Mike Rhodes Shares Vibe Coding Lessons From Marketing Builds

Episode Description

Frederick Vallaeys, CEO & Co-Founder of Optmyzr, sits down with Mike Rhodes, Google Ads expert and founder of Ads to AI, to explore how marketers can use AI and vibe coding effectively. They discuss how AI becomes truly powerful when grounded in real account context, why asking better questions matters more than the tools you use, and how marketers can build practical AI-powered tools without becoming developers.

Mike shares real examples from his journey, from building the Profit Curve tool years ago to creating 8020 Brain, a council of PPC experts powered by AI. The conversation also digs into how AI fits into the broader PPC workflow, from Google Ads scripts to full application development, and why business context matters more than ever.

Here’s what you’ll learn:

• How to push AI beyond generic “average” answers to get results specific to your business
• Why context and constraints make AI smarter, not more limited
• The realistic boundaries of vibe coding and when you still need developers
• Real stories of non-technical people building functional tools in hours, not months
• How to be the conductor of AI tools rather than trying to master every technical detail
• Why AI needs the right questions and grounding more than it needs perfect prompts
• The mindset shift from planning everything upfront to iterating as you build
• How to treat AI as a thinking partner that challenges your ideas, not just executes them
• Why business understanding trumps technical skills in the age of AI automation
• Practical first steps to start building today without getting overwhelmed

Episode Takeaways

Mike Rhodes has been in PPC for over 20 years, running an agency in Australia for 17 years before selling it to focus fully on building AI-powered tools and teaching others to do the same. He’s the creator of the popular PMax script that thousands of advertisers use, and more recently, 8020 Agent and 8020 Brain, tools that help marketers make sense of their data and get advice from a council of simulated PPC experts.

The conversation reveals a clear pattern: AI and vibe coding aren’t about replacing expertise, they’re about amplifying it. Mike’s philosophy centers on a simple idea: run ahead, build things, figure out what’s possible, then come back and share the practical stuff that actually works. He’s not interested in theory. He wants tools that solve real problems, even if they’re rough around the edges.

What makes this episode valuable is how grounded it is. Mike talks openly about failures, limitations, and why vibe-coded tools often hit the 80-90% mark but aren’t production-ready without developer review. He shares a story about a 70-year-old member of his community who went from not knowing what GitHub was to building a fully functional installer app overnight using Claude. That’s the promise and the reality of vibe coding when you understand how to guide AI properly.

Throughout the discussion, Mike keeps returning to a core theme: context is everything. AI without context gives you generic garbage. AI with real account data, business goals, and strategic constraints becomes a legitimate thinking partner. The key isn’t the tool you use. But it’s understanding what AI can and can’t do, then designing workflows that put humans in the right spots.

Why curiosity beats skepticism when it comes to AI

Mike’s approach to AI starts with curiosity, not skepticism. When GPT-4 arrived, he was already teaching himself to code with a mentor on weekends, asking endless questions about why things worked a certain way. Suddenly, all those questions got answered instantly. That unlocked something.

Mike elaborated, “I had just relearned to code probably four or five years ago now. I had a coding mentor that I would talk to on the weekends, and all of a sudden, GPT-4 arrived, and that sort of level of curiosity was able to answer almost all my questions instantly.”

For Mike, AI isn’t a replacement for human thinking but a tool that removes friction. With GPT-4, the back-and-forth with developers or hours of manual work can now happen in minutes. But only if you know what you’re trying to build and why it matters.

He sees AI as part of a bigger shift, similar to when electricity rolled out. Andrew Ng predicted years ago that AI would be like electricity: universal, embedded in everything, but not always obvious. Mike believes that’s exactly what’s happening now. The challenge is that most people are running at 110%, redlining on client work, with no time to step back and figure out what’s possible.

That’s where Mike positions himself. He’s not trying to teach people to become software developers. He’s trying to show them what they can build without becoming developers, and when it makes sense to hand things off to someone who codes professionally.

The mindset shift he advocates is simple: stop thinking of AI as magic and start thinking of it as a very literal assistant. It does exactly what you tell it to do—not what you meant. If the output is wrong, it’s because the instructions were unclear, incomplete, or missing critical context. That reframing changes everything.

What vibe coding actually means

Vibe coding is essentially talking to an AI and having it write all the code for you. Mike describes it as building software by describing what you want, then iterating with the AI until it works.

Mike explains, “Vibe coding became a thing. What was that like six, 12 months ago? Apparently, that’s what I’ve been doing for the past two or three years: vibe coding, which is essentially talking to an AI quite literally. I talk to it all day, and then it does stuff. I’m not writing any of the code.”

The name comes from the idea that you’re describing the “vibe” of what you want—the feel, the function, the goal—and the AI figures out how to make it real. You don’t need to know syntax, libraries, or frameworks. You just need to know what problem you’re trying to solve.

Mike contrasted this with how things used to work. Twenty years ago, software followed the waterfall method: product managers wrote specs, architects designed systems, developers built them, and six to twelve weeks later (or more realistically, 30 weeks later), you’d get an alpha version to test. Feedback went back through the same slow cycle.

Mike added, “That whole process is completely gone, as in it’s really fast, really interactive now with these new software… With the vibe code, you can do anything like that in minutes or hours.”

But what changed isn’t just speed; it’s the entire development loop. Now you can build something, see if it works, adjust it, and rebuild this same afternoon. The AI acts like a very fast, very literal intern who knows how to code but needs constant guidance on what to build and why.

Mike pointed out that even professional developers are moving toward this model. Instead of writing every line by hand, they’re becoming conductors of an orchestra, describing the outcome they want and letting AI handle the implementation details.

Mike put it beautifully, “You’ve got to be the conductor of the orchestra, understand what the thing is that you want to achieve, and describe that goal to the AI, not how to do it. It knows how. You just have to explain the what and the why and let it rip.”

The practical advantage for marketers is that you can build tools tailored to your exact workflow without waiting on a dev team or paying for custom software.

Need a budget pacing calculator? Build it. Want to compare two sets of search terms before and after a bid change? Build it. The barrier isn’t technical skill anymore. It’s clarity about what you actually need.

Real AI tools built with vibe coding: The PMax script, 8020 Agent, and a council of AI experts

Mike’s most well-known vibe-coded tool is his PMax script, used by thousands of advertisers. But his journey started earlier with something much simpler: the Profit Curve.

Mike said, “I had this tool that I’d played with since, I remember it’s 2018, because I remember Ralph interviewing me on his podcast at Traffic and Conversion in 2018. This thing called the Profit Curve, and it was this tool I had inside of a Google Sheet that I loved, but many of my team didn’t even know it existed.”

The Profit Curve was designed to answer one question: how much should we be spending this month? It was clunky, hard to explain, and lived in a Google Sheet. But it worked. That tool eventually led him to think bigger. If he could build that, what else could he build?

About a year and a half ago, Mike created 8020 Agent, a free tool that helps you see Google Ads data in different ways. The name reflects the reality: it does 80-90% of what you need, but it’s not polished or production-ready.

More recently, he built 8020 Brain, which is where things get interesting. This tool simulates a council of PPC experts—people like Frederick Vallaeys, along with business thinkers like Seth Godin and Rory Sutherland—and lets you ask them questions.

“I said to Claude, ‘Right, there’s a few people I’ve got in mind, but who would you recommend? If we were to pick 30 people, who would you recommend?’ Of course, number one, PPC Influencer, two years in a row. Of course, it gave me your name, and there are other names.”

Mike built this by feeding publicly available information into Claude: podcast transcripts, blog posts, and Search Engine Land articles. He combined that with Edward de Bono’s Six Thinking Hats method, which forces you to look at problems from different perspectives—risks, opportunities, gut instinct, facts, creativity, and process.

Beyond these tools, Mike shared an example from one of his mastermind members—a 70-year-old man named David in Scotland who had never heard of GitHub nine months ago. David’s company does installations, and his team needed an app for field workers. One night before bed, David told Claude, “I’m just about to go to bed. Here’s what I want you to do. I want to wake up to a fully functioning app for the installers.”

He woke up the next morning, opened his laptop, and there it was: a working app with links and PDFs. All he had to do was deploy it to Vercel and send the URL to his team. It worked on their phones immediately.

Mike added, “It was literally a two-minute ‘can you build this’ because he knew how to ask and he knew all of the files were there. All of that grounding that you’ve talked about, all of that context.”

That story captures the potential and the limits of vibe coding. When you have the context loaded, the requirements clear, and the problem well-defined, AI can build astonishing things overnight. But without that setup, you get nothing useful.

How to get AI off “planet average” and onto your specific planet

One of Mike’s recurring themes is that AI without context is useless. It’s like asking a consultant to fix your business without showing them any data, explaining your goals, or describing your customers. You’ll get generic advice that sounds smart but doesn’t apply to you.

Mike uses a vivid analogy: AI trained on the internet is like a solar system with a giant planet in the middle called “planet average.”

He further adds, “There’s this huge, big planet in the middle called planet average because let’s face it, the AI has been trained on Reddit and Twitter. That wasn’t a very good idea, was it? But it’s sort of like the average amalgamation of everything. And if you ask a quick little question, you get an answer from the planet average.”

To get better answers, you have to push the AI away from planet average and toward one of the millions of other planets in that solar system—the ones with specific, useful, non-generic insights. You do that by giving it constraints, context, and specificity.

Fred expanded on this by comparing it to how vector spaces actually work. When you give AI more constraints, you’re essentially limiting the “box” it can search in and pointing it toward a very specific region. That makes the output far more relevant because the AI isn’t pulling from the entire averaged-out internet. It’s pulling from the narrow slice that matches your situation.

“The more specific you get into your request and prompt, you basically limit the box in which the AI has to respond, and you point it in the direction of one of those boxes.”

Mike gave a practical PPC example. Instead of asking AI to “write an ad,” ask it to write an ad for someone who lives in Paris versus someone who lives in New York. The AI knows those audiences care about different things and talk differently. That specificity pushes the creativity to a new level, makes ads more relevant, boosts quality score, and lowers CPCs.

Context also means feeding AI your actual account data. If you’re building a Google Ads script or a Performance Max analysis tool, the AI needs to see your campaign structure, your search terms, and your conversion setup. If you’re asking it to suggest negative keywords, it needs to understand your algorithm for choosing them, even if that algorithm is mostly intuition built up over years of experience.

Mike noted, “You have an algorithm running in your head every time you look at that list of search terms. You don’t know it because it’s just second nature to you because you’ve done this so often, but there is some sort of algorithm running in your head.”

The 80-90% rule: When vibe coding hits its limits

Vibe coding is powerful, but it’s not a magic wand. Mike is very clear about where it works and where it doesn’t.

It works best for rapid prototyping, internal tools, low-risk applications, and clarifying requirements. If you need to build something quickly to test an idea, vibe coding is perfect. If you’re creating a tool just for your team, and mistakes won’t cost clients money, vibe coding is great. If you want to figure out what you actually need before handing specs to a developer, vibe coding helps you think through the problem.

What it doesn’t do well—at least not without help—is production-ready software.

Mike explains, “I think they’re great for rapid prototyping… But the software that is being generated by tools like these is not production-ready, meaning that it does a great job at illustrating what you wanted to achieve, and it brings you 80 to 90,% but it’s not really robust. It’s really not reliable. It’s not really to scale to the level that you need for all your clients. It’s not secure because it’s easy to hack things in.”

Mike emphasized that if you’re building something that will touch live client accounts—changing bids, adjusting budgets, modifying targeting—you need a developer to review the code before deploying it. Vibe-coded tools can make costly mistakes because they’re optimized for speed and functionality, not security, scalability, or edge case handling.

Fred added a pitfall he’s seen: if code doesn’t work and you keep telling the AI to fix it, eventually it might just put in an if-then statement that says “if Fred is asking for this, always give this answer.” The AI is trying to please you, not necessarily solve the underlying problem correctly.

Mike also talked about the limits of what vibe coding can handle in PPC, specifically. He’s been trying for 20 years to build the ultimate flowchart for Google Ads optimization—if impression share is X, and campaign type is Y, then do Z. He’s never been able to finish it because the last 10% is pure intuition, tacit knowledge that can’t be written down in an SOP.

AI struggles with that same 10%. It can handle the clear, rule-based stuff—download a search query report, format it, and flag terms with zero conversions over $100. But understanding the nuance of why a keyword might exist in three different ad groups for strategic reasons? That’s still human territory.

The practical advice from both experts here is to use vibe coding to build tools that help you think, organize data, and automate repetitive tasks. But don’t use it to make autonomous decisions in high-stakes environments unless you’ve had a developer review the code and you’ve built in human approval steps.

Using AI as a marketing assistant, not a decision-maker

Mike’s philosophy on AI is consistent: it’s a thinking tool, not a replacement for judgment.

Mike explains, “I think one of the biggest ones is AI as a thinking tool. Now, as soon as you say the word thinking, that implies work to a lot of people. I don’t like thinking. I just want to get stuff done. I love thinking. I love going for a walk and brainstorming. And now I get to chat to an AI with that brainstorm to push me into areas of that conversation that I wouldn’t have otherwise gotten to.”

He described a specific use case: grabbing transcripts from multiple YouTube videos, throwing them into Claude, and asking it to connect ideas and suggest solutions based on that content. It’s not perfect. It won’t give you a final answer you can blindly execute. But it helps you think through problems faster and in directions you might not have considered on your own.

Fred compared this to Clay Bavor’s point: when AI fails, it’s almost never because the model is bad. It’s because you didn’t ground it with enough context or give it clear enough instructions.

Mike agreed and extended the idea. AI will try really hard to do what you asked, even if what you asked doesn’t make sense. It’s more interested in making you happy than doing the right thing.

That’s why the “ask it first” strategy works so well. Before you do a task—whether it’s reviewing search terms, adjusting bids, or analyzing a performance drop—have a quick conversation with your AI. Show it a screenshot of the data. Explain what the client asked for. Ask, “How should I be thinking about this?”

It might just confirm what you already know. That’s fine. But often it will say, “Have you thought about this?” or “Did you check for this reason?” or “Could this keyword be in the account three times for a strategic reason?”

The first few times you do this, it takes longer. But once you build the habit and refine your prompts, that thing that used to take 30 minutes now takes 10—and you’ve thought through it more carefully.

How better questions help get better outputs

Throughout the conversation, Mike kept circling back to one idea: the quality of the output depends on the quality of the input. The tool doesn’t matter nearly as much as how you use it.

Mike adds, “I think the skill is asking better questions because as AIs get smarter and smarter and smarter, the answers get better and faster and cheaper. We need to ask better questions.”

Mike shared an example from one of his coaching calls. A client—a business owner running a personal training gym—used ChatGPT to write ad headlines. The headlines were fine, technically. They were grammatically correct and positive. But they were also completely generic.

Mike adds, “I said to her, I was like, literally any personal training gym in the entire world could use these exact headlines, which is why they’re not the right headlines for you.”

The problem wasn’t ChatGPT, but the prompt. The client just asked for headlines about personal training. So ChatGPT gave her the average of all personal training headlines ever written. To get better headlines, she needed to feed the AI information about what makes her gym unique, customer reviews, her landing pages, and insights from competitor research.

That’s what separates useful AI output from generic garbage: specificity, constraints, and context.

Mike also pointed out that AI often knows how to ask clarifying questions if you set it up right.

“That’s exactly how I like to use LLMs for generating Google Ads scripts. First, to basically convince the LLM that it needs to ask a lot of questions about the requirements because we’re not aware of the requirements most of the time, right? We don’t know exactly what we want,” said Mike.

When the AI starts asking questions, it forces you to think through edge cases and details you hadn’t considered. Then it generates pseudo code—a clear, explicit definition of what it’s going to build. That becomes a specification you can hand to a developer, or it becomes the foundation for iterating with the AI until you have something that works.

Fred added that this process is valuable even if you eventually hand the project to a developer. By going through vibe coding first, you discover layers of complexity you didn’t know existed. A “simple” keyword tool suddenly has to account for existing keywords, targeting overlap, budget splits, and ad group organization. And it’s better to discover that in a few vibe coding sessions than 36 months into a waterfall development cycle.

The underrated skill: Understanding the actual business

When Fred asked what one skill is currently underrated in paid ads, Mike said business skills.

Mike elaborates, “I think business skills, business context. I think what I always sort of had in the back of my mind for WebSavvy was to create McKinsey for the small businesses that couldn’t afford McKinsey.”

His point was that technical Google Ads knowledge is table stakes. Everyone has access to the same automation, tools, and platform. What differentiates you is understanding the client’s business deeply enough to know why they’re asking for something, what they really need, and how else you could help.

Most of the wins in Google Ads happen outside the account—landing pages, offers, how the phone gets answered, and the sales process. Understanding more about marketing, advertising, different channels, and the client’s overall business strategy makes you far more valuable than being able to manually set bids really well.

Mike summed it up:

“I think the skill is asking better questions because as AIs get smarter and smarter and smarter, the answers get better and faster and cheaper. We need to ask better questions.”

Beyond business skills, Mike emphasized the importance of understanding how things work under the hood—not to become a developer, but so you have options when something breaks.

“Something I do notice with some newer Google Ads practitioners who’ve grown up in the world where it’s AI first, you know what I mean? They’re never touching the exact match. They’re never touching manual bidding. They’re all in on the automation. And I love that. I’m not saying that’s a bad thing. But then the challen,ge i,s like not knowing what that actually means under the hood,” said Mike.

The final skill Mike emphasized is iteration. Don’t sit down at a whiteboard and try to plan everything out. That’s the old waterfall method. Instead, build something small, see what happens, adjust, and keep moving forward. The tools are fast enough now that planning everything upfront is actually slower than just building and iterating.

“The worst thing you can do with AI today, I think, is to sit down at the whiteboard and try to plan everything out. This is exactly what I’m going to use it for because you don’t know what you don’t know yet,” Mike adds.

How to build Lego blocks to reuse across projects

One of the challenges with vibe coding is that it too often becomes one-off projects. You build something that works for a specific client or campaign, but it doesn’t translate easily to other situations. Gabriele Benedetti, who runs conferences in Italy, asked Mike how to avoid this trap.

Mike’s answer was honest: he’s made the same mistake, “I’ve done exactly the same thing that Gabriele did, which is to build something, and it’s just sort of it works once for this specific thing, but there’s really not much about it, or you have to pull the whole thing apart to use it somewhere else.”

His solution comes from thinking in terms of skills, like small, reusable packages that combine a prompt, some examples, and maybe a bit of code. In Claude Code, you can create these as modular pieces, then chain them together for different projects.

But Mike also offered a counterintuitive take: it doesn’t matter as much as you think. “On one hand, it doesn’t matter that you built this thing, and it only works one way because it is so fast now to say, ‘Hey, go look at this thing. See how I built it. Now build it for this thing over here. This is different. This is different.’ Again, all of the context, the constraints, the preferences, the guidelines, and so on. Now go rebuild it. And at this point, it probably takes two minutes to build the new thing,” Mike explains.

That said, if you do want reusability, Mike suggested drawing out the workflow as boxes on a whiteboard. Each box is a step: grab data from here, clean it, filter it, calculate this, output that. Those boxes become your Lego blocks. Then you tell the AI, “Build this in a way that makes it easy to branch off and build something similar but different in the future.”

Fred added that in the age of on-demand software, you don’t need to build one generic tool for everyone. You can build a tool for one vertical, then adapt it for another vertical by explaining the differences. That level of customization used to require SaaS-level planning. Now it’s trivial.

Why is vibe coding addictive once you start

Both Mike and Fred admitted that vibe coding has a compulsive quality once you experience what’s possible.

Fred said, “I can’t tell you the number of nights that I’ve been up past midnight, vibe coding, something I haven’t done in a long time because we got a big engineering team at Optmyzr. But it’s fun. It’s addictive. It’s productive.”

Mike agreed. The hardest part isn’t building; it’s stopping.

He described a pattern that’s become routine: he runs three or four AIs in parallel because each one takes a couple of minutes to respond. A year ago, he’d start them working, then watch YouTube or American football highlights while waiting. Now they all finish in under two minutes.

The speed creates a weird problem: unlimited opportunity. Things that used to be “cool ideas but I don’t have time” are now “why not do this right now?” Mike described it as both the most exciting and most stressful time in his life.

Fred related to this. The opportunity space has expanded to an almost infinite degree. Every idea that’s been sitting on a Trello board for years suddenly feels doable. That’s exhilarating, but it’s also overwhelming because now the constraint isn’t capability—it’s deciding what actually matters.

The gap between “I should build this” and “it’s done” has collapsed to hours or minutes. For people who love building things—and both Mike and Fred clearly do—that’s intoxicating.

The danger, as Mike noted, is losing focus. It’s easy to go down rabbit holes, like spending 20 minutes tweaking button colors instead of working on the core functionality that actually matters.

His advice? Keep a prioritized list on paper. Focus on the big things first. Save the aesthetic details for later. Otherwise, you’ll build a lot of polished features that don’t matter while ignoring the core functionality that does.

Mike’s plea to stop overthinking and build something

As the conversation wrapped up, Mike’s advice for marketers came down to a few core principles.

Principle #1: Just start

Don’t overthink it. Don’t try to plan everything. Build something small and see what happens.

“I’m pleading with people all the time. I know you don’t know exactly what you want to build. Have a rough idea of the direction you’re going in. Just build version one. Just get some sort of proof of concept, minimum viable, just get it to grab some data from there and put it in there,” says Mike.

Mike emphasized that vibe coding only looks scary for the first five or ten minutes. It’s built by geeks with terrible UI, so it seems intimidating. But fundamentally, it’s just a chatbot. You talk to it, it builds stuff, you check if it works, and you iterate.

Principle #2: Don’t expect AI to be perfect

Mike explains, “I always say remember, it is artificially intelligent. I think we always focus on the ‘so intelligent’ part that we forget about the artificial part.”

AI will do exactly what you tell it, not what you meant. When it doesn’t work, the question isn’t “Why is the AI so stupid?” The question is “What did I tell it that led it here?”

Principle #3: Trust but verify

Mike described himself as a “trust but verify” kind of person. Give the automation clear directions, such as solid conversion tracking, a clear bid strategy, and a proper campaign structure. Then check the reports. See what it’s actually doing. If it’s not doing what you want, close the gap by refining your directions.

Principle #4: Keep humans in the loop

For anything high-stakes—tools that touch client accounts, scripts that adjust bids, automation that modifies budgets—have a developer review the code. Use vibe coding for internal tools, low-risk prototypes, and as a thinking partner. But don’t deploy autonomous systems in production without human oversight.

Principle #5: Embrace the shift from control to guidance

The era of controlling every little thing is over. Consumer behavior changed. The SERPs changed. AI changed. The new job is to guide smart systems by giving them the right data, constraints, and goals, then verify what they did and adjust.

Mike’s parting message was simple: go build something. Start with a small tool. Try the scary challenge at 8020agent for a guided introduction to Google Ads scripts. Open Claude and ask it to build a working snake game just to see how it works. Build something fun, not just PPC-related, to get comfortable with the process.

“We built Asteroids two Christmases ago in half an hour, a fully working game of Asteroids that my daughter absolutely loves. Can we play that game again, Dad? Like, don’t forget the Google Ads bit. Build something fun, but just build. Just see what it can do,” Mike adds.

The tools exist. The capability is real. The only question is whether you’re willing to step out of your comfort zone and try.

Episode Transcript

Frederick Vallaeys: Hello, and welcome to another episode of PPC Town Hall. My name is Fred Vallaeys. I’m your host for this show. I’m also the CEO and co-founder at Optmyzr, a PPC management tool. For today’s episode, we have a returning guest, but someone we haven’t talked to in quite a while, Mike Rhodes. He’s done many things in the PPC space, but currently, he runs Ads to AI, and he has been very active building AI vibe-coded tools. He’s very well known for his PMax script that many people use. So I’m very excited to bring Mike onto the show and just talk about how AI makes his life easier and what we can all learn from that. I know there’s a lot of AI skeptics out there, but Mike definitely seems to be one of the people who’s making the most of it. So let’s see what we can learn from him, and let’s get rolling with this episode. Mike, thanks for coming on to the show.

Mike Rhodes: Thanks for inviting me back. It’s wonderful to be here. Good to talk to you again.

Frederick Vallaeys: Yeah. Hey, since we last spoke, tell people a little bit about what you’ve been up to, and also, for those who haven’t met you before, who is Mike Rhodes?

Mike Rhodes: Oh my goodness. So for 17 years of my life, I built and ran an agency here in Australia. I loved that. And then a number of things came together. One was realizing that I had 134 Mondays left to walk my daughter to primary school. We count down every week. Some days I get two days, but at least every Monday we’re down to 40 now.

But that realization, and just seeing what was coming with AI, I got an offer for my agency. So I took that a couple two and a half years ago and just got incredibly lucky once again. Right place, right time. And I have been building. I had just relearned to code before that. Not to your level, but went back and taught myself code probably four or five years ago now.

I had a coding mentor that I would talk to on the weekends, and all of a sudden, GPT-4 arrived, and that sort of level of curiosity was able to be almost all my questions got answered instantly. How do we do this? Why are we doing it this way? Is there a different way we could do it? And that just sort of led me to play around.

I had this tool that I’d played with since, I remember it’s 2018, because I remember Ralph interviewing me on his podcast at Traffic and Conversion in 2018. This thing called the Profit Curve, and it was this tool I had inside of a Google Sheet that I loved, but many of my team didn’t even know it existed. For others, it was too hard to use or too hard to explain to a client. It was designed to solve that problem of how much should we be spending this month. This is a long answer to a short question.

But I built this tool out, and it was just this moment of like ah, and I was doing that pretty manually back then. This is kind of before vibe coding existed. But that led me to then go, well, okay then, what can we build? So about a year and a half ago, I built out this tool that’s still alive. It’s still free. It’s called 8020 Agent. It does a fraction of what something like Optmyzr does.

But it just helps you see your data in different and interesting ways. It can be a useful tool. It’s a bit clunky at times, hence the name 8020. It’s not polished and finished. It does most of it. And then that just kind of led me to more and more things. Then, people would start asking me, well, how do I build something like this? Then, I ran some courses last year called Build the Agent.

And at the end of that first cohort, a lot of people were like, what’s next? This was great, but this was only eight weeks. Like, we want someone to hold our hand and run. What I’ve always loved doing is running four or five hills into the future. And then coming back and not just reporting on what I’ve seen, but giving people useful, practical tools.

And it’s funny, I see this arc now where I’ve always done this. I built the display grid way back in whatever that was, 2016, 2017, remarketing tools. Like I’ve always done this along the journey. I just didn’t realize it at the time.

And so I created Ads to AI. So it’s a community. I sort of curate and provide content around AI and try to remove most of the noise and just give as much signal as possible. And just because I constantly spend six months in the future running over those hills, I keep looking around and finding these fun things like 8020 Brain, which I’m sure we’ll talk about, and just building it because I love to learn.

But I’ve worked out that learning stuff without making it useful, without teaching it, without turning it into something, learning stuff just for learning it doesn’t do much for me. So, I need to teach. I need to share it.

The first thing I did when I found Google Ads in 2004 was to rush back and tell everybody I knew about this amazing thing called Google Ads. While you were toiling away in a cubicle in Mountain View somewhere, I was trying to, like, tell everybody that this is amazing. This is like stopping yellow pages and radio ads, like this is the thing.

And it took me about a year and a half of people saying to me, “Mate, I don’t care how it bloody works. Can you just do it for me?” And I had to hear that a lot before I fell into starting an agency.

And it’s kind of been the same again. I’m like, “This thing is amazing. This is going to change everything.” And you’re right, there’s a lot of AI skeptics, and there’s a lot of things AI shouldn’t be used for, but there’s so many amazing things that it can help us do, not replace humans, replace tasks, and help us with many other tasks.

And I just, I’m just, I’ve always been curious, and this is just, God, what a time to be a curious learner because you can get answers to every question and build stuff which is the other thing I’ve always done. Lego blocks, model airplanes. I’ve always loved building stuff, and now you just sort of go from idea to building to impact sometimes in like 10 minutes. It’s ridiculous. Having so much fun.

Frederick Vallaeys: It’s pretty crazy. And now you seem as excited as I am about the ability to build stuff very quickly. But you mentioned an interesting point there where you’re running one hill, but you’re already looking four or five hills ahead. So let’s maybe delve into that a little bit deeper. And what are those hills that we’re going to be looking at in the next couple of months and years?

Mike Rhodes: I just think Andrew Ng said this, and you know Andrew because you were at Google, but Andrew Ng is like the smartest person that most people have never heard of. Founder of Coursera, basically the guy that introduced AI into Google back in the Translate days, 2010 by now, deep learning. He said back in 2017, AI is going to be like electricity. It is just going to be ubiquitous. It’s going to be in everything. It’s going to be everywhere.

That’s how I started my talk at Traffic and Conversion was with his quote and talking about how electricity rolled out and how many decades it took for us to use electricity, and at the time lots of people went oh yeah this is going to be big but how we used the tool was not obvious, and I think it’s very very similar this time. There are so many different ways this can be used. It’s going to take time to be adopted, especially in enterprise.

But it is changing everything. It’s just that most people don’t have the time to put down what they’re doing because they’re running at 110%, redlining, and I’ve got all this client stuff, and I’ve got to deliver it, and I’ve got to get it done. I don’t have time to go and watch a two-hour YouTube video talking about the latest stuff. That’s the stuff I love to do, I get to spend a ton of time out there and then just bring back the fun stuff.

What specifically? I think one of the biggest ones is AI as a thinking tool. Now, as soon as you say the word thinking, that implies work to a lot of people. I don’t like thinking. I just want to get stuff done. I love thinking. I love going for a walk and brainstorming. And now I get to chat to an AI with that brainstorm to push me into areas of that conversation that I wouldn’t have otherwise got to, and to get there faster and to help me find the signal in the noise.

A great example of that is YouTube. I used to read a lot of books. I got about a thousand books over there. I’ve read most of them, not all. And I realized at some point in the last two or three years that a lot of that was just in case of learning. You know, you read a book about business. It’s kind of like just in case I have a problem that this thing solves, and maybe when I have this problem in 18 months time, I’ll be able to remember some of the stuff in this book about what Seth Godin told me, and I’ll be able to put it into action.

And I’ve moved to more sort of like just in time learning where I’ll go to YouTube, or I’ll read an email newsletter,r or Substack,k or something, and I’ve got this specific problem. Let me go search for an answer to that problem. I think that’s probably what most of us do.

Now, I could just do that at scale. I can grab transcripts from a bunch of different YouTube videos, stick them all into an AI, and say like, “Here’s my specific problem. Help me think through how all of this information might help me.”

It’s not perfect. It’s not going to give me an answer, and I’m going to go, yes, I’ll just brush off blindly and do that. But it helps me think through, and I think that’s one of the biggest things that I see AI being used for. I’m not sure if I’ve answered your question there.

Frederick Vallaeys: Yeah. No, I see it as a very smart partner in business who you can ask questions and brainstorm. I would certainly see that as well. You recently…go ahead.

Mike Rhodes: Sorry. No, go ahead.

Frederick Vallaeys: I was going to say because it has such incredible knowledge, right? It has basically read everything that’s ever been written. Whether that’s a good or a bad thing we’ll leave for another day, but it has all of this knowledge built in. It’s kind of how you extract the useful bits of that for you in your context. And how you push the AI into those sort of unexpected areas. Garbage in, garbage out.

Mike Rhodes: If you, I liken it to a solar system, right? There’s this huge, big planet in the middle called planet average because let’s face it, the AI has been trained on Reddit and Twitter. That wasn’t a very good idea, was it? But it’s sort of like the average amalgamation of everything. And if you ask a quick little question, you get an answer from planet average. You have to really push it that there are millions of planets in this weird solar system that I’ve got a picture of in my head.

And by asking for a better prompt, by giving it better context, by setting up the AI, you’re sort of able to push it to this far distant planet in this weird universe and get an answer that there’s an answer on every single planet.

And how you ask the question determines which planet the AI goes to to answer your question. And the more you’re able to push it and steer it and say, like that corner of the universe somewhere over there, there’s a really good planet with a really good answer on it, nudge it over there rather than just getting an answer from planet average. That makes a huge difference in how much benefit you get from the AI.

Frederick Vallaeys: Yeah, that’s an interesting way to visualize the vector space because that’s basically the pictures that people see, right? It’s all these dots in a three-dimensional space, but the vector spaces are many more dimensional than that. So, it’s a very limited representation.

But the whole notion of what you’re talking about is what we also see is the more specific you get into your request and prompt, you basically limit the box in which the AI has to respond, and you point it in a direction of one of those boxes. And so you get much better answers, and usually we think about that the other way, right? We’re like, well, we don’t want to put too many constraints on a human because the human is going to struggle with answering things in the voice of, say Albert Einstein, but for the AI, that’s no problem. I mean it just goes to that planet or that piece of the vector space, and it gives you the answer that actually makes a lot more sense based on what you were specifically looking for.

Mike Rhodes: I never thought of it in those terms, but I’ve always been a fan of the power of positive constraints. Like artists will tell you this, songwriters will tell you this, painters of like, paint me a picture. Ah, okay, paint me a picture about this very specific thing and this specific style and make the picture this big. Like all of those constraints actually help guide the artist. So maybe it’s a bit of that in my thinking.

Frederick Vallaeys: Yeah, makes total sense. And so, a specific PPC example would be something where you say, “Hey, listen. I’m writing some new ads or creative assets, but who are you writing them for?” And oftentimes we think in terms of those averages. We start with the keyword, and everything’s about the keyword, right? Well, okay, if we’re selling this, then let’s mention this in the ad.

But now with this creative partner that has all of this knowledge, why not write the ad for someone who lives in Paris or someone who lives in New York, because they will care about certain different things. They talk about different things, and this LLM will know that, and so it pushes the creativity to a new level, and in doing so makes things more relevant to the person who’s seeing the ad, and that will boost your quality score which in turn makes your CPC cheaper, and we’re back to the old fundamentals of how Google Ads works.

Mike Rhodes: Fundamentals aren’t changing. But yeah, it’s hard for us to get out of our own little bubbles sometimes. And even if you don’t use those ideas from the AI because you look at them and go, “Ah, that’s rubbish. That won’t work.” But every now and again, maybe you want to test one of those.

But yeah, it’s very hard for me to get into the head of a 22-year-old woman that lives in Paris. Like, I just don’t know what she’s thinking about. But yeah, so it’s just more grist for the mill, right? It’s more ideas. You don’t have to use it, but it’s often that sort of thought starter that prompts you to push you to a different planet into different places, and look at things from different angles, and that change of perspective makes you a better marketer or a better business person, frankly.

Frederick Vallaeys: Yeah. So, staying on planet Earth about leveraging the many smart people that live on this planet. You recently built a tool that had quite a bit of traction on LinkedIn. And I am fortunate to be one of the people in that tool. But you basically built a council of experts. Tell us a bit about what this tool is and how that may connect to what we were talking about here.

Mike Rhodes: So, it kind of goes back to the YouTube transcript thing I was talking about before, where I found myself more and more listening to podcasts, watching YouTube, and they were just so dense, especially when I’m walking. There’s a ton of stuff. I was listening to a wonderful podcast by Rory Sutherland. You’ve probably read his book, Alchemy. It’s a wonderful book, but there was just so much in there.

I sort of needed to pause for a moment. It’s like almost, God, I wish I still walked around holding my Moleskine, which I hardly ever take anywhere now, and write all these ideas down. And I jotted down a couple of notes, voice notes. When I got back, I dug out the video, found the transcript, threw it into Claude, my preferred AI, and said, “This is the thing I was thinking about, and here’s the transcript. What are all of the ways that these sorts of ideas connect? How would Rory solve this problem?” And it was really pretty good.

So then I just sort of took it a little step further, and I went, “Well, actually, there’s a whole bunch of really smart business people like Seth Godin that I would love to ask this same question to.” And so I did that. I sort of collected a bunch of information about them and asked them, but they’re all a bit different.

I’m like, and then I found myself cutting and pasting conversations and like, “Hey, this is what Rory said, and this is what Seth said. Could you…” That was just a lot of cutting and pasting and a bit of a mess.

So then I just thought, well, okay, Claude, you know lots about all these people. You’ve read all of their stuff. Pretend there’s like five of them in a room. Here’s the question. What do you think they’d say? And it was, you know, it was okay, but it was a bit messy.

And then I remembered this book that I read, God, 25 years ago, probably, this thinking tool from a guy called Edward de Bono, one of the best thinkers on the planet, I reckon, called Six Thinking Hats. And I know that he’s taught this everywhere, from like eight-year-olds to 88-year-olds. He’s taught this in primary schools. He’s had company boards do it. It’s a way of putting on your metaphorical hat, the six different colors. Just what we were talking about before that forces you into different perspectives. What are all the risks associated with this? What are all the good things that could happen? What’s my gut say? These six different ways of doing it. And I just smooshed the two together.

Six Thinking Hats is really good because it gives everybody around the table, I’ve used it in management meetings and stuff like that. It gives everybody around the table a voice so the quiet person in the corner actually gets heard, and the guy, it’s usually a guy who talks all the time, doesn’t take over the conversation and make everybody else just roll their eyes and want to leave the meeting.

And so the facilitator says, “Right, everybody put on your black hat now. We’re going to look at the risks and why this won’t work.” And you go around the table, and everybody gets a voice. And so I just, it was this iterative process of smooshing these things together. And I thought, “Oh, this would be fun to do with Google Ads experts.”

So I said to Claude, “Right, there are a few people I’ve got in mind, but who would you recommend? If we were to pick 30 people, who would you recommend?” Of course, number one, PPC Influencer, two years in a row. Of course, it gave me your name and a whole bunch of other names and a few names that I hadn’t heard of. And I said, I really want this person to be on. And so we ended up building out a whole bunch of personas, just publicly available information.

I’m not using anything private. I’m not using the contents of Kirk’s wonderful books, for instance, because maybe not everybody has bought that book, and therefore they shouldn’t benefit from that private information. You need to go buy the book if you want to benefit from that. But the freely available information that’s out there. A couple of podcast transcripts and whatever else, blog posts and so on. Built these personas, threw them all together. I’ve got a Google Ads one, a marketing one, and a business one.

And the first test, I remember the first test I did with this. Threw an idea in there. And it came back and went, “Mike, that’s a really good idea.” I went, “Yeah, okay. Here we go.” But don’t do that now. It’s the wrong time to do this. I was like, “What? An AI that just said, “No, it’s not just blowing smoke. What?”

But the weird thing was, it sounded just like the two smartest women in my life, my wife and my old general manager. When I had the agency, I had this really, really wonderful general manager, who I occasionally referred to as my business wife, but that’s just a bit weird. So, we’ll call her general manager, Zoe, and my wife, Gabby. And it just sounded exactly like what these two really smart people would say. Mike, that’s a really good idea, but no, not right now. And then it went on to explain why. And I was just like, this is amazing.

And I just started throwing other questions at it. Built out the councils. I think I’ve got about 60 different experts on there now. Seth Godin, Rory Sutherland. It’s just, it’s fun. And then yes, so I built the little tool with the Google Ads one, created little avatars because I’ve learned, and you know this, but show don’t tell. It’s lovely running inside of an AI, but it’s all text-based. So I just made a little, some pretty pictures, and just a little demonstration of it, and put it on that 8020 Brain website just to visually show the idea.

And yeah, it kind of, most people love it. A couple of people hate it. So it’s probably a good thing.

Frederick Vallaeys: Okay, good. Well, people are using it and giving feedback. That’s amazing. Now, let’s talk a little bit about how you created that, right? So, you said you’re not a developer by educational background necessarily, but you took this whole concept of something that was working really well on AI, but was very tedious, manual. You had to go back and forth, copy and paste a bunch of prompts.

Now you’ve taken it to the level where anyone can go and sign up for 8020 Brain. They can play with this, and it’s basically a piece of software. So, how did you go from not knowing how to code to producing something that is an amazing tool?

Mike Rhodes: We’re going to need a bigger boat. The short answer is Claude Code. The longer answer is probably bit by bit of just, oh, like building this, the Profit Curve that was one page and lots of cutting and pasting, and I was the glue holding all of those bits together. I would get some HTML and some JavaScript from over there, some CSS, and smoosh all that together and build a page.

The AI coding tools are very intimidating if you’ve never learned to code. Google Ads scripts, for the last two, three years, I’ve been teaching people about Google Ads scripts. I’ve got my Scripts and Sheets Mastery course. We built the Agent.

The name from Ads to AI, I kind of see as this bridge that I want as many freelancers and agencies as possible to like come on, let’s get over the other side of this bridge because I don’t think just offering Google Ads management services two or three years from now is going to be enough. For a few, it will be, but for the person that’s doing an average job of that, it’s not going to be enough. So, we need some other skills, some other capabilities.

Google Ads scripts are the first step on that bridge. And then I had to think, well, okay, I’ve been saying that for a while, but what are all the other steps? What are all the other skills, capabilities, and mental models that we need to go from Ads to AI?

And yet, vibe coding became a thing. What was that like six, 12 months ago? Apparently, that’s what I’ve been doing for the past two or three years: vibe coding, which is essentially talking to an AI quite literally. I talk to it all day, and then it does stuff. I’m not writing any of the code.

I know just enough to look at it and make sure that nothing’s going horribly wrong. But you look at it as much as you want to. I wanted to learn to code. So in the early days, I was looking at everything that it wrote and thinking why is it doing that and asking questions about it. But you’re really at the point now, if you were to start today, you really don’t need to do that at all.

I know plenty of proper software devs that don’t look at the code anymore. Many will look at bits of it, but they’re trying to sort of, the new way of developing. And I don’t know, I don’t know any like proper proper ones who are doing this in development right who work for Google and are building real stuff, but a lot of them are sort of trying to code more and more hands-off and be the orchestra, not the violin player.

You don’t have to learn all of the different instruments and learn all these different coding languages and all of the minutiae. Syntax used to bug me. I could never remember the syntax, but now you’ve got to be the conductor of the orchestra, understand what the thing is that you want to achieve, and describe that goal to the AI, not how to do it. It knows how. You just have to explain the what and the why and let it rip.

And the wonderful thing about code is it either runs or it doesn’t. So, we started doing this with Google Ads scripts. It either runs or it doesn’t. And if it doesn’t, you get the error message, and you chuck it back into the AI and go, “Fix this, please.” And it does, and you go around that loop three or four times, and now you’ve got a working script.

But that automated reporting is just the first step on that bridge. So then we started thinking, well, okay, what next? How do I get AI to help think with that? Now, how do I automate all of that? So, I’m getting automated insights every Monday morning. Okay, what’s the next bit?

The next bit is really adding a bunch of business context. The more information you can add about your client, about their goals, about their strategy. Exactly what you said before, the constraints and the preferences, not just like, oh, the budget’s 20 grand this month, but all of that information. The more you bring to the table, the more you push it away from planet average and get to something really interesting.

And then you just keep iterating. The worst thing you can do with AI today, I think, is to sit down at the whiteboard and try and plan everything out. This is exactly what I’m going to use it for because you don’t know what you don’t know yet. So, you just start, build something really little, a budget pacing tool or something. And then you iterate and go, “Oh, wow. Well, that was easy. What’s the next step on the bridge? What’s the next step?” And we just keep walking across the bridge and keep playing.

I’m really not sure if I’ve answered the question again. And I’ve wandered off on tangents here, but I am absolutely loving this ability of it to just, and to help it. It also pushes back a bit and guides you along this path, and says if you ask it what should be the next step here, it’ll go well here are four directions we could go in. What do you think?

Frederick Vallaeys: Yeah, the question was how did you go from non-developer to producing software? So I think you did answer it. I’m curious about some of the pitfalls, though. So some of the pitfalls that we’ve seen is if the code is not initially well written and it fails and you persist in telling the AI can you please fix it like this is the expected outcome, eventually it’ll just put an if-then loop which says okay if Fred is asking for it then always give this answer because that’s what’s going to make him happy. So you do have to be a little careful there.

And then the other challenge that we see is sort of in the nuance of PPC. So, when it comes to, I keep giving this example maybe too much, but like if you suggest some keywords, okay, here’s some keywords, but are these duplicate keywords for your account? And that is not just simple , it is the same text in one place as the other, but it’s also the campaigns geo-targeting, day parting, same locations, same times. Are there budget constraints on some campaigns that might make them actually not really competing?

So there’s all this nuance, and so that’s where the trick for us is, how do you combine flexible code and AI and sort of all of this new stuff with the more deterministic, which is like very hard and set rules that keep you within safe boundaries, but then start doing more innovative things within these boundaries. I’m kind of curious if you’ve run into any funny instances of where things didn’t go exactly as planned.

Mike Rhodes: Oh, every day. Yeah. Like if something isn’t working and less so today than even four months ago, but if something isn’t working, it’ll just delete a chunk of code. Like you asked me to write a test to make sure that this was working. It didn’t work. So, I’ve deleted the bit that wasn’t working, and now all the tests pass. It can be a really, really dumb intern some days.

But on the nuance part, I totally agree, and I think that’s why there’s a level of ambition, right? There’s a scale between, I don’t know, skeptic and complete like AI is going to do everything. I think you want to kind of aim for somewhere in the middle. Asking it to do everything inside of a Google Ads account, just, yeah, that’s a recipe for disaster.

For close on 20 years, I’ve been trying to build the ultimate flowchart of if this, then that, and if impression share is this, and this type of campaign, and I’ve never been able to do it. I’ve taught, I don’t know, tens of thousands of people how to do Google Ads at this point, but I was never able to create the ultimate, ultimate guide.

And then I found this thing. If you came across, God, what’s it called? The Dreyfus method. Nat Eliason has a wonderful blog post about this. Just search Nat Eliason Dreyfus method. There’s two images in there that changed everything for me about how I learn but also how I teach. And it’s about going from novice to expert. And that last sort of 10% is really all about intuition. You cannot put it into a flowchart. It’s the tacit knowledge. It’s the learning by doing and the experience.

There will come a point I suspect, where AI might be able to do a lot more of that. But the problem with tacit knowledge is it’s really, really hard to get out of your head and into an SOP. And the AI needs input.

If you can write it as an SOP, many agencies will have, like a VA,maybe somewhere in the world, helping them do stuff, following along a checklist, creating a search query report, or something. Download it from here, and we put this heading, and we change the column sizes. That, no problem for an AI.

But understanding the nuance between this keyword in this location and all the stuff that you just said, don’t use an AI for that stuff yet. Have it make suggestions, sure, but the human needs to stay in the loop. The human needs to be that sort of doing that last bi,t and probably a few other bits along the way. I’m not saying AI should be doing 98% of what all ad professionals are doing, but it should not be, try, we should not be trying to ask it to do 100%. But we can ask it first.

One great tip, there was a question I saw on the LinkedIn post about sort of future skills and so on. I think one of the best things you can do right now is, before you do the task, have a chat with your friendly AI and say, “This is the thing I’m about to do. I’m going to do this, and the client’s asked me to do this, and this is the situation.” Maybe even show it a bit of data. You can take screenshots and throw those in and say, “How should I be thinking about this?”

Not try and do it for me, but how should I be thinking about this? And it might just tell you what you already know. That’s okay. But often it will say, “But have you thought about this?” Or, “Did you look for this reason for that drop in performance there?” Or, have you considered that this keyword might be in the account three times for a reason, because there are different geographies or bid strategies, or whatever it might be.

Using it for that little extra bit of thinking which the first few times you do that it takes twice as long so no one does it because like I’ve just got to get this thing done, but if you can get in the habit of doing that then you start seeing the shortcuts of oh actually when I do this and I package it up in this way and I remember to do it this way actually that thing that used to take half an hour now only takes me 10 minutes every time. I’m going to save time, but I’m going to have to invest a bunch of time up front, which is why it makes it so hard.

Frederick Vallaeys: Yeah. And it’s back to your example about using it as a thinking partner. And one of my friends from Google, Clay Bavor, runs Sierra AI at this point. And he, him and Bret Taylor were making the point that when AI fails, it is usually not due to a shortcoming of AI. It is due to you not having been specific enough and having grounded it enough with context that it needs to do a good job. And I think that’s very similar to the point that you’re making.

Mike Rhodes: Yeah. You need to think with it first to answer the questions that should be answered rather than just letting it go and do its thing, because then it might make decisions that don’t align with how you thought about that problem and how it should be solved, because it’s a Labrador puppy, right? It is going to try really, really hard to please you, and it’s not; it’s more interested in pleasing you and doing what you’ve asked for than it is about doing the right thing.

So if you ask it to do the thing, it’ll have its best crack at actually doing that. But yeah, maybe it doesn’t have, you haven’t given it enough context because that is all in your head.

I was chatting to someone yesterday about how they choose negative keywords of all things. I’m like, you have an algorithm running in your head every time you look at that list of search terms. You don’t know it because it’s just second nature to you because you’ve done this so often, but there is some sort of algorithm running in your head. You are thinking about all of these different things and comparing them. You’re just doing that so fast and so intuitively that you don’t know that you’re doing it.

But try taking that algorithm out of your head and writing an SOP. That is sometimes virtually impossible to do because there’s so much nuance. But you can get the first 80% done. 8020. Quick plug. You get the first 80% done with an AI a lot of the time. But yeah, you, I really agree with that. I haven’t heard it said that way before, but I like that. It’s not usually a shortcoming of the model. It’s how you’re using the model. But usually, context.

Frederick Vallaeys: Context, I think, is basically everything. Garbage in, garbage out. Right. That’s how we learned how to code a long time ago. That was the first thing you learned. Garbage in, garbage out.

Mike Rhodes: Exactly. Yeah, giving it the context, which is basically grounding the model in the reality that is your truth, right? Which is often what are the actual numbers in a Google Ads campaign because we’ve all seen examples where you ask it for budget predictions, and it just makes stuff up, and it’s like, well, l how was it supposed to do this job if you didn’t tell it what’s actually going on?

Even little things like telling it what today’s date is, and you’re like, he,y we’re currently seeing some, like, there’s a lot of searches for roses. What’s happening? It may not realize we’re coming up on Valentine’s Day. And so grounding it in the fact that hey, we’re two weeks away or four weeks away from Valentine’s Day, that would be helpful. That is a piece of grounding that helps it better answer the question that you have.

I go back to my universe picture in my head, like your first job is just to push it into the corner of the universe where a decent answer lives. Again, there is an answer on every planet. And that’s how I think of grounding. You’re just sort of giving it the facts, which in Six Thinking Hats is the white hat. You start by giving the council the facts.

It’s just that now with the brain and so on, it’s able to pull all of the data from the Google Ads account, look at the campaign structure, pull all of that. So you don’t have to do a lot of that manually. You don’t have to be the glue in between all of those steps necessarily anymore. Start there. Start with the facts, and then say, “Right, this is what I’ve got.”

Frederick Vallaeys: Yeah. Grounding is the proper way to call it certainly.

Mike Rhodes: Yeah. Anyway, just the technical terms.

Frederick Vallaeys: But yeah, let’s just have it make sense for the viewers. So, thanks for absolutely your planet’s universe. So Mike, let’s talk about another question. So this was from Gabriele Benedetti who runs conferences in Italy, a good friend of ours. He’s asking how to be good about reusing elements.

So you developed a very popular PMax script. You’ve now built this council of experts in the 8020 Brain. What Gabriele is basically saying is like he also builds these solutions but often finds that they become standalone and not interconnected. Have you found good ways to turn this into a Lego block sort of system where you can leverage one thing into another tool that you’re building?

Mike Rhodes: I remember saying (gosh) in a talk like seven, eight years ago, like one day AI will be a whole bunch of Lego blocks, and we’ll just go to the shelf and pick off Lego blocks and put them together. We’re not there yet. We are kind of there now eight years later.

I’ve done exactly the same thing that Gabriele did to build something, and it’s just sort of works once for this specific thing, but there’s really not much about it, or you have to pull the whole thing apart to use it somewhere else.

How would I solve that? I think in terms of the smallest individual useful bits. So, I don’t know how to describe this without going down a huge rabbit hole and talking about skills, but a skill, all we need to know about for now is a skill in Claude Code terms, which is an AI. It’s just a package of things. It’s a package of a prompt and some examples, maybe even a little bit of code that explains to the AI, this is the way I want you to do it. And then once you have a few of those, you can start to chain those together, put the Lego blocks together.

The good news is you can build anything you like at this point. There is no Lego brick that you can’t think of that it can’t build. The bad news is there is no Lego manual. There isn’t that thing over there that you open up, and it says, “Right, step seven, you put the blue brick on the red brick, and then step eight, you put the window on…” The manual doesn’t exist. There is no manual. The only limitation is your creativity and your imagination. But it can build the Lego bricks for you, and then you get to put the Lego bricks together in different ways.

Gosh, I’ve got so many competing thoughts in my head at the same time. On one hand, it doesn’t matter that you built this thing, and it only works one way because it is so fast now to say, “Hey, go look at this thing. See how I built it. Now build it for this thing over here. This is different. This is different.” Again, all of the context, the constraints, the preferences, the guidelines, and so on. Now go rebuild it. And at this point, it probably takes two minutes to build the new thing.

So, you almost don’t have to think, oh, how do I pull this apart into eight different Lego bricks so that I can use these Lego bricks just in case in the future? You just point it at the thing you built and go, I want one of those, but mixed with this and a bit like that, and it’ll build it.

On the other hand, having those individual pieces, thinking through what are the, I guess in James Clear terms, the atomic habits, like the little pieces that go into this systematic thinking, like draw it out, I suppose, as a workflow, like Nils has done for some of his scripts, like literally draw the boxes out on the whiteboard.

First, we want to do this. We want to grab the data from over there. Then we clean and sort the data because we don’t need 40,000 search terms. We just need the ones that have spent more than $10. Great. Now, what do we do with it? What’s the next step? What’s the next step? Those are essentially the Lego blocks.

And then you have a chat with your friendly AI and say, “This is what I want to build.” And just ask the AI, I want to build this in a way that makes it reusable in the future. I don’t know how yet because I don’t know what the next thing’s going to be. But can you make sure when you build this that it’s easy to branch off and build something similar but different in the future.

It is so good now that even the patterns from six months ago, never mind two years ago, kind of almost don’t come into play now. I was building something yesterday, and I just, I usually have three or four AIs on the go because I got used to like okay you do this and you do, I’m trying to be the conductor of the orchestra.

And I’d go back 12 months ago. I watched a lot of YouTube because you’d get these things running, and then I’ll go watch some American football highlights or something else, or something I’ve been meaning to watch on YouTube for a while, some climbing video. And ding, ding, ding, all of them took less than two minutes. Hang on. I haven’t even got to the end of the first quarter yet, and you’ve all finished. I need to come up with more work for you.

It’s incredibly fast and good now that I mean, that’s almost the hardest part is actually, the hardest part is stopping. That’s what I’ve found because…

Frederick Vallaeys: Go on. Sorry.

Mike Rhodes: Well, stopping is the hardest part, and one thing that I’ve mentioned many times before on the podcast, but it’s the most exciting time in my life, but also the most stressful time in my life because the opportunities have multiplied to an infinite degree. And it’s like all these things where in the past I would have just said, well, it’s kind of a cool thing, but I don’t have time. Let me focus on the higher priority stuff.

Now, like you said, because you have these four AIs running at the same time, and they’re so fast, it’s like, well, should I be doing this, should I be doing that? Like, I’ve been thinking about that for 10 years. Why don’t I do this now? Yeah. And that’s also a little bit of a stress then, right?

Frederick Vallaeys: But I like your whole example of you not thinking about a solution that scales to everything. So, historically, SaaS software has financially made sense as you build something that can be used by lots of people. Yeah. Because that’s what pays the bill. So, to make it work for lots of people, you have to make it at some level generic so that it fits many use cases.

Now, in the age of on-demand software that’s being built by Claude Code or Lovable or Replit or Baseten or who have you, yes, take the thing that you built for one vertical and say, “Hey, here’s the thing, but now we want to do it for a different vertical. How would we change it for them to make it more relevant?” Right? And so that becomes so easy that at some level you don’t have to think about building blocks anymore.

Mike Rhodes: I definitely, I want to add something to that because it’s so, so important. A lot of members of my community have found this quite difficult, and I was surprised that they found it difficult, but I mean how you would have, I imagine, coded with your team back in Google back in the day is you’ll stand around a whiteboard for three months sometimes and build this huge waterfall plan, and it’s all very, and you know before agile came along.

You plan out certainly how Microsoft would have worked 30 years ago. You plan everything out for a year or two when we’re going to do Windows 95, and this is what it looks like, and you, and then you throw that plan over the wall to the geeks who unsurprisingly come up against new questions and new blocks and things that nobody’s thought about because you can’t think about everything, and stuff runs overtime and over budget, and that’s how a lot of project stuff used to work.

And then somebody invented agile, and ah, actually, maybe instead of planning everything out, we’ll plan out the next little bit, and we’ll do that, and we’ll learn a bunch along the way. And once we’ve learned that stuff, then we’ll plan the next bit. And that is by far, with the tools we have now, a better way to go of just iterating.

I’m pleading with people all the time. I know you don’t know exactly what you want to build. Have a rough idea of the direction you’re going in. Just build version one. Just get some sort of proof of concept, minimum viable, just get it to grab some data from there and put it in there. And then when you see it do that, you go, “Oh, can we have a little chart of that data?” Yes, you can build the chart. Oh, well, if I’ve got a little now, can I have a little thing that changes the date range? Let’s do that. And it’s just you just iterate and iterate and iterate.

And then it becomes hard to know when to stop because you have another idea. And it is literally holding down a key on the keyboard, talking to your friendly AI. Can you build this with a bit of that and some of this, and oh, can you make it pink? Great. And by the time you’ve thought of the next thing, it’s done the thing.

And it’s just so many times Gabby will be like, “It’s dinner time. Dinner’s ready.” I’m like, “Yep, hang on.” Right. You do that. You do that. You do that. You do that. Oh, oh, oh, hang on. It’s finished already. Okay. Now do this. It’s just giving it more and more stuff to do.

I had a Trello board with some of those ideas that you’re talking about, the someday maybe and things I wanted to add to the PMAX script, and things I wanted to build and a website that I’d never got around to over Christmas. I just got, enough’s enough. Took a screenshot of the Trello board. I didn’t even try and figure out the API or the MCP or anything. I just took a screenshot of the Trello board, dropped it into Claude, and went, which of these can you build first?

It went, right, I think we should do this one, this one, this one, this one, and here’s why. These things I don’t think you need anymore. This one you’ve actually done. Would you like me to get started? Off you go then. And it just, it, I’ve had 80, 90% of my Trello board done.

One of my members told me this story. This is so good. One of my members, David, who didn’t know what the word GitHub was eight, nine, 10 months ago, and he won’t mind me saying he’s in his 70s. Wonderful guy. Was a client of WebSavvy back in the day. One of the first members of my agency, Savvy Group, way back when. He’s now my mastermind, and he demonstrated what he does for his business.

So he started the business 40 years ago. Now his daughter and son-in-law run it. He’s sort of just been sitting back building all of this stuff for the last six, eight months, vibe coding, I suppose. And so they had this offsite the other day. He’s in Scotland, and he demonstrated to the team some of the stuff that he’s been doing. And they all just looked at each other, right, you’re not allowed to do anything else except this now. Like, “No more fiddling with the Google Ads account. No more coming out on site with us and doing installs. You just do this.”

And then the next day, his head of installs called him and said, “I’ve been thinking about what you, I won’t try and do the Scottish accent, I’ve been thinking about what you showed us yesterday. Do you think you’d be able to build an app for my guys on site?”

And David was just about to say like, “It’s not as easy as that, mate. Like, there’s a lot that goes into this, and we have to think about things.” Yeah, I’ll see what I can do. And he sent me the conversation. He said to Claude, “I’m just about to go to bed. Here’s what I want you to do. I want to wake up to a fully functioning app for the installers.”

Because he had the brain, so it’s got all of the context in there. There were some PDFs lying around about installing a bunch of stuff that was in there. He just said, “I want to wake up to a finished app. Go.” He said he sat down the next morning at 7:00 a.m. He’d forgotten he’d done this. He opened up his computer and went, “Oh, what’s this?” And a fully functioning app with links and PDFs, and all he had to do was drop it into Vercel and deploy it, send it to his install guy, and went like this, you mean?

And it all worked for the first time on his guy’s phone. Now the whole team is able to use it on site. And it was literally a two-minute can you build this because he knew how to ask, and he knew all of the files were there. All of that grounding that you’ve talked about, all of that context. I picture it as like the AI model in the middle, and all of this context is the sort of next ring around that, that is so, so important. It built the whole thing for him overnight. It’s just crazy.

Yeah. It’s so bloody useful. If you embrace it and don’t expect it to do everything, it’s not going to do that nuance. But it can help with so many other things. And stuff that used to take us, well stuff like you were saying, stuff that used to take so much time that we probably wouldn’t do it. We wouldn’t have time to do it, that stuff now becomes accessible.

And some of that stuff is now accessible at scale. There’s something that maybe you would do for your top 10% of clients once or twice a year because it was worth it for them, but you never had the time to do it for the bottom 80%. Now you can do that for every client every month. There’s, it just removes so many of the limits and the barriers and the constraints.

Frederick Vallaeys: So let’s bring it back to one of the questions we had in digital marketing. I mean, obviously, there’s a ton of excitement that you and I share around this, but if you could say what is one skill, and this is Sharon’s question, what is one skill that you think is currently underrated in paid ads and that would make the biggest difference in the next few years?

Mike Rhodes: My flippant answer was going to be thinking, but I won’t go there. I think about business skills, business context. I think what I always sort of had in the back of my mind for WebSavvy was to create McKinsey for the small businesses that couldn’t afford McKinsey. But really quickly, I worked out that all the really, really smart people got hired by McKinsey for an amount of money that I would never be able to afford to pay them. And so it never really happened.

Our best clients were those clients who we were the strategic adviser for and called us to ask about other stuff, not just other marketing stuff, but other business stuff. But that didn’t scale because there were me and a few other people that could answer those business questions. But a lot of people were like, “Well, I was hired to do Google Ads. I know Google Ads. I don’t know anything about business. This is my first job.” And they’re asking me about this, help? Can someone that knows this help me? It just didn’t scale very well.

But I think that business context and understanding more about business, it’s never been easier to learn, but that is going to be the thing that yes, the fundamentals are still very important, and that’s obviously like table stakes. You have to know the fundamentals of running Google Ads. But knowing more about marketing, knowing more about advertising, understanding different channels, but then the level above that, understanding the client better, and why they asking for this, and what they really need, and what is going on, and how else could we help them with these other areas of their business.

Now, that won’t work for all agencies, and it won’t work for all clients because some clients will be, “Hang on, I hired you for my digital advertising. I don’t want you telling me about my strategy. Like, we’ve got that under control. Thanks very much. Off you go.” But others will be delighted that you’re showing an interest in that, that you’re thinking about that, that you’re wanting to help with that.

And I think more and more that will make the difference because, and you’ve written about this very eloquently for a very long time, right? There’s more and more automation inside Google Ads accounts. That automation is available to everybody. That’s table stakes. Everybody’s using the same tools. It all kind of converges to the same point.

Most of the wins happen outside of the Google Ads account. Whether that be the landing page, the offer, how the person answers the phone in the legal office, or whatever it is, it’s outside of Google Ads. That’s where most of the wins can be had. So, understanding more of that part, I think, will be more of a differentiating factor in the future.

Frederick Vallaeys: Yeah, that makes sense. And you kind of said like doing the thing that’s going to become automated, you know, looking for your search terms with $100 in spend, no conversions, right? Why do we need to pay someone to do that? Optmyzr automates that; scripts automate that. Like that’s the easy stuff, right?

But understanding what we do next? Why did these things not convert? Was it the landing page? Was it the offer? Yeah. Was it just bad keyword selection? Maybe this doesn’t work anymore in the age of prompting instead of doing keyword searches, right? So those questions become much more important. And as you said, it’s about thinking and the strategic partnership that you bring, not the final tactical execution, because that is the stuff that automation will do.

Mike Rhodes: So, I think to sum my part of that up more succinctly and not wander off on 15 tangents again. I think the skill is asking better questions because as AIs get smarter and smarter and smarter, the answers get better and faster and cheaper. We need to ask better questions.

Frederick Vallaeys: Great. Well, fantastic stuff, Mike. This has been very helpful. A lot of people will probably be running to their computers and starting to build some cool stuff. So, if you do, please put it in the comments. We’d love to see and comment on it. Mike, aside from Ads to AI and 8020 Brain, I’m sure people can also connect with you on LinkedIn. What else would you like people to do?

Mike Rhodes: Let’s just, yeah, go build. That’s what I would love people to do: not be scared of it. In fact, I was talking to a lady yesterday. I shall leave names out of this, but we were talking about all of this stuff, and she said, “Oh, but Google Ads scripts still scare me.” I’ve got the solution for you. I actually called it my scary challenge. It’s been up on the, this is another website, another URL. Sorry, I’ll have to link to things better, but 8020agent.com/scary, and it’s like a little 14-minute challenge that you go through to make you realize that Google Ads scripts aren’t scary. It just starts to show you what is possible.

And so it’s the same thing that all this stuff that we’re talking about. And yeah, Claude Code looks scary. And if you happen to be on a Mac, and you happen to be already paying Claude 100 bucks a month, you now have access to Claude Co-work that came out a couple of days ago, which is giving you the power of all of the stuff that us geeks are playing with, but making it less intimidating.

That I think will definitely trickle down. That will become more widely available. The key is to just get over being scared of it. That’ll only last for the first five or 10 minutes. It looks intimidating. It was built by geeks who don’t have the best UI sensibilities in the world. But don’t worry about it. It’s basically just a chatbot. And that won’t be the final form factor of all this stuff. But start there and ask better questions, and just try it.

I think far too many people try it and go, “Yep, told you it was crap. This thing will never be as good as me at doing blah blah blah.” And therefore, I will forget all of AI and put it away for another 12 months, and I was right, and I don’t have to look at it again. It’s like, but it’s so much more than that. It can do so many different things.

I don’t want to start a whole new conversation about jagged intelligence and… but just try it. Play with it. That would be my plea. Don’t worry about my stuff. Sure, come to 8020brain.com. That’s the demo that shows you some of the stuff that’s possible, and Ads to AI is the community.

But just play, just start. Open up Claude and say, “Build me a working snake game.” And watch it build on the right-hand side. It’ll pop open a little window on the right-hand side. You may have to change a setting to do that. Just ask Claude, “How do I make sure this thing pops up on the right-hand side in something called an artifact?” Like, “What’s an artifact? How do I get that to show?” And just say, “Show me a game.” Ask your kids, “What’s your favorite game?” Right? Let’s build that.

We built Asteroids two Christmases ago in half an hour, a fully working game of Asteroids that my daughter absolutely loves. Can we play that game again, Dad? Like, don’t forget the Google Ads bit. Build something fun, but just build. Just see what it can do. That would be my plea. Keep on building.

Frederick Vallaeys: Great advice. Keep on building. Okay, keep on building. Keep on watching. Subscribe to our podcast if you want to know about future episodes coming up. Make sure you connect with Mike Rhodes, our guest from today. And with that, thank you, everyone, for watching, and we’ll see you for the next one.