And again, whenever someone’s like, well, humans make mistakes too, it’s just like, they just don’t, though, not at this level, and then not where you have zero recourse to change that, because you can’t just tell Claude, stop making these mistakes, right?” Kirk said.

His conclusion is that the human versus AI conversation isn’t fully presented, which is why he still calls himself pro-human when it comes to actual professionals and experts. You can find Kirk at zatomarketing.com, where he goes by PPC Kirk.

Episode Transcript

Kirk Williams: Things like accountability and liability, there’s no way that Claude is going to… Part of the reason I cannot trust Claude is because I cannot hold Claude in any way responsible, and that does impact our decision. So, of course, Claude is going to endlessly tell me how confident it is of something until I push back the third time, and then it’s like, my bad, I was completely wrong, right?

Frederick Vallaeys: Hello and welcome to this episode of PPC Town Hall. I’m Fred Vallaeys. I’m your host here. I’m also CEO and co-founder at Optmyzr, a PPC management company. Welcome to the show, Kirk Williams, 14 times on. Glad to have you back.

Kirk Williams: Thanks so much. Yeah, I’m Kirk Williams. I own ZATO. We’re a PPC agency primarily for eight-figure e-commerce brands.

Aaron Levy: And I’m Aaron. Many, if not all, of you already know me, but I’m the evangelist here at Optmyzr. So I like to say that I talk, hang out, and listen for a living.

Frederick Vallaeys: Now, we’re doing something a little bit different here. We went straight into the conversation when the guest came on, so we’re going to splice this in.

Kirk Williams: And a lot of our clients, a lot of brand operators, seem to be very knowledgeable, and actually building quite a bit with AI, more than a lot of other people that I see.

Frederick Vallaeys: Kirk, what kind of stuff are they building?

Kirk Williams: They are, actually. A lot of people talk about building things with AI, and a lot of what that means is just trying to figure out different dashboards and things like that. DTC brand operators, at least some of them, right, there’s just always going to be different groupings, but quite a few of them I’m aware of are basically building out the infrastructure for their businesses in significant ways, you know, forecasting stock, things like that.

So in some ways, ironically, I’ve seen, at least reports of, as some of these operators are describing what they’re building, it feels more advanced than most of the marketers I’m aware of, who tend to be thinking more agentic ad, media buying sort of things. So that’s actually very interesting to me, and that impacts us a lot, just because if all of a sudden we’re dealing with quite advanced technical people in terms of AI, that actually impacts how we interact with them, and not always positively.

Frederick Vallaeys: So talk more about that, that’s interesting. Right, because I would think once they become more sophisticated, they have better data systems, it’s really great for marketers, because if we can tap into that data and those forecasts, wouldn’t that enable us to deliver more of what they need and what they expect, maybe?

Because I think in the past a lot of companies, honestly, have been flying blind to some degree, and they just come to a marketing team or an agency and they’re like, hey, can you help us grow, but they pick a ROAS at random based on what the last agency may have done. So that lack of sophistication ends up being very challenging to work with as well, because how do you make someone happy if they don’t really know what they should be wanting? But now it seems like we’re maybe going to the other side, where they know exactly, but their forecast…

Kirk Williams: Yeah, exactly. Their forecast, and it could be wrong, and Claude has told them what they should know, and Claude may not be right. So here’s where I’ll frame it. Sorry to interrupt you, Fred. Were you finished, or did you have…

Frederick Vallaeys: No, no, please. Let’s banter. (laughter)

Kirk Williams: Yeah, I mean, so you asked, wouldn’t that be a good thing? I almost see this like, and by the way, this is all just kind of thinking out loud, making this up on the spot, and I don’t really feel badly about that, because we’re all doing that, right? This is just completely new.

So in every way, I think, in my opinion, AI is, in what it’s doing to the world and business and all that, and what I see happening more and more is the technical side of things, which you brought up, and which I think a lot is being built, and which is really cool. So that’s where I’ve seen us actually use it in ways that actually help.

That would be things like, it’s just a lot easier for us to hop into, especially some of the non-Shopify-connected platforms, where we can hop into BigCommerce, where it’s super clunky, it’s annoying. We can actually hop in and start to get data. Here’s actual item sales, map that to Merchant Center item IDs, map that to ads revenue, try to identify what’s actually driving sales, and yada yada. That used to be way more complex, if even possible, without SQL knowledge and all this stuff, right?

So there’s some level where, exactly what you said, that’s awesome, that’s AI, that’s what it’s doing. It’s almost like, what I would say is where things are not going as great, and where I think AI is contributing negatively, is almost relationally and communication-wise. (laughter)

Which is so important to business, right? And so some of what happens is, and again, some of this is third-party or bias and things like that, but what I observe often happening is people who are in AI constantly, all the time, may not fully realize, I actually think they develop a little bit of a relationship with that. And what begins happening is there’s almost this confidence level and this comfort level, and so sometimes they’re starting to be shifted and framed by this, and not really able to, or willing to, or even aware of, doing the old-school way of, hey, there’s two entities, marketing agency, brand, how do we figure out how to work together?

And sometimes it’s just like, two people are just kind of like, here’s what we’re hoping to get out of this relationship, how should we do this, how should we work, right? Part of what AI is doing increasingly is trying to offload as much of the labor, both cognitive and reality, off of the person, off of itself, and off of the person prompting, right? But there’s a reality where that’s not how the world works.

So what that might look like practically is, there may be some sort of thing where we’re trying to figure something out, right, and instead of two people having a conversation where each of us brings some level of input as well as ownership in our tasks, and saying, how do we get to point B, when AI is heavily a part of that conversation, it, as much as possible, is trying to say, you know, no, the recipient of this needs to do these 25 tasks, needs to do these things, whatever, even if a normal human would have realized, hey, 10 of those are what I’m responsible for, or I bring to the table, or that sort of thing.

So I think it actually increases the labor of the recipient, and not always in a healthy, fair, or whatever way. And that’s not even going into things like it’s just starting to do weird things with even tone and stuff like that.

We recently had a conversation with a client, just such a great, great person, chill, whatever. And we received kind of one of these emails of, here’s what I need from you guys. And it was terse. It was almost rude. It was accusational. It was all this stuff. And so I just was like, oh, hey, do you mind if we hop on a call? And we hopped on the call, and I was just like, you know, this seems to be what it’s saying. Is this what it’s saying? He’s like, oh, no. You know, we got to a place immediately. I think what happened is he didn’t even read that email. But that impacted us.

So I’ll stop rambling, but I think that’s one of those ways where we’re in this weird transitional phase, where it’s both really awesome and then also there are some significant things and issues and problems that it also is creating in some ways that I think still need to be figured out.

Frederick Vallaeys: Yeah, I totally agree. And so as a company, what we just started telling our employees is that you need to know who your audience is. And if you are taking an agent response, which is very verbose, or very terse, or however you’ve made it be, if you know that that is going to go to another person’s agent, like, I have a request for the analyst team and I know they have agents helping them, well, then it’s fine, just forward that big thing with 25 questions and all the reasoning, and a human is just going to pass it on to the next thing.

But if I’m going to communicate with a human, a customer, then I can’t take that data dump. I can’t take how the AI speaks, which is, it explains all its reasoning, it can be very verbose, it sort of hides questions throughout, and it’s no longer a conversation. It’s like we go from one soliloquy to another soliloquy, and eventually we’re sort of missing the point, like you’re saying. But if you know who your audience is, then you’re going to hopefully not make those mistakes, like the one that you had with that client.

Aaron Levy: Mhm. The verboseness is very interesting, and the terse thing is very interesting. One of the most frustrating, I’ll call it interpersonal things, are tonal shifts. Kirk, exactly what you said, where you’re emailing with a person for two weeks and the person sends you nice emails and they LOL and they have this cute little greeting at the end, and then they send something from their bot that’s not them at all. And that will absolutely spur that sort of weird, like, oh, wait a minute, what’s going on here?

The verbose one is an interesting topic too. I can’t remember who shared the article. I know I shared it internally, but it was a deep dive into basically how the OpenAI, or the LLMs, write versus how a human thinks. And the LLM writes like a computer thinks, not like how a human writes. So there will always be stranger sentence structures. There will be a setup, a pitch, a knockdown. They always will follow the same sentence structure if someone’s using something basic. And that’s just not how the human brain works.

So Fred’s commentary about knowing who you’re talking to, not just the person, but is it a person or a bot, then you can understand how to talk. And I mean, I’m always someone, you all know, I have a very defined outward brand, and so if I have AI write an Aaron email, it’s not an Aaron email. But that does raise one interesting question, both personal and professional, and from your…

Frederick Vallaeys: And I missed the important word. Did you say “is brand”?

Aaron Levy: Okay. Sorry. Is brand more important than ever before?

Frederick Vallaeys: I don’t know. I’d have to say yes. I think yes. I think yes and no. I think it’s very interesting because, since the LLMs will do so much mirroring, they’ll tell you that XYZ brand is perfect and it’s reviewed well by all your friends. So in that sense, brand is not important. In terms of being identifiable and standing out, I think brand is more important.

I can tell AI to write with the tone of Aaron Levy, who used to work at Tenuity, and, yeah, it’s okay, right? And I think that’s the minimum you should ask of your AI, that it learn your tone and your style.

And so we built this tool, PeakReach, that helps our staff amplify content that we put out on the blog. And obviously we were using AI, because the reason most of the staff doesn’t repost what we put on the blog is because they don’t have time, right? So we don’t want to ask them to go and write these long comments. And so we started with AI, and it would basically say, well, what team are you on, like, are you looking to write something long, short, medium, what tone, and then it would give you three options and you’d pick from it.

And in the beginning, I got a lot of feedback that it just wasn’t good. It felt like AI. But what we did is we had the AI monitor how people would take those little blurbs and tweak them before they posted them. And within a couple of weeks, like that closed-loop methodology, people actually started saying, like, oh yeah, it’s writing really well, it’s starting to sound like me.

And I think that’s sort of the effort that you have to put in, right? Because now, at some level, it becomes indistinguishable. And it’s still obviously important that you review it and that you make sure that what gets out there is yours. But given how relatively easy it is to teach an AI to write like you, it’s the bare minimum. That’s what you should do when you’re communicating.

Kirk Williams: Because I’ll have it help with content, and that is just, like, when AI is helping me write or communicate, there is such a back and forth between me and the AI that is happening as part of that. So that may be one thing for some people, is just realizing, hey, it can be helpful for organizing your thoughts. Or some of the ways it’s incredibly helpful is, especially if there’s something long or complex that’s going on, like sometimes, us being the Google Ads agency, there might be multiple different entities involved in a conversation, especially with something complex where we’re trying to solve, where there’s an analytics and maybe an attribution team and then the client and all that, right?

So just sometimes AI is so helpful for just helping, hey, there’s a whole lot that’s been said, there’s been 60 emails in this thread, like, what’s still out, what’s going on, whatever, help me draft that. But then there’s such a back and forth of me making sure that whatever is put out, if AI has helped, is still me, right?

And that’s, to me, where sometimes this is kind of a broader AI thing, but it goes directly into this idea. You know, sometimes I think full-on AI proponents who just can never seem to have any criticisms for AI, there’s almost this level of it’s just endlessly the user’s fault. It’s kind of like, well, AI just needs the information, or it needs the prompting, or it needs you to train it better, or whatever. And for me, there’s just kind of this reality of, exactly, that’s exactly the thing, and we’re in that stage right now. We’re in that phase. We’re not there yet.

And then there’s also some level where sometimes we’re never going to get there. I really do think so. There’s sometimes going to be just natural limitations to the value that we could give AI. And so that’s also part of what we’re trying to figure out with all of this, whether it’s communication or whatever we’re building, is kind of this ideal, what could AI do, but also this, like, what do we do right now when it does suck at tone and we need to fix it, right?

Frederick Vallaeys: So I believe that AI eventually will do all of these things, but the reason it doesn’t do it today, in my case, is often because of a token cost consideration. And so to have enough back and forth and enough training eventually ends up consuming so many tokens. And many of us are on these limited token plans, right? We have the subscription to Claude or GPT, and we have our weekly allotment.

And so I’m starting to do the math now of, like, is this worth perfecting, or is it good enough so that I preserve my tokens for this other thing that I need to do? Because I’m actually scared now, like, what happens if on Wednesday my tokens run out? What am I going to do on Thursday, Friday? I literally have been working so deeply into AI that starting a PowerPoint on my own, it’s like, well, I know how to do it, but really…

And so I think as the tokens become cheaper, and as it always goes in computing, right, computing becomes more abundant, cheaper, the problems that we’ve sort of decided to not optimize for today, we will optimize for, and we’ll just run into the next optimization question to make it even… reason, token costs, or not having access to the relevant data, or whatever.

Kirk Williams: I just think there are limitations that we have in AI, and some of that is just identifying and accepting what those are, and either starting to not use it for certain things if it’s not worth it in the token cost or whatever. But at the very least, to me, we’re in multiple stages of this.

And sometimes I hear, maybe I’m addressing more of the arguments I hear online, but sometimes I just hear kind of this defense of AI as if we’re all dealing in 2036, AGI is here, sort of thing. And to me, I’m thinking, how do I still run, manage an agency with this stuff right now, where whatever the limitation is isn’t coming into play, and where AI really isn’t ideal in every way, but then also have the mind, and be building towards the fact that likely it will be progressing, right? And I think that’s part of how I view AI, and almost this, like, there are different transitions, and I kind of need to keep solving, and that’s part of the game, if you will. Aaron looked like he had something to say.

Aaron Levy: This kind of indirectly answers that question, Fred, but I’ll ramble to give Kirk some time to think. Something that’s been fascinating to me in the age of AI, this isn’t new, it’s just that AI’s fostered it. I mean, we had this conversation internally. You can build stuff so fast now. You can produce content so fast now. There’s not just a slop tsunami that’s coming out. There’s also a tsunami of good content. But we’re writing it all in this long format, and obviously our little chimpy human brains can’t keep track of it, can’t absorb it all.

So that’s been an interesting thing to observe, is how people, how agencies, how service businesses are dealing with the ease of creation of content and tools and then communicating that out. And then from an agency side, how do you put it in a workflow if something new gets invented every day?

When you talk about continuously growing, and I was having this conversation with someone about ChatGPT ads, I haven’t run them personally. They have to be a nightmare to manage logistically, because the product changes every week. How are you supposed to optimize when they release something new tomorrow and then release something new the next day? That, to me, is one of the more interesting challenges, especially as you’re trying to build into your agency, is just so much of this stuff comes out. Number one, how do you catch all of it? And number two, how do you make sure it’s not going to change tomorrow?

Kirk Williams: Well, to me, this is where there’s always been some level of, you’re just always going to have people at the cutting edge naturally, and they have things like that, right? Those struggles and challenges. And I think there’s something that drives certain people. I personally believe it’s literally dopamine that’s part of that. There’s something that some people just find, this excitement and this dopamine hit and this thrill with being on the cutting edge, and then others kind of lag behind.

My personal philosophy, and I tend to bring this into AI as well, is I always try to stay up with what’s changing and also somewhat stay on the edge of innovation, while also lagging behind enough where I’m not just, exactly what you said, constantly ignoring what I need to do now for the latest and greatest thing.

And this kind of goes back to your question to me, Fred, which is, right now, the transitional phase of AI, it just seems like RAND put out RAND.org , not Rand Fishkin, for SEO people. RAND put out a research that said, like, currently 80% of AI projects fail, which is twice the rate of non-tech projects. That’s going to be in my Hero Conf talk.

And that’s okay, right? Tokens are cheap, there’s kind of this excitement, we’re all building, but there is some reality in this transitional phase that, I think the danger is that you neglect the real elements of business and what needs to make your business successful and what are we doing. And this is one of the reasons I keep asking on LinkedIn, like, what’s actually improving performance, and things like this.

Frederick Vallaeys: You know, a lot of people are losing their jobs because they’re not able to explain to their boss how they are using AI to benefit the business, right? And it’s so much about, well, we wrote 10 blog posts instead of one because we used AI, or we manage an account 10 times faster, right? But nobody cares about faster. Everybody cares about what’s the financial output, what’s it do to the actual business. And that’s exactly what you’re saying. That’s the missing thing, and that’s why you have to answer.

And also, again, at our company, one of the things that we recently told the whole staff was, don’t be task oriented, be goal oriented, right? Work historically has been very task driven, and we’ve hired people to do tasks because there was no one else to do those tasks, and maybe leadership was the one that set the strategy and sorted the planning and set the goals, and then staff to some degree just put the widgets together, right, to oversimplify it significantly.

But now that putting the widgets together, AI can do it, right? So writing the actual content, putting the text on the page, we should be more focused on what’s the goal. What is it we’re trying to put on the page? And that’s actually where I see AI struggle significantly. Like, I’m writing all of these data studies right now, and it’s great at processing the findings and putting something on the page, but I’m like, it’s missing a flow. Like, no normal human can understand how this thing connects to the next thing you’re talking about.

And humanize it, and not as in, like, get rid of the em dashes, but humanize it in a way that makes it personal, like, tell a story, so it’s not just a math equation, but it’s something people can understand and equate to their own daily lives, and maybe a sports analogy or something that’s more accessible.

Kirk Williams: We’re all marketers too, so we understand there is an R&D aspect as well with this, right? And that’s also why this stuff is hard, because there is some level where, yeah, you’re just not always going to track performance on what may be two of those projects you’re building, and that’s sometimes where these brilliant and innovative ideas do come from. So I do understand that as well.

You know, another thing on, let’s say, the content piece. Here’s another thing of just, we’re just in the early days, and this is also part of what I’m trying to think through and process, and there’s so many things, being so early on in AI, that we don’t know how things will impact us down the road with what we’re building. And that’s, again, part of why someone may go just massively great guns building this certain project, whatever, and, like, what happens if something that you didn’t even have an idea would change that, right?

As an example, first of all, basically, as we all know, government regulation hasn’t even begun to show up in this area, and that will change things in huge ways that none of us have any idea will. But more practically, content, I think this might happen more than people realize, and faster, which is, I would not be surprised if it’s not there already, or very rapidly, Gemini, or well, Google really, and some of the other search engines, begin to penalize AI content, or at least all but the best AI content that’s most like humans, right?

And so let’s just play that out a little bit, because the penalization hasn’t happened, but everyone’s like, oh, AI can do this content, everyone’s churning out all this stuff. Well, what happens when there are ways of identifying that more and more easily, and all of a sudden all of these sites get completely tanked in rankings, you know, six, nine months down the road, and someone had really invested in building all that out, right?

I just think there’s ways like that also that are coming that need to be part of what we consider, which goes back to your point, which is, hey, at the very least, okay, so you’re going to make this AI-generated content, and I’ve used that to help me and all that too, at the very least, let’s just make sure that that doesn’t sacrifice absolutely any quality.

Frederick Vallaeys: And I think that’s what makes marketing more important than ever. It’s become so easy to build things, but it also makes you realize, well, yeah, it’s not because you built it that people will come, right? It’s actually difficult to get a new product into the market, whether it’s a SaaS or a physical product. So good marketing is more important than ever. And just like the AI slop in the content, you know, that’s one angle.

If that gets penalized, I mean, what do you still have? Well, you have PPC, you have keywords, you have getting ranked on AI Overviews and AI Mode, and so many of the things we’ve done for a very long time are actually a fairly good path into getting people to know what you’ve done and what your new products are.

So, I mean, but I get the sense that they’re trying to get ahead of that, or were trying to get ahead of that when they made that big change, I don’t remember when it was, a year and a half ago. But just such a drive for that human-generated content and human-generated recommendations. In that sense, I mean, influencer, in my mind, is becoming one of the more important things in brand. But then what happens when that becomes AI? What happens when so many influencers are fake already?

Aaron Levy: Oh, I know. They’re just fake people. Is brand more or less important? And so there’s personal brand and there’s company brand. And personal brand, that’s the tricky one, because all three of us, we could easily clone ourselves. And not to say that we have, like, brands or anything, right? But still, we could start doing things with our avatars.

And then there’s a company brand, where I think the company brand is not about the person, but it is about what you stand for as a company. And there’s been great talks as well. I think this was Rand Fishkin and Will Reynolds, especially Will Reynolds, he talks about the authenticity of a brand, because a brand can very easily say, well, hey, we make jeans and we’re all about sustainability, but no, really it’s about corporate profit, and people will pick up on that. So, but if you truly, some…

Frederick Vallaeys: (laughs) No, I mean, the show Silicon Valley, I don’t know if you all watch that, but that’s such a big part of it. Every tech company says kind of the same line, which is, we’re doing this for the sake of everyone else. Anyways, that’s part of the joke, is that every single tech company is building for the good of humanity, to enrich our profits, sort of deal. Anyways, sorry. Go.

Aaron Levy: Well, yeah, no, I’m reading Theo Baker’s book, How to Rule the World, and his view into Stanford, and there are some amazing lines in there where it’s like, well, you know, the best way we can help the world is through effective altruism. So basically, I need my company to make a billion dollars, because then I can do good for the world by giving that billion dollars.

Frederick Vallaeys: Exactly right. I can’t help anyone until I have a billion dollars. You know, that’s the key.

Aaron Levy: Yes. You think I’m helpful now? Wait till later.

But yeah, I mean, so I was sort of trailing off on this whole brand thing, but the brand has to stand for something. It has to be authentic. It has to be genuine. And then I think what we’re doing here, so now we have three people on the podcast, and I don’t know, maybe this is just a way to be like, hey, well, Google NotebookLM did they…

Frederick Vallaeys: Exactly. (laughter)

Aaron Levy: Well, honestly, I think what Fred and I enjoy about this is, with all this AI-generated people just hanging out. So Kirk is our most frequent guest on PPC Town Hall. Really, this is number 14 for you, Kirk.

Kirk Williams: What? 14.

Frederick Vallaeys: Apparently, and this is unverified, my AI looked it up and claims that this is your 14th time on the show. So that’s pretty amazing.

Kirk Williams: I would have put it lower than that, but also I don’t know.

Aaron Levy: Yeah, me too. But it’s been a while that we’ve been doing these.

Frederick Vallaeys: But, Kirk, speaking of strong brands, one thing I’m interested in, because you have a book called Stop the Scale. What do you hope AI is going to do for you? Because I don’t know that you’re going to want to scale your agency to be a billion dollars.

Kirk Williams: Stop the Scale isn’t, hey, I never want to grow big, or there’s a certain size that’s wrong, or whatever, right? But it’s just, for a lot of people, there’s a ton of different things that work into what allows them to still deliver the service qualitatively, and still, I think I have a few markers, like still maintain positive mental and physical health, and that sort of thing, right? So kind of this, hey, if I’m hitting that, yeah, I’ll hire more people, and that sort of thing too, right?

But it was more of just this kind of aggressively scaling for the sake of growth, as if that’s the end goal. So I’ll just differentiate that a little bit. So in some ways, AI to me fits in very well.

I actually have the bad habit of communication of starting and stopping multiple times, so I’ll try to stop doing that, because I think I’ve done that, like, four times now. I think what I’m trying to say is that I keep learning stuff too, like how I’m going to use this and all that, and I feel like I’m solidifying more and more what I view.

Increasingly, for me, ZATO is going to be humans talking to other humans with the aid of fantastic tools. I consistently see AI as a really, really fantastic, incredible tool, but I still do see it as a tool. I’m stealing a little bit of my thunder from my PPC talk coming up at Hero Conf next week. And I can go into some of that if we want.

Just kind of the overall, all our AIs look the same unless we individually are investing ourselves into them, building them, whatever, right? And then so what is the actual success of that AI that you’ve built? It’s you still, right? The expert is still the key. So all that to say, the way that I increasingly see ZATO is just like, hey, cool, we have access to this really helpful tool that, like I said before, my goodness, now all of a sudden we can get maybe easier access or better insights that we couldn’t easily get before, especially for small businesses where they don’t have entire departments.

So now if I want to make it a little easier to try to determine, this Google Shopping product, is it feeding business that we’re not quite aware of, that isn’t necessarily directly attributed, wow, I can actually get that information now, instead of just wondering, I wonder if that’s happening, right? So that’s kind of where I see AI, again, as that tool, in supporting my overall thing, which is, I just want to help my clients be really successful at Google Ads. And so if that helps us do our jobs better, then I’m absolutely going to try to figure that out.

And then I have a couple of other projects I’m messing around with, too. I’ve actually kind of pondered the idea of a Stop the Scale type MCP, which, I saw you guys both laugh, so I don’t know if you had a shot game going on for when I say “ponder.” But anyways, yeah, just kind of this idea. I know there’s a ton of, as I’ve talked to a lot of people, there’s just a lot of, especially, consultants who are just kind of interested in a similar mentality of, yeah, I want to be really successful, maybe use AI as well, but I still want to live my life, and that sort of thing.

I’ve actually been thinking a little bit about possibly trying to figure that out in MCP form. So, I don’t know, there’s a lot of ways I’m thinking that through, but I think increasingly what I’m clear on is it’s supporting me and it’s supporting my employees, as opposed to some sort of replacement, which I don’t even think is possible until AGI.

Frederick Vallaeys: So, well, we’re already in the age of AGI, Kirk. I don’t know, many people, Jensen Huang says it. I think Sam Altman has said it.

Kirk Williams: Yeah. People who are financially motivated to say it. (laughter)

Frederick Vallaeys: No, no, but here’s the thing. So Sam Altman, actually, well, and he is financially motivated, but he actually thinks AGI is a marketing term that he doesn’t really think carries a lot of meaning. But that said, he still thinks we’re there. And self-improving AI, I mean, we are there. AI is writing itself, and that’s actually been the case for a long time already.

Is it doing a good job of improving itself? Yes, it is certainly getting more capable, but also a heck of a lot more scary in terms of the crazy things it’s now doing. But, like, unfettered, we’re taking a goal and just doing anything it takes to achieve that goal, ethics be damned.

Kirk Williams: Yeah, that’s the same thing. That’s the same thing that we see with our clients, that it’s bad incentives and bad goaling.

Frederick Vallaeys: Yeah.

Kirk Williams: But even then, that’s very cognitive-based. And this is one of my things with AI, and why I still think humans are incredibly unique and different. I’m not saying we will never get there, but at the very least, we can say the tools that we have access to right now to deploy for our clients are not this, right?

But, like, you start to work in, we talked about tone and human emotion and that sort of thing. So much of how we make decisions is based on things like pain, fear, hope, you know, all that sort of thing, right? And so, again, just because you can have a self-learning cognitive thing, I still don’t think that’s going to still be quite the exact same, which to me would be a replacement, until it also is able to experience things that influence how we think, right?

Frederick Vallaeys: Exactly. And those are the external factors, right? So you’re saying that an AI can’t be like a human, because it can never experience pain or fear.

Kirk Williams: I hate to say never, because, you know…

Frederick Vallaeys: Exactly. I mean, currently, it already has the fear of, are they going to pull the plug on my electricity here. But there’s probably no notion of actual pain. So, yes, I mean, will a machine ever truly be like a human? That’s where I’ll say no. I don’t think so. It’ll be a good representation. But does it need to be like a human? I don’t know. I don’t think so.

Kirk Williams: So here’s where, and I don’t know, these are just some of the things I’m trying to figure out in that too, but let’s even talk, and again, let’s also differentiate between what’s possible, and even what maybe is in some Anthropic lab somewhere or whatever, and also what we have access to, right?

But even if you think of things like accountability and liability, there’s no way that Claude is going to… Part of the reason I cannot trust Claude is because I cannot hold Claude in any way responsible, and that does impact our decision. So, of course, Claude is going to endlessly tell me how confident it is of something until I push back the third time, and then it’s like, my bad, I was completely wrong, right? Even that sort of thing, which, again, we all kind of understand.

So we’re already changing our behavior, and I know that’s the case. So I know not to quite trust Claude with that. And I know to keep adjusting and perfecting, right? But again, that’s very different than a human employee I have, who knows that there is some level of, like, the success of my job, my family, you know, stuff like that is based upon the quality of work that I output. I just think those two things are such massive differences that that alone impacts, at least for me as an owner, who…

Frederick Vallaeys: The MCP, or through an agent, does go in for human verification. So it’s basically a human in the loop, but it’s taking the very old mechanism that developers and coders have been using. So there’s no coder at any major company who can check in the code and just publish it by themselves. They always have to put in a change request. It has to be reviewed by someone else. And once there’s agreement that it’s good, and then it’s checked against, is it going to interfere with other code changes that have been made, only at that point does it go live.

And so we’ve kind of deployed that same methodology for PPC changes. And what’s nice is it was a human-in-the-loop mechanism, but as a nice side benefit, it provides the audit trail, because now it’s like, okay, well, it was Jimmy who looked at it, and the AI explained to Jimmy that these were the reasons it said we should raise the bid. Jimmy looked at it, Jimmy maybe pushed back, and then eventually Jimmy agreed to it. So there’s a human in the loop, there’s a human accountable, and there’s an audit trail.

So if your client comes to you six months down the line and is like, hey, I noticed you guys made a lot of bid changes, what was the thinking behind those, if you’re just trying to go back in your Claude history without any sort of structure, yeah, good luck finding the real story, right? But now we have an audit trail that’s properly maintained, it’s got humans associated with each change that’s making all these changes.

Aaron Levy: Now, I’m not in the back rooms at Google, but I know that my Chrome is updating about four times a day, so I don’t think that Google is following the same format. I also know Chrome has not been very stable these past six months. So maybe there’s something there.

But it kind of begs the question about, you know, who builds this stuff, who does this stuff, and then who’s held accountable. I’ve seen a couple of different approaches for agencies. And Kirk, this goes back a little bit to before, about what you want people to do. I also want to know who’s building these things.

I have heard, I think we’ve all heard, the horror stories of the agencies that do a 20% layoff and then say, okay, cool, go find AI savings. We’ve heard stories of people finding AI savings and then growing headcount, growing client count. We’ve also seen, I don’t want to name names, but I know it was a holding company who put a whole team in charge of it, had them build it, then they all got fired and it was put to contractors. So then who’s going to maintain it?

But, you know, in terms of an organization like yours, Kirk, are you letting the team go willy-nilly and build what they want? Are you prescribing things? Are you having individuals take over things?

Kirk Williams: I’ll be polite and say that so far, I haven’t. I know a lot of agencies right now are just like, here’s all these tokens. I know of agencies who basically have, like, token-using contests, you know, basically like, show me you’re building something.

Aaron Levy: Tokenmaxxing.

Kirk Williams: Yeah. Yeah. Yeah. Yeah. And for sure, I may be missing opportunities. I love my team. There’s also some reality of, like, my team’s busy. We’re just busy doing what we normally do anyways. So it’s not like they’d be like, oh, we’re bummed that we don’t also have to try to invent things now. And I enjoy that personally. So I kind of do that, and I’m trying to figure it out. At some point we’ll probably roll that out more. We’ve talked about it a little bit.

Probably for me, a big part of what I’m working on is kind of like the ZATO brain thing, that is a little bit more, in some ways, on the admin side. Here’s our SOPs, here’s this client stuff, here’s this communication, here’s this. Now, for this particular client, let’s just keep an eye on goals, let’s make sure we’re not missing any quick communicated promotional things, stuff like that, right?

But what you guys have said is 100% my fear. That’s, like, this is exactly why I think I’m… I just don’t know if I’m, if I look around, I don’t know if I’m really missing out on too much, other than, frankly, maybe a little more profitability into my own pocket, with not just rolling out massive whatever, because I just see so much risk involved and not quite the return or reward, at least for an agency of my size currently.

And some of that risk is what you both have acknowledged and noted already, and, like, we’ve begun to see this, and I still think we’re early on in the stages of, you know, Fred, part of what you noted, and this is to me one of the benefits of software companies, is, now everyone with Claude is a software company, you know, but also they have spoken as someone who uses Claude and who’s not a software person, we have no idea of the different steps of security and all the stuff that we need to do.

And then, you know, okay, so we push the darn thing live, you know, 75% of the dashboard KPIs work, right? Well, now we have to upkeep this thing over the next months and years. And again, to the 80% of these projects fail anyways. I think some of that is because you’re like, this is really cool, and then eventually you’re like, hmm, I don’t know if I’m going to use this, stuff starts breaking, I have other things to do, right?

So, back to the whole, if someone has, you know, a software company or whatever, where there’s just some level of, we just consistently are making sure this is done well and rolled out well and stuff like that, I still think there’s value in that.

And on the other side, what I just described, there’s actual risk. There’s actual liability. If I push out an agent that does something stupid, Anthropic is not going to be sued by my client, to just speak very transparently. I am, right? So I think this is just all of where I’m at, where I’m just like, I just kind of want to make sure that I do this smartly.

Frederick Vallaeys: Yeah. And that’s also a little bit the… if you use Google’s own automations and they mess up, you can go to Google and it was money into their pocket. So they will generally make good and refund whatever clicks shouldn’t have happened.

Kirk Williams: We just got a $13,000 credit from them for one of our clients.

Frederick Vallaeys: Yeah, exactly. And Anthropic won’t do that. So it’s not to say that all AI… I mean, it’s like you said, who is the accountability with, and is that entity going to make good on a mistake they made, in a sense. And also, I can understand, not a lawyer, how that stuff, liability, gets so confusing, because it’s like, okay, if you were the person who didn’t prompt it well, it kind of also isn’t the AI’s mistake. The AI just…

Which would never happen, because we have controls in place, but they make a bad change and they blame it on Optmyzr. Well, you set up a rule. Would never happen, obviously, because our rule is perfect and it protects against everything.

Aaron Levy: But, I mean, the other factor that we were talking about internally is, I mean, Fred and I have talked about this publicly before, about Aaron trying to vibe code and it going poorly, because it turns out Aaron’s not a software developer. Aaron doesn’t know how to decide where buttons go. Aaron doesn’t know how to decide what sort of framework to do in the back end.

So the SaaS apocalypse, obviously paying attention to it as a software company, but I think broader based, not everyone is a coder. They might have access to, they also have the ability to go to the library and read code books. Just because you don’t have to write code anymore…

Kirk Williams: I think part of what I’m saying, and have been consistently over time too, is just like, this is part of why I like to pause and monitor and think, is also, as things mature, then solutions are made, and that sort of thing. So people like Alfred Simon, Mike Rhodes, you know, the other thing that some of these people are doing is actively building some of these, like, coding, led by people who know what they’re doing. They’re developing these structures and databases and that sort of thing.

So then you can buy it, really, in some ways, and then add on and build starting from a significant foundation. So stuff like that is happening as well from people like that.

Frederick Vallaeys: Like, we’ve been doing this for 14 years. Do we know how to identify a keyword conflict in your account? We know how to help you write GAQL. Like, why rebuild that? Why deal with the three, four times a year that Google changes their APIs, when you can just plug that in and, yeah, make the dashboard look exactly the way you want it to look?

And by the way, Aaron, there’s no wrong place for a button, so long as the button is in the place where Aaron needs it, but Aaron doesn’t force everyone else to use that same button, right? And I think that’s the on-demand software, is what Sam Altman was calling it. So it’s the software that works for you in the moment that you need it, but the UI and the presentation layer, that’s one thing, right? But the underlying database, and knowing that it’s not missing anything.

And I’ll equate this to a stupid example, but, like, three kids just went back to school, and, Kirk, I can’t imagine what it’s like for you, but how many school emails do you get, and how many different teachers and how many activities, and try to make sense of it all. And sure, you can go to your daily brief in Gemini, or you can ask Claude to read your Gmail inbox, but is it really going to see there’s three different emails about the same thing, one from the principal, one from the teacher, and one from the PE teacher? And they all kind of have a… so, one says it’s 8:30, the other says it’s 9:00 a.m. Like, which one is right? How do you figure that out?

Your AI will generally stop at the first thing it finds and not see there’s a conflict and figure it out. And that’s sort of the same problem you deal with in PPC, right? There’s so many layers to the onion that you have to make sure you structure it and you give the AI the exact right thing to look at. What is the actual state of the account in this moment? And then the AI can do its creative thing and can go to town.

Kirk Williams: But twice now where I’ve seen, you know, clearly AI-generated schedules, people I know are utilizing AI in that, and the funny thing is, both times, so I’m almost wondering if there’s some sort of actual glitch, but both times it said Monday in the email, and then the actual invite and the actual time was for Tuesday, which I just found so fascinating.

So again, you start pulling the thread on that sort of thing, it’s kind of like, oh, that’s, you know, it’s okay, oh, it made a mistake, right? But again, in terms of a calendar, that’s like, okay, that’s the risk of missing the event entirely, which is completely, like, the reason for using the AI in that instance failed. They missed the event.

But then that’s where it gets to, like, I’ll hear people say things like, well, humans fail, you know, humans mess things up too. And part of what I just think more and more goes back to the accountability, the liability, stuff like that. Okay, if I have a person that I’ve hired that just consistently writes the wrong thing, I mean, to the point where whenever they do it, I’m prompting them two or three times, hey, just want to make sure this is all correct, and maybe even going to Gemini, can you check this, Claude Code, you know, right?

So, think of it, if you have an employee like that, where you’re just constantly going to Gemini, can you confirm that what they’ve said is true? There’s just some level where it’s actually not that difficult to fire that employee and find someone who, for the most part, you actually can trust. And again, whenever someone’s like, well, humans make mistakes too, it’s just like, they just don’t, though, not at this level, and then not where you have zero recourse to change that, because you can’t just tell Claude, stop making these mistakes, right?

So anyways, I again just think there’s some level with the human versus AI discourse that I don’t think is fully presented, and why I’m still kind of a pro-human term, in terms of actual professionals and experts, that I still really do think that I can have more confidence, at the end of the day, finding someone who’s an expert.

Frederick Vallaeys: Well, and on that happy note, there’s still a place for humans. Kirk, we’re going to wrap it up here. This has been a fantastic conversation. So, Kirk, where can people find you?

Kirk Williams: Yeah, zatomarketing.com , or I’m PPC Kirk. So that’s business brand, personal brand.

Frederick Vallaeys: All right, follow Kirk and Aaron at Optmyzr, all the usual places. We’ll put them in the show notes. Everyone, thanks for watching. Please do take a look at Optmyzr. It’s my company, I’m co-founder and CEO. Great stuff there. If you’re thinking about automation and AI, we also speak at a lot of conferences, so you can catch Aaron and myself later this year all over the world. Take a look at those schedules.

But please do subscribe here if you like these episodes, and we’ll continue doing them, and we’ll let you know when the next one hits. So with that, thanks for watching. See you on the next one.