How Lars Maat Built an AI-Driven Agency: Tools, Workflows, Team Adoption, and Risks

Episode Description

Running an agency that’s all-in on AI isn’t just about using ChatGPT for blog posts. It’s about replacing entire parts of your tech stack, building custom tools in weeks instead of months, and getting a team of 33 people to actually use AI in their daily work, not just nod along in training sessions.

Frederick Vallaeys, CEO & Co-Founder of Optmyzr, sits down with Lars Maat, founder of Maatwerk Online in Rotterdam, to talk about what it actually looks like to run an agency that’s fully leaning into AI. Lars has spent the past year systematically evaluating every software subscription his agency uses and asking: Could we vibe code this ourselves?

The answer, more often than not, has been yes.

If you’re trying to figure out how to bring AI into your team in a practical way, this is a good one to watch.

Here’s what you’ll learn:

• How AI expands agency capacity instead of reducing hours worked
• How Lars replaced SaaS subscriptions with vibe-coded internal systems
• Real marketing examples: from holiday campaigns to herb-powered SEO
• When “good enough” is actually good enough (and when it’s not)
• Can AI replace tools like Optmyzr?
• Build vs buy: when to replace software with AI (and when not to)
• Why MCPs and connected systems beat standalone tools
• AI tools in practice: Claude, Cursor, voice workflows, and constant switching
• How to get 33 people to actually adopt AI
• Real use case: scaling SEO for 48 parking locations across 25 cities
• Why hourly pricing breaks in the AI era (and what replaces it)
• Who should innovate inside an agency

Episode Takeaways

Lars Maat didn’t set out to build a 33-person agency. He started as a freelancer in 2012 after leaving an agency. People kept asking him for help with digital marketing. One client became two. Two became five. A hobby became a business.

Now he runs Maatwerk Online in Rotterdam, a name that translates roughly to “custom work,” which has become oddly prophetic in 2026.

Fred opened the conversation by posing a hypothetical to Lars: if you were starting as a freelancer today, how would AI change the trajectory? Lars’ answer was immediate:

“I would probably get more work done, work for more clients, and probably make more money as well, in maybe the same amount or even less amount of time.”

But the reality, as both of them know, is more complicated than that. AI doesn’t reduce workload. It expands what’s possible. And if you’re the kind of person who gets excited about what’s possible, you end up working more, not less, just on different things. Things that used to be shelved as “someday” projects now get built in two weeks, and the bottleneck shifts from execution to imagination.

This conversation is about what happens when you stop treating AI as a productivity hack and start treating it as infrastructure. Lars has spent the past year systematically replacing software subscriptions with vibe-coded tools, building internal systems that do exactly what his team needs, and figuring out how to get 33 people to actually use AI instead of just nodding along in training sessions.

How AI expands agency capacity instead of reducing hours worked

Fred started with the question everyone’s asking: Does AI actually give you more free time, or do you just end up doing more work?

“I see myself as an idiot in the business. I want to know everything what’s possible, and I find everything interesting,” he said. “And I think you are probably quite alike.”

That’s the trap or the opportunity, depending on how you look at it. When Gemini or Claude releases something new, Lars dives in immediately. When Google Ads adds a feature, he wants to test it right away. When vibe coding makes something possible that wasn’t possible last week, he starts building.

“Before you know it, it’s like 1 or 2 a.m. in the night, and you’re supposed to go to bed because I’ve got some little kids as well who tend to come a couple times in the night,” Lars said. “But because it’s that interesting and you are so enthusiastic about the possibilities, you start working with it, and you don’t want to stop actually.”

Fred agreed completely. He’s been up past midnight more in the past year than he has in a long time. Not because he has to be, but because he’s building things that used to be impossible without a team of developers and months of planning.

The promise of AI is that it’ll do all this work for you, freeing up your time. But the reality is different. AI expands capacity and lets you take on projects that would have been shelved indefinitely. It also removes the friction of needing to coordinate with designers, developers, and product managers just to test an idea.

But that means the constraint shifts. You’re no longer limited by execution time. You’re limited by how many ideas you can pursue simultaneously without losing your mind.

Fred pointed out that he’s investing time in building AI processes that will 10x his team’s capabilities in the future. Individual team members might use AI to finish tasks faster, but they’re not the ones staying up late. That’s the founders and technical leaders who see the possibilities and can’t stop exploring them.

How Lars replaced SaaS subscriptions with vibe-coded internal systems

Lars described his current obsession: systematically evaluating every software subscription Maatwerk Online uses and asking, “Could we vibe code this ourselves?”

“We are looking into every subscription, every software that we are using for stuff, and we just try to vibe code it ourselves,” Lars explained. “We try to make one big environment, call it a portal, where every piece of software and equipment that we are using third-party suppliers for now comes into the portal, vibe coding it ourselves.”

The benefits are twofold. First, cost savings. Building your own tools is cheaper than paying monthly SaaS fees, especially when those fees scale with team size or usage. Second, customization. You can build exactly what your team needs instead of working around the limitations of off-the-shelf software.

Lars gave a specific example of a staff engagement software. His agency used a tool that let them survey employees anonymously. When someone gave negative feedback, the team wanted to dig deeper, ask follow-up questions, understand context, and get examples. The vendor said that wasn’t possible in their software.

“I’m like, how is that not possible?” Lars said. “So we built it ourselves. And I’m sorry for that supplier, but we have to cut you loose, and we are using our own program nowadays.”

Fred immediately understood the pattern. He’s been making similar decisions at Optmyzr, not because the software is bad, but because it can’t move fast enough to keep up with what they want to do.

This is the shift happening across the industry. Software used to be a constraint. You adapted your workflow to fit what the tool could do. But now, with vibe coding, the tool adapts to your workflow, and if it doesn’t, you just build a new one.

Real marketing examples: from holiday campaigns to herb-powered SEO

Fred shared a marketing example he’d vibe-coded the night before: a calendar of fun holidays paired with an AI image generator.

The idea came from a conversation with friends who run a jewelry business. One of their Facebook ads had unusually high engagement. When Fred asked why, they said it was tied to a niche holiday—National Pizza Day or something similar. People who didn’t care about jewelry cared about the fun holiday, and that expanded their reach.

Fred thought, “I should build something with this.” So he did. He created a system that pulls fun holidays from a calendar API, takes a character (either a photo or a generated character like a yeti), and uses AI to create images of that character celebrating the holiday. Then it writes social media copy to go with the image.

“Yesterday I actually made a yeti for Earth Day who had a recycling can on his head, and then he was juggling soda cans and crumpled pieces of recycling paper,” Fred said. “And that’s what’s been so exciting to me is that this is a thing that the idea came about two weeks ago, and now I have an image generator that’s far better than what you get in platform in Meta or in Google.”

Lars jumped in with his own example: an e-commerce client selling herbs and spices. They ranked well organically for product searches, but traffic had plateaued. The team needed a way to bring new people to the site who weren’t actively shopping for herbs yet.

“We vibe-coded a structured template in HTML where recipes for each of the herbs and spices they are selling could be placed,” Lars explained. “We used AI to come up with recipes for all of the products they are selling. We used AI to generate images for the recipes. We pushed everything into the HTML template. We used AI to generate the structured data for SEO results.”

They ended up with 400 new recipe pages added to a 450-product site. Within weeks, organic traffic spiked. Though the onversion rates for recipe visitors were lower than product page visitors, the recipes created a remarketing audience and familiarized new people with the brand. Plus, each recipe page linked directly to the products needed for that recipe.

Fred loved the example since he had done something similar with an Advent calendar during the holiday season. The look and feel screamed “vibe-coded”—purple color scheme, familiar UI patterns—but it worked. It was functional, and it only took an evening to build.

When “good enough” is actually good enough (and when it’s not)

That led Fred to an important question: do we eventually need to spend more time polishing vibe-coded projects to remove the “AI signature,” or is good enough actually good enough?

“I think it depends on the case. In the example of your Advent calendar, you just wanted to do something fun for the holidays. You wanted to use vibe coding, and you came up with it. I know probably once you’re going to an event and you want to build something to introduce your company to potential clients, you will probably just do way more in terms of look and feel,” Lars answered.

The standard shifts based on context. If you’re building an internal tool or testing an idea, good enough is fine. If you’re building something client-facing or representing your brand at a conference, you’ll invest more time in polish.

“If you are using your first project, if you are building your first project while vibe coding, as soon as the system builds something that you are okay with, you are probably so enthusiastic, and you’re going to share it,” Lars said. “And once you are at your 50th project, you will probably say, ‘Okay, it’s working fine, but the look and feel is not that nice. So let’s adjust that.’”

Experience matters. The more you build, the higher your standards become. But that doesn’t mean you should delay shipping. Ship the thing. Learn from it. Improve it later if it matters.

Can AI replace tools like Optmyzr?

Fred runs a software company. His customers are asking: why should we pay for Optmyzr when we can vibe code our own tools?

It’s a fair question. And Fred has thought deeply about the answer.

He broke software into two categories. First: tools that are essentially just a UI layer on top of an existing platform like Google Ads.

“We’ve seen a lot of our competitors come out, and they just kind of reskin the capability that Google Ads natively has,” Fred explained. “Their whole pitch was, ‘Well, the Google Ads interface is complicated. It’s hard to find the thing you need to do. So here’s maybe a simpler version with just exactly the buttons that you need.’”

In an AI world, that category of software dies. Why pay for a simpler UI when you can just tell Claude, “Pause this campaign” or “Increase this budget”?

The second category is different: tools that do deep analysis and business intelligence that’s genuinely hard to replicate, like pulling data, adding to spreadsheets, VLOOKUPs, and comparing it to other sources of data.

This is where Optmyzr lives. The value isn’t the UI. It’s the 10+ years of knowledge encoded into what a “duplicate keyword” actually means, or how to forecast budgets while accounting for seasonality, campaign structure, and bidding strategy interactions.

“That’s where we think we are, a business intelligence layer where we’ve spent over a decade coding up what that actually means,” Fred said. “And it’s deterministic. Deterministic is very important because AI code is not deterministic. It’s going to make slightly different decisions every time you run it.”

When you’re managing million-dollar ad budgets, deterministic matters. You want guardrails. You want to know that if you run the same optimization twice, you get the same result.

But the majority? They want convenience. They want a tool that just works. They want updates pushed automatically when Google Ads changes. They don’t want to maintain their own codebase.

“As a customer of Optmyzr, as soon as something new is being released, you guys are working with that, and the fact that everything is in place for me and it’s convenient. I don’t have to worry about stuff, I think at least at this time, the majority of people out there want that,” Lars said.

The software that survives is software that has depth, data, and determinism. The software that dies is software that’s just a prettier face on someone else’s API.

Build vs buy: when to replace software with AI (and when not to)

Lars explained his team’s decision-making process for evaluating software: “It’s not only a budget thing. It’s mainly a thing like, ‘Okay, this is what the software is doing, but we actually want it to do this two or three steps further.’ And the software isn’t able to do that at this moment.”

If the software does what they need, they keep it. If it costs a few euros more per month but saves time, that’s fine. But if they hit limitations repeatedly, they build their own version.

The other reason to build: demonstrating capability to clients.

“We want to show our clients that we are able to build those kind of things, build APIs together, automate stuff,” Lars said. “Because if we could do it for ourselves, we could do it for our clients as well.”

Most software is bloated. It serves millions of users, so it has features you’ll never use. But it also lacks features you desperately need. That’s where frustration builds.

Vibe coding lets you build exactly what you need with just the thing that solves your problem today.

Why MCPs and connected systems beat standalone tools

Fred shifted the conversation to MCPs—Model Context Protocols—essentially APIs that enable AI systems to interact with tools and data.

“For listeners who don’t know what an MCP is, it’s kind of like an API, an application programming interface, but for modern AI,” Fred explained. “It enables an API to understand your data layer and interact with your tool in some sort of a programmatic AI way.”

Optmyzr has an MCP. That means you can tell Claude, “Enable the Optmyzr MCP,” and now Claude can access Optmyzr’s tools, data, and 10+ years of PPC knowledge without you needing to rebuild that logic yourself.

“Your AI can start to focus more on, ‘Oh, well, what is it that Fred needs today? What is he trying to achieve for his business or his client?’ And then it can connect to Optmyzr and actually get that done faster,” Fred said.

Lars shared a frustrating counter-example: a client showing him a sales reporting deck in Excel. Lars asked for API access so he could pull the data automatically. The client said the reporting tool didn’t have an API. It didn’t even have an export function. Everything had to be typed manually.

“I was like, ‘Are you kidding me?’ If you are a software company and you don’t have those kinds of things, I can’t really understand how is that still possible at this moment,” Lars said. “And do you still believe that you are a company in one or two years?”

The lesson: software without APIs or MCPs is dead software walking. In an AI-first world, if your tool can’t connect to other tools, it’s isolated. And isolated tools get replaced.

“As long as you work with companies who are evolving at the speed of AI and come up with things like MCPs, then you are working with the right software,” Lars concluded.

AI tools in practice: Claude, Cursor, voice workflows, and constant switching

Fred asked Lars which tools he’s using day-to-day. The answer revealed something important: the ecosystem is moving so fast that even experts are constantly switching.

“That’s also the problem I’m having with AI,” Lars admitted. “If you’re talking about being enthusiastic and getting stressed out, there are so many tools out there. And every time a tool gets a new update, you’re like, ‘Hey, this is the tool I’m fine with.’ And then, oh, but on the other hand, this tool got this update. I need to start using that.”

Lars has been a fan of Cursor. Then he switched to AI Studio from Google because it was more convenient at the time. But then he realized Cursor can take over your computer while AI Studio can’t. Now some of his team is using Claude Code to build the internal portal that connects all their software together.

The point isn’t which tool is “best.” The point is that the tools keep improving so rapidly that what was best last month might not be best today.

Fred mentioned he’s not typing anymore, he’s using voice commands extensively.

Lars agreed, “I’m using Whisper Flow to just tell the system this is what I want, and you see the screen flickering and changing, and pop-ups are popping up everywhere, and then all of a sudden you get something on your screen, and it’s done.”

Whether that’s Claude Code or another tool doesn’t matter. What matters is using the tools and getting better at them.

How to get 33 people to actually adopt AI

Fred asked the question every agency leader is struggling with: how do you get your team to actually use AI instead of just nodding along in training sessions?

Lars was brutally honest. “At the beginning, the problem I was having was the fact that nobody was using it. I was seeing so many potential things that we could do, not only in terms of efficiency but also in terms of creativity for our clients. And I was being really enthusiastic, and I was sharing screen videos of what I was doing in Cursor, for example, for a couple of clients and what I did and what the outcome was. But I was constantly demotivated because nobody else was picking it up.”

Sound familiar? That’s the reality for most agencies. The founder or technical lead gets excited. They build things. They share demos. And then… nothing. The team keeps doing things the old way.

Lars tried a different approach: a 12-week internal course. “We said, ‘Okay, everybody needs to follow this course.’ It’s something we built ourselves, and it’s mainly just guidance through AI and Cursor and those kind of things and everything that is possible.”

But here’s the key: they didn’t just teach tools. They gave each team homework.

“Each team—so for example, the SEO team, the SEA team, social advertising team—we gave them homework. Come up with things that you do repeatedly every day, every week, or every month. Come up with things you hate to do in your work. Come up with things that you think we could do faster or better,” Lars said.

Every team came back with a list. Then came the assignment to get the list and start working with number one and fix it with AI.

“That was the moment everything switched,” Lars said. “All of a sudden, everybody was using it, or at least start thinking about the possibilities, and they came up with some great tools, internal tools we could use.”

The breakthrough wasn’t teaching AI. It was connecting AI to problems people already had. Once they saw that AI could solve their actual frustrations, adoption became organic.

But that also created new problems, like how to analyze or update certain things. And that’s when they created a guidance document: what we can do with AI, what we can’t do with AI, how we think about these things, and how we proceed from here.

Real use case: scaling SEO for 48 parking locations across 25 cities

Lars shared another example that shows what’s possible when you combine AI, APIs, and a clear business problem.

A client runs 48 parking lots across 25 cities in the Netherlands. If you search “parking lot Rotterdam,” they rank well. But if you search “parking near Rotterdam museum,” they don’t show up. The problem: they weren’t targeting activity-specific searches.

“We were like, ‘Okay, how can we change that?’” Lars said. “And we used AI to come up with a Python script. We didn’t have any Python knowledge whatsoever, but we knew we had to use Python.”

They built a system that:

1. Used the Google Things to Do API to scrape activities in each city
2. Used the Google Maps API to identify which activities were within walking distance of each parking lot
3. Pulled search volume data from the Ahrefs API to prioritize high-traffic activities
4. Used AI to generate SEO-optimized text for landing pages targeting each activity

“All of a sudden, we’ve got SEO-friendly text on the website, completely built with Python scripts and API connections, where we only used AI,” Lars said. “And those are all kind of things that we couldn’t come up with ourselves a couple of years before. And even if we did, we wouldn’t have it online within one or two weeks.”

Fred’s mind immediately went to the next step, “Do you now have structured data? Why not feed that into Optmyzr’s Campaign Automator and now you can also have automatically dynamically generated ads for those parking lot locations near museums?”

“That’s what we’re doing right now,” Lars confirmed.

That’s the power of this approach. The thing that would have been shelved as a “someday” project got built and deployed in two weeks. And once it’s built, it becomes infrastructure for the next idea.

Why hourly pricing breaks in the AI era (and what replaces it)

The unprecedented efficiency of AI has made the traditional hourly fee model obsolete, requiring agencies to pivot to a value-based model.

Lars Maat stated that in this new environment, “If you are still using pricing like an hourly fee model, it’s not going to work anymore. We are working like what we call the value-based model. This is what you want, you pay this amount, and this is what you get. And we are not talking about how many hours of work are in there.”

The payment is for the knowledge and experience that enables rapid delivery, transforming the agency’s speed into a value-added feature, not a reason for discounted labor.

To manage client expectations and fund continuous development, agencies should be transparent about their pricing structure. Maatwerk Online tells clients, “X percent of the fee that you are paying us is being used for internal innovation time, and that is where we come up with those kind of things. Of course, you are paying for that, but you are getting the benefits of that as well.”

This approach not only justifies the value but also grants the agency leverage to dedicate staff time purely for innovation, which is crucial for long-term survival in the AI era.

Who should innovate inside an agency?

Fred pointed out that he plays the innovator role at Optmyzr—trying different tools, figuring out which ones work, then pushing the team to adopt them. It’s the “PPC doctor” role he wrote about in one of his books: understanding the problem, knowing which solutions exist, understanding the tradeoffs. However, agencies do not need every employee to embody this mindset.

Lars Maat provided a realistic assessment, “At the same time—and I think that’s something that is in every company the same—there are maybe 80 or 90% of the people who don’t want to be an innovator. They just want to show up, do their work, talk with colleagues, fix the problems for the clients, and that’s fine.”

The key to successful agency-wide AI adoption is focusing energy on a small, core group of enthusiasts.

According to Lars, “As long as you’ve got four or five people in the room who are exciting about this and testing with this, then the other 30 people will be fine as well.”

These innovators are responsible for building the systems, documenting the lessons, and translating that knowledge into usable systems for the rest of the team. This ensures that whether a person thrives on exploring new possibilities or simply wants to execute good work, the necessary systems are in place to support both groups.

Episode Transcript

Frederick Vallaeys: Hello and welcome to another episode of PPC Town Hall. My name is Fred Vallaeys. I’m your host. I’m also CEO and co-founder at Optmyzr, a PPC management company.

So for today’s episode, we’re bringing in another of the top 50 most influential PPC experts, and this one is based in the Netherlands. He’s been in the industry for quite a long time. And the reason that I brought him on to the show today is because he’s another big fan like myself of AI, vibe coding, and just figuring out new ways to do things with all of this cool technology that we have.

So our guest today is Lars Maat. I don’t know how you say that in English because I originally speak Dutch as well. Also, Lars Maat from Maatwerk Online and also the co-founder of The Social Club. So can’t wait to hear what Lars has to share with us. And with that, let’s get rolling with this episode of PPC Town Hall.

Lars Maat: Thank you for coming on. Yes, it was a good intro. You pronounce the name right.

Frederick Vallaeys: Oh, good. Good. How do people in English say your last name?

Lars Maat: Maat, Matt, I think. When I’m at an event and they are announcing me as a speaker, I think everybody just gets fine along with Matt.

Frederick Vallaeys: Okay. Yeah, it’s not that hard to say, right? But then that also leads into your company name, which is Maatwerk Online. That probably doesn’t make a lot of sense to our English listeners, so maybe explain that one a bit.

Lars Maat: Yeah. So I’m not sure what the English word is for Maatwerk. But it basically means that we are a digital agency based in Rotterdam in the Netherlands, and we don’t work with standard packages for every company. We try to make a plan based on the needs of the company.

So there—it’s not like, “Hey, if you want us to run your PPC campaigns, it will cost you X a month.” Every company has its own price, plan, strategy. And in the Netherlands, that’s what we call Maatwerk. So that’s basically where the name comes from.

Frederick Vallaeys: Yeah. So custom basically, right? Is would be the translation that I would think of. Yeah, that’s great. So how long has Maatwerk Online been in business?

Lars Maat: Well, actually, funny story. I worked at an agency before, but it was not like the standards of the agency were not like my standards anymore. And at that time—and I’m talking around 2012, 2013—a lot of people were asking me questions from, “Hey, you know, you are working in digital marketing. Is this something you can do for us? Could you help us with that?”

And it basically was the reason that I decided to start working for myself as a freelancer. And I never intended to have like staff or a real agency. And here we are in 2026. I’m sitting in a big office in Rotterdam, and we’ve got 32, 33 people working at the agency. So it’s kind of like a hobby gone wrong.

Frederick Vallaeys: Well, congratulations on the success of your hobby gone wrong. But it’s—we’re in interesting times now, right? Because obviously you went from being a freelancer to hiring all of these people because presumably you couldn’t do the work of 33 people by yourself. But say that you had started the agency or the freelancing company today. How do you see AI as an enabler in that?

Lars Maat: Well, it’s difficult because, you know, when you start to think of how I did it in the early days and how I would do it now, it’s completely different. There are so many more opportunities to do stuff way better, faster, more precise with not only AI but also automations because I think a lot of people are using the word AI while they are working with automations. So there’s obviously a difference between them.

But yeah, I would probably get more work done, work for more clients, and probably make more money as well, in maybe the same amount or even less amount of time that I used to do back in the days.

Frederick Vallaeys: Yeah. Well, and to me, this is interesting because some people talk about AI doing things faster, right? So you can do what you normally would have done in a normal workday in one or two hours. But then the question is, do you take these extra six hours, or if you have a long workday, these extra 10 hours, and do you just enjoy some leisure time, time with the family, or do you end up just doing more things? Do you take on more clients? Do you do more projects?

And for me, where it’s been exciting is that it’s just enabled me to bite off projects that in the past I would have shelved because it just would have taken too long. I would have had to talk to too many people to get it done. I would have had frustrations about maybe they don’t understand my vision. And then the back and forth is just endless.

But I find myself more excited but also more stressed out than I’ve ever been. And stressed out just because there’s more things I end up doing. But the excitement is in being able to do these things. And like for you, where does it fall in terms of the AI? Like are you taking less time off, or are you just doing new things altogether?

Lars Maat: I think the excitement is also the main reason of the increasement of the stress. Because I see myself as an idiot in the business. I want to know everything what’s possible, and I find everything interesting. And I think you are probably quite alike.

So the problem for me is as soon as Gemini or Claude is releasing something new, or Google Ads is releasing something new, I want to dive into it right away. And nowadays with all the information that is available and all the possibilities that are available in terms of vibe coding, for example, you start testing it right away. And before you know it, it’s like 1 or 2 a.m. in the night, and you’re supposed to go to bed because I’ve got some little kids as well who tend to come a couple times in the night.

But because it’s that interesting and you are so enthusiastic about the possibilities, you start working with it, and you don’t want to stop actually. At least that’s what I’m experiencing.

Frederick Vallaeys: Yeah, totally the same here. I mean, the number of times I’ve been up unusually late, past midnight at work, it’s a lot more than it used to be. But it’s exciting, right? So but talk to me about what is the current project that’s keeping you up at night and programming throughout the day.

Lars Maat: Yeah. So what we currently are doing at the company is we are looking into every subscription, every software that we are using for stuff, and we just try to vibe code it ourselves. And we try to make one big environment, call it a portal, for example, where every piece of software and equipment that we are using third-party supplies for now come into the portal, vibe coding it ourselves. And it’s really amazing not only what is possible but also the amount of money you can save by building your own programs.

And at the same time, you can custom-made those programs for your company. So for example, we’ve got a software we are using for—how do you say it?—staff engagement and staff happiness between the company. And it’s totally anonymized, but we can ask questions, and people at the office are responding to it, but we don’t know who is responding.

But as soon as somebody fills in a negative answer, we want to dig deeper into that answer. Okay, why is this? Do you have examples and stuff like that? And the supplier is basically telling us, “Ah, that’s not possible in our software.” And I’m like, “How is that not possible?”

So we built it ourselves. And I’m sorry for that supplier, but we have to cut you loose, and we are using our own program nowadays. And that is what’s really exciting. And I know that it’s not like a PPC example, but the possibilities with all the things you can do nowadays in terms of vibe coding and using AI, it’s amazing.

Frederick Vallaeys: Yeah. And let me give a marketing example maybe on this. But I totally get what you’re saying, right? So you look at the status quo, and you rebuild a little bit of the status quo in vibe code, and then it’s like this blank canvas where you can just prompt to do one more thing.

And what sometimes frustrates me is that I see people being stuck in the status quo. So they’re like, “Oh my God, I just vibe-coded a version of, say, PowerPoint. Great, but what was it about PowerPoint that frustrated you in the first place that you wanted to do more with?”

So the marketing example of what I vibe-coded last night was it’s a calendar of fun holidays. So like National Donut Day or National Pizza Day or National Scratch Your Back Day. And then an image generator that takes a character. So it could be either a picture of you, or it could be a character that you generated like a yeti or a dog, and then some things. And then the AI puts together what could that character be doing on National Pizza Day.

And then it starts generating the image for that. It puts together the character with some accessories using the Llama Nano model. And then it writes a little description of like what could you put on social media that this character is doing that’s related to that special fun holiday event.

And so then I got that, and I’m like, “Oh, the next cool thing would be let’s make this into videos, right?” And so just one more prompt. And now yesterday I actually made a yeti for Earth Day who had a recycling can on his head, and then he was juggling the soda cans and the crumpled pieces of recycling paper.

And so it was a beautiful image, and I was like, “Okay, one more thing. Make it into a video.” Now he’s actually juggling these things. And that’s what’s been so exciting to me is that this is a thing that the idea came about two weeks ago when I was talking to some friends. And these friends, they told me—I was looking at their Facebook campaign, and there was one ad that stood out as having had much higher engagement. And I was like, “Well, what was it about that ad?”

And my friend was like, “Well, it was just—it was related to a special holiday, and just for some reason, that resonated with a whole different group of people that we usually target. And they didn’t care about the jewelry that she was selling, but they cared about the fact that it was about this special fun holiday.” And I was like, “Oh, I should do something with this.”

And so that’s what that became. It became an evening of vibe coding and saying, “Let’s do one more thing, one more thing.” And now I have an image generator that’s far better than what you get in platform in Meta or in Google or in any of these other places.

Lars Maat: Nice. I’ve got a marketing example as well. So we are working for an e-commerce store who is selling herbs and spices, and they are in the business for quite a while now. So if you are looking SEO-wise, when you start searching—at least in the Netherlands—for some of the herbs and spices they are selling, they are constantly number one, number two, number three in the organic results.

But the problem is it’s not decreasing, but it’s not increasing anymore as well. They are like on a plateau. So we thought about, “Okay, what can we do to increase sales, get new people to the website and stuff like that?”

So what we did is we vibe-coded a structured template, an HTML template, where recipes for each of the herbs and spices they are selling could be placed. We used AI to come up with recipes for all of the products that they are selling. We used AI to come up with an image generator for the recipe that they were talking about. And then we pushed everything into the HTML template.

We used AI to generate the structured data for the SEO results. And all of a sudden, we’ve got this herbs and spices website which had 450 products on it, but now they have 400 recipes online as well. And within a couple of weeks/months, all of a sudden they were generating a lot more people to the website. And the conversion rate for those people is way lower because they are not searching specifically to buy the herbs. They are searching for a recipe for that herb.

But when they are looking at the recipe at the website, obviously we made sure that the products that they are using in the recipe, you can buy them directly from that page. So it’s generating new people who are getting familiar with the brand. It is generating sales obviously. It is generating a remarketing audience for our Google Ads. And it’s all built with AI.

And it’s just mind-blowing that we could do this ourselves. We didn’t need a developer. We had the plan, and the execution was in like two weeks. It’s amazing.

Frederick Vallaeys: Yeah, I love that. I had a similar one. So when it was the holiday season, Christmas, the Advent calendar, which surprisingly actually a lot of people in the United States don’t know Advent calendars. But I was like—

Lars Maat: I saw—I know the example you’re going to tell me. I saw the Advent calendar, and I knew right away he vibe-coded this. The look and feel was totally vibe-coded.

Frederick Vallaeys: Exactly. Well, it’s funny how anything that’s vibe-coded usually starts with the color purple. And I was talking to some people who work at OpenAI, and they’re like, “Yeah, ChatGPT and how it writes code, it really loves purple.” So just like you can pick up on AI-written articles, you can really pick up on AI-generated vibe-coded projects as well.

But yeah, so with this Advent calendar, obviously I wanted to make it a very quick project, so there was still that sense—like you’re saying, you could tell that it was vibe-coded. But how do you think about that? Do you think that we will eventually spend more time vibe coding to get rid of that AI, those AI signatures, or do you think it’s good enough and people like it? Or where do you stand on that?

Because my concern is that eventually we end up spending way more time to get rid of those AI signatures to make it look like it was built by a human, and we end up eating up all of that time that we initially saved.

Lars Maat: I think it depends on the case. In this example of your Advent calendar, you just wanted to do something fun for the holidays. You wanted to use vibe coding, and you came up with it. And I know probably once you’re going to an event and you want to build something in order to introduce your company to potential clients, you will probably just do way more in terms of look and feel of what you vibe-code it to make sure it stands out. It’s more connected to the brand.

And I think that that’s the case for every example. If you are using your first project, if you are building your first project while vibe coding, as soon as the system builds something that you are okay with, you are probably so enthusiastic, and you’re going to share it. And once you are at your 50th project, you will probably say, “Okay, it’s working fine, but the look and feel is not that nice. So let’s adjust that.” And I think that that’s the case at this point.

Frederick Vallaeys: Yeah. Okay. So I run a software company, and my customers are certainly asking me questions. It’s like, “Why should we pay for an Optmyzr subscription when presumably you can vibe code?”

So I’ve got an answer, and I want to run it by you and see what you think. So I think there’s two categories. There’s more than two, but let’s just break it down to two categories of software. There’s one software that’s just kind of like a UI layer on top of Google Ads.

And so when we’ve seen a lot of our competitors at Optmyzr come out, we see that they just kind of reskin the capability that Google Ads natively has. And their whole pitch was like, “Well, the Google Ads interface is complicated. It’s hard to find the thing you need to do. So here’s maybe a simpler version with just exactly the buttons that you need.” But fundamentally, it still does the thing that you could have gone to Google Ads and just done right there as well.

And then there’s another category of tools, which is maybe a little bit more like Optmyzr, where it’s like, “Well, we’re not reskinning a pig. We are actually doing very deep analysis that’s a bit more complicated, which tends to involve for an agency a lot of pulling data, putting it in spreadsheets, doing VLOOKUPs, comparing it to other sources of data.” So it’s this backend work.

But then how you present that, it’s like sometimes it’s a little clunky, right, because there’s so much going on that it is difficult to put a really nice user interface on that. And so for many years, we struggled because people were looking at these beautiful UIs and saying, “Oh, well, why can’t Optmyzr be exactly like that?”

But now I think it’s shifting because now the UI layer, the presentation layer, is really going away. That’s what you vibe code. And why even vibe code? Why not just tell the AI what it is you want to do? That is your user interface to specifying how things should be.

And but then it needs to go and do the work which is deep down in the ad system. And it could connect to the Ads API, but there’s a lot of nuance, right? Like, okay, wait, if you said we need to remove a duplicate keyword, but what is a duplicate keyword? Is that just looking at the text of the keyword, or do we need to look at the targeting of different campaigns and the day-parting and the budget exhaustion at certain times a day? So there’s all these things that go in.

And that’s where we think we are a business intelligence layer where we’ve spent over a decade coding up what that actually means. Like, okay, you want to do this thing. Here’s the very sophisticated analysis and manipulation of the data that gets you to that thing that you wanted to do. And it’s deterministic. And deterministic is very important because AI code is not deterministic. It’s going to make slightly different decisions every time you run it.

But when you’re dealing with million-dollar ad budgets, you want deterministic. You want to know that if you said, “Oh, do this keyword optimization,” it’s fine to deploy one of Optmyzr’s tools to do that because that gives you guardrails. And so that’s where I think software is going to start distinguishing itself.

And it’s the software that has your history of knowledge, your data repo basically, or it has this business intelligence layer that’s really hard to replicate. So as you are building things, as you are making decisions about what software to keep, talk me through: does that map to how you think about it, or are there other things you look at?

Lars Maat: Yeah. Yeah. I can totally agree with what you are saying, and specifically looking at the Optmyzr tool. I think the difference is there are probably a couple of guys and girls out there who are so keen on vibe coding and are so like geeks in a good sense of the word, and they are like, “I want to do this myself.” And okay, you may lose those people as a client.

But on the other hand, I think a lot of people are basically using your tool because of the fact that you guys have so much data, and you guys are pushing in so many updates. I think you updated yesterday or the day before something new as well, if I’m correct. And as a customer of Optmyzr, as soon as something new is being released, you guys are working with that. And the fact that everything is in place for me and it’s convenient—I don’t have to worry about stuff—it’s all there.

And I think at least at this time, the majority of the people out there—we as geeks are talking about vibe coding and doing things ourselves, but 80 or maybe even 90% of the marketers out there aren’t using it at the moment, or at least not at the sophisticated way that we are using it and the possibilities that are already available.

So in terms of what we are doing when we are looking at software, it’s not only a budget thing. Of course, it’s mainly a thing like, “Okay, this is what the software is doing, but we actually want it to do this,” and it’s two or three steps further. And the software isn’t able to do that at this moment because of the way it is built in the past or whatsoever. “Why can’t we just do it ourselves?”

And if the software is able to do what we want and it will cost us a couple of euros more each month, it’s also okay. So that’s one reason. And another reason is we want to show our clients that we are able to build those kind of things and build APIs together, automate stuff, because if we could do it for ourselves, we could do it for our clients as well. And they benefit of those steps as well. So that’s mainly the reason why we are looking at this.

Frederick Vallaeys: Yeah, and it’s actually brilliant, right? Because your company name, Maatwerk—custom—we are headed into an era where software is custom. And Sam Altman is saying it’s on-demand software. And what that really implies is that it is software for the exact thing that you need at that exact moment in time.

And it’ll just build the thing that is necessary right there. And when you look at so many softwares, they are all bloated because they need to satisfy the needs of millions of people. And so they have all of these features. But honestly, what do you use in the average software suite that you deploy, right? And so you only use a fraction of what they offer, but then do you need all of these extra things that they don’t offer? And that’s where the frustration comes into play.

Now, as I think about this scenario as well, so I’ve made similar decisions as yours about softwares that we no longer need to invest in, not because they’re bad, but just because they can’t move fast enough at the speed that we want to move. But let’s talk about MCPs and data connectors, right? Because again, there are things that these companies are doing well.

So if I can build on top of what they are doing but have it do just a little bit more, that is actually useful, right? So for that, there’s the notion of an MCP, which is kind of—for listeners who don’t know what an MCP is—kind of like an API, an application programming interface, but for modern AI web. So it’s a Model Context Protocol.

Basically, it enables an API to understand your data layer and interact with your tool in some sort of a programmatic AI way. And this is where Optmyzr then has, say, an MCP. And so now you can talk to Claude Code, and you can say, “Enable the Optmyzr MCP,” because that’s enabling the Optmyzr skill.

So now you’re not asking Claude to figure out how to do a keyword manipulation or a budget forecast for that. It can look at Optmyzr and the 10-plus years of knowledge that we built into how those tools work correctly within guardrails, right? But Claude, your AI, can start to focus more on, “Oh, well, what is it that Fred needs today? What is he trying to achieve for his business or his client? Oh, okay.” And there is a tool from Optmyzr that we can connect with and actually get that done faster.

So that’s where I’m excited about these MCPs. Talk to me a little bit about how you’ve been thinking and using MCPs.

Lars Maat: Well, I think you’re totally correct. And the beauty is you, as an owner of the company, are so involved in the possibilities. And you know that as soon as MCPs arrive, you probably knew instantly, “We need to have an MCP.”

And there are still a lot of softwares out there who don’t have an MCP or even an API call. I was literally in a call two or three hours before this meeting where a client was showing me a reporting deck in Excel. And I was like, “Okay, I need to have access to this because I want to know”—it was the sales reporting—“I want to know how sales is going because based on that, we can adjust campaigns.” And he was like, “Yeah, but I have to type it manually over from the reporting tool.”

And I was like, “Are you kidding me? It doesn’t have an MCP. It doesn’t have an API. It doesn’t even have an export function.” And I’m like, “If you are a software company and you don’t have those kind of things, I can’t really understand how is that still possible at this moment. And do you still believe that you are a company in one or two years?”

So I think as soon as—as long as you work with companies who are evolving and evolving in the speed of AI at the moment and come up with a thing as an MCP, then you are working with the right software, I think.

Frederick Vallaeys: Yeah. Now, okay, that’s—okay, mind blown that this example that you just gave. But luckily there are solutions, right? So let’s talk about maybe Claude Co-work or Claude Code. Is that something that you use? Because the way that I see that is somebody says, “Okay, here’s a spreadsheet, and we can’t export it,” or “There’s—move data from one place to another and no API exists.”

Well, that’s where I’ve been using Claude Code. And maybe the API even does exist, but I’m just like, “Hey, Claude, I’ve got these files in this place. Can you open them up? Can you look at what’s in them and put them in a spreadsheet and do some sort of a classification?” And it literally runs on my computer, takes control of my computer, opens these files, puts it in a Google Sheet, does whatever I need.

So to me, this is brilliant, right? Because we’re no longer dependent on what someone has decided to expose through an API, right? If there is a presentation UI layer that my computer can go and click, it can do that. Now, historically, there have been tools that are button clickers, but they’ve been very complicated because you always have to specify exactly what is the button name, where is it on the screen.

AI is flexible. It can just take a screenshot of your computer monitor and be like, “Oh, it looks like the button—it’s not exactly the text we thought, but it looks like it might be that one. So let me try it.” And that’s what’s brilliant, right? So but I want to hear from you. Have you used Co-work or Claude Code or any of these?

Lars Maat: That’s also the problem that I’m having with AI. If you’re talking about being enthusiastic and getting stressed out, there are so many tools out there. And every time a tool gets a new update, you’re like, “Hey, this is the tool I’m fine with.” And then, oh, but on the other hand, this tool got this update. I need to start using that. So that’s why I’m constantly switching tools.

So I was really a fan of Cursor. Then I actually made a switch to AI Studio from Google, which I think was way more—how do you say it?—it was more convenient for me at that time. But the problem was obviously that Cursor can take over your computer while AI Studio is not able to do that. And then you got Claude Code obviously.

I haven’t used it the way you just described it, but I know that some of my colleagues are actually using it to build the portal I was just mentioning in the beginning where we are connecting all kinds of software together for, for example, reporting purposes. But yeah, it’s so amazing that you could just basically tell the computer—you’re not even—well, at least I’m not typing it anymore. I’m using Whisper Flow to just tell the system, “This is what I want,” and you see the screen flickering and changing, and pop-ups are popping up everywhere. And then all of a sudden, you get something on your screen, and it’s done.

And whether that’s using Claude Code or another tool, it doesn’t matter as long as you are using it and try to get better at it.

Frederick Vallaeys: Yeah. Yeah, that’s great. So super useful. And then how do you think about your team’s usage of AI? Do you put limitations in place? And how do you think about that as a business in terms of potential liability and mistakes happening?

Lars Maat: That’s a good question. At the beginning, the problem I was having was the fact that nobody was using it. I was seeing so many potential things that we could do, not only in terms of efficiency but also in terms of creativity for our clients. And I was being really enthusiastic, and I was sharing screen videos of what I was doing in Cursor, for example, for a couple of clients and what I did and what the outcome was. But I was constantly demotivated because nobody else was picking it up.

And then we came up with a 12-week course, internal course, in which we said, “Okay, everybody needs to follow this course.” It’s something we built ourselves, and it’s mainly just guidance through AI and Cursor and those kind of things and everything that is possible. And each team—so for example, the SEO team, the SEA team, social advertising team—we gave them homework.

“Come up with things that you do repeatedly every day, every week, or every month. Come up with things you hate to do in your work. Come up with things that you think we could do faster or better.” So every team came up with a list. And then we just said, “Okay, you now had this course. You saw what’s possible with all the toolings. Get your list and start working with number one. Try to fix it using AI.”

And that was the thing that was the moment everything switched. All of a sudden, everybody was using it, or at least start thinking about the possibilities. And they came up with some great tools, internal tools we could use. But again, that was the point that we saw we were growing in the uses of the credits. We came up with questions like, “Hey, I’ve got this data file for my client. I want to analyze. Can I just upload it, or how does it work?” stuff like that.

So now we have a guidance document on, “Okay, what can we do with AI? What can we not do with AI? How do we think about these kind of things? And how do we proceed from here?” But that’s obviously really important. But that’s basically how our journey was here at Maatwerk Online in terms of using AI. And now we are doing amazing things.

For—I already told the herbs and spices example. I’ve got another great example. For—it’s for a company who are running parking lots. So they’ve got 48 parking lots in I don’t know, 25 cities in the Netherlands. And the thing is, if you start searching for parking lot Rotterdam, for example, they are going to be found where there are a lot of things you can do in Rotterdam. So when you start searching for parking lot or parking place for this specific museum, you’re not going to find that client.

So we were like, “Okay, how can we change that?” And we used AI to come up with a Python script. We didn’t have any Python knowledge whatsoever, but we knew we had to use Python. And we used AI to come up with a Python script. And the Python script used all these kinds of APIs to not only scrape the Things to Do API from Google to see, “Okay, what are all the things you could do in, for example, Rotterdam?” but also use the Google Maps API to see, “Okay, there were 100 things you could do in Rotterdam. We’ve got three parking lots in Rotterdam. Use the Maps API to see which of the things you can do are in a radius of the parking lot,” because you’re not going to park over here to walk 10 miles over there.

And then we used another Python script to push all the information into a spreadsheet. So all of a sudden, we’ve got this database saying, “Okay, these are the things you could do. These are the things you could do within a 10-minute walking distance of parking lot X.” We used the Ahrefs API to see if there are search volumes for those things to do. And then we used AI agents to come up with text, SEO-optimized text we could place on the website.

And all of a sudden, we’ve got SEO-friendly text on the website, completely built with Python scripts and API connections, where we only used AI. And those are all kind of things that we couldn’t come up with ourselves a couple of years before. And even if we did, we wouldn’t have it online within one or two weeks. And I think that that’s really amazing if you could come up with just those kind of things.

Frederick Vallaeys: Yeah, I love it. And so my mind immediately then goes to, do you now have structured data? Why not feed that into Optmyzr’s Campaign Automator, and now you can also have automatically dynamically generated ads for those parking lot locations near museums?

Lars Maat: That’s what we’re doing right now.

Frederick Vallaeys: Yeah, exactly. And that’s what’s so fun, right? The thing that would have been shelved as a project that we’ll do at some point, now you can actually get it done within the quarter. But maybe it’s a little scary too, because once the client picks up on the fact that, “Oh, we asked for this, and they did it in two weeks and at a reasonable cost,” what else can we ask them? You’re going to have to hire more people.

Lars Maat: That’s a funny discussion, and I’m having that discussion with other agency owners as well. What do we need to do with our pricing? But to be honest, as an agency or as a freelancer as well, if you are still using pricing like an hourly fee model, it’s not going to work anymore. We are working—we call it the value-based model. This is what you want. You pay this amount. And this is what you get. And we are not talking about how many hours of work are in there, because because of the fact that we could do it in two weeks or in one week or whatsoever, it’s because of the knowledge that we have or the experience that we have. And you are paying for that knowledge and experience as well.

But to be honest, we are actually really upfront with our clients, telling them, “Okay, X percent of the fee that you are paying us is being used for internal innovation time, and that is where we come up with those kind of things. And of course, you are paying for that, but you are getting the benefits of that as well.” And that also enables me to talk to the staff and say, “Okay, X percent of the time you are working at the office is purely for innovation. Come up with new kind of things.” And I think that in this era that we are living in right now, that’s really important.

Frederick Vallaeys: Yeah. And that makes me think here, so I find that I play a lot of the innovator role in Optmyzr, trying the various tools and sort of then pushing the team to use one or another. But you’re right, because I’m mostly coding on Lovable, but then yesterday I see something about Replit, or I tried Google AI Studio, and sort of it’s that PPC doctor role that I described in one of my books, which is, “Listen, we understand what the client needs or what the problem is we’re trying to solve, but what are the different medications or solutions that we could deploy for that, right? Which one is best for each scenario? What are the side effects?”

There’s a lot of time that goes into that, and that is valuable. And that is something that as an agency or software vendor, whoever you are, there is time that needs to go into that. But then I also wonder, to what degree do the individuals end up being the innovators themselves versus me saying, “Okay, here’s the thing that I know is now possible,” and then I just have someone go and implement it. And it sounds like you give a lot of that creative freedom to your team as well, so that they individually become innovators rather than it being centrally driven from one person.

Lars Maat: Yeah. But at the same time—and I think that’s something that is in every company the same—there are maybe 80 or 90% of the people who don’t want to be an innovator. They just want to show up, do their work, talk with colleagues, fix the problems for the clients, and that’s fine. As long as you’ve got a couple of people in your team who are really enthusiastic about this and are willing to do the extra mile in terms of learning these things and try to be innovative and make sure that that knowledge is being passed on to the other members of the team, I think that’s okay.

In the beginning, I wanted everybody to have the same mindset as I was having. “This is really exciting. Why are you not testing with this?” And they’re like, “It’s not that exciting for me.” And at the beginning, I found that really hard. But now I know as long as you’ve got four or five people in the room who are exciting about this and testing with this, then the other 30 people will be fine as well.

Frederick Vallaeys: Yeah, makes total sense. Hey, well, this has been fantastic stuff. I feel like we’re going to have to bring you on again in a couple of weeks, as I’m sure all of the AI tools will have had new updates and new things we can do with them. So I’d love to keep this conversation going and hear what you’re up to. And so thanks for sharing everything so far. But for people who want to hear more and find out about what they can do with you, where should they go?

Lars Maat: Probably follow me on LinkedIn. I’m most active on LinkedIn. I’m speaking at a lot of events as well. So probably you will find me at some of those events as well. But yeah, feel free to give me a message on LinkedIn.

Frederick Vallaeys: Great. Yeah. And we see each other on the speaking circuit once in a while. So can’t wait to see you in person again. Hey, but Lars, thank you so much for sharing all your knowledge. And thanks, everyone, for watching. Hopefully you’ll give a thumbs up to the show, start following us, subscribe, give a review. And there’s also the comment section. So if there’s anything you want to get the link to, something that Lars or I mentioned, we’re happy to reply to those. But with that, thanks for watching PPC Town Hall, and we’ll see you for the next episode.