Spotting the difference between hype and productivity gains with Marcel van der Meer

Hey, hey, hey, surprise! It's Ali's Alley. That's right, it is Friday, 1 PM US Eastern time. Not the same time where my friend Marcel is, but it's still Friday hopefully there, right? It hasn't changed.

It is Friday at 7 PM on our side.

Friday evening, Friday day. That's right. So wow, you guys are in for a treat. We have someone from across the pond, although I think a lot of my listeners are also from other places than the United States. At least that's what my Google stats say anyway. Apparently, you know, people would rather listen to me from other countries than my own country, but that just, you know, is what it is. So for those of you that are wondering who my twin is, we are basically identical twins. Marcel van de Meer is an automation— I guess I'll say automation expert. Long-time recruiting expert. So first and foremost, he's a recruiter. He works and lives in Amsterdam. You're in Amsterdam or are you outside of it?

No, no, outside Amsterdam. Is it hours away from Amsterdam? That's near to the German border.

Oh, okay. What's the name of the city?

Emmen.

Emmen. Okay. So anyways, that part of the world. So it's beautiful there. And I've had the pleasure of visiting a couple of times. It's always a good time. and also he's come to speak and do presentations at many of the EREs here. So expert in recruiting, expert in automation, and expert in a very interesting topic that I wanted to dive into today. So say hi, first of all. Hi, Marcel.

Hi.

Hi.

Thank you for having me, by the way, Shadi. Thank you for having me.

My pleasure. Yeah, it's always fun when we talk. You have such great perspective— not perspective, how do I say, point of view. Like, the way you see things is a different angle than a lot of other people. And so, getting an opinion from you is always expanding my universe. You know, like, it gives me more— I'm able to brainstorm more and come up with creative ideas because of our conversations. So, it's always a pleasure.

And back at you because you're my identical twin. That's right. Same counts for you, Shelly.

Sometimes I look in the mirror and I start speaking Dutch. I don't know how that happens. So we agree on a lot of things, but we have very different perspective for it. One of the things that we definitely agree on is that you think that AI is not about replacing humans. It's not about replacing recruiters, whatever, but rather it is about replacing the tasks, or how I say it, it's alleviating you from having to do some of the very mundane, repetitive, formula-based things that we do. Not everything that we do in recruiting is a formula, but there are formulas for many things. And it's easier to have a machine that is very good at applying a formula do that for you correctly and accurately, much faster, even while you're actually busy doing something else, like maybe a podcast or something. So right now I have my robot sourcing and doing some searches for me. So we agree that it's an enhancement. And I wanted to start by saying, this has come up a couple of times on the show before, but I have a very strong opinion about generally speaking about this, what they call a revolution. Okay. In my opinion, you have very advanced techniques and you coach people on how to do this. So you're very familiar with this. In my opinion, this is not a revolution. To me, this is really an evolution because you may remember me talking about Power Automate and a lot of other business process automation and robotic process automation. And that was not— I mean, it was like 2 or 3 days before ChatGPT launched. I was already teaching classes on how to do this stuff. And it's no different. The, the same concepts of how do I take a task and make it repeatable with macros in Excel, going all the way back to I don't even know how long ago, you know, macros in Excel and VBScript. We've been doing that for decades.

Yeah, yeah.

So this is really not that big of a difference. What What is different is the accessibility, is the fact that before you had to know how to do VBScript or you had to know how to do Google Script or you had to know how to even just interpret HTML code. Yeah. Even if you didn't know that, you could still like have to read it so that you could tell the RPA tool what to do. And so now it's a lot easier, but all that means is it's a lot faster. The problem is the same as before. People don't have an idea. They don't know what works, right? So the biggest problem with automation, the biggest problem with AI, the biggest problem with all of these tools is that people just don't ideate. What could I do with this?

I always say that the limitation of your possibilities is within yourself. It's not with the tool.

That's my point. Exactly right.

Yeah, it's limited in yourself. They look at a blank screen or a blinking dot They don't know what to do.

And you and I used to be like, well, we got to figure out a better way to do this because I'm going to spend 5 hours creating a VBScript for an Excel task that only takes 2 minutes because I'm going to do 20,000 times that task.

Correct.

But people don't think that. They just think, I'm just going to do the task the way I always do it, and they do it manually, manually. So that's the really— that's the biggest difference is that it's not an evolution because it's already been there. But what's different is We can now just automate it so much easier. However, how do you think of the ideas to automate, right? And that brings me to the big question that I wanted to ask you from all the conversations that we've had in the past. How do you tell the difference or how do you decide? Let me just say this. People say AI saves time, right? But how do you tell the difference or separate the flashy demo from an actual measurable productivity gain in your day-to-day, or even if it's not day-to-day, even just your week-to-week? But what is that? Is there a formula? How do you tell it's a good idea to do this even though it's investing a lot of time to do this because it's going to save me time in the long term versus, oh, I got to do this because everybody else says it's cool.

That's true. The last part is absolutely true. People build custom GPTs or whatever, or they do some prompting within ChatGPT or Claude or Gemini just for the sake of it. And then you see how much they iterate to get the real good answer out of it. And that's not automation.

And sometimes that's just fun experimentation.

It's fun, of course.

You're not going to put that into your business process, right?

And the evolution within the AI is also that you can interpret unstructured data or unstructured text with AI. So that's also the difference between our time. I automated a lot with Zapier or with Make. You did Power Automate. But it's the same.

Which used to be, you know, if this, then that.

Yeah, if this and that.

Yep.

So all the same principles, but, um, and it's more easy because it's more drag and drop and you can ask, uh, your AI assistant to, to, uh, I'm running into this issue, how can I solve it? So here's a script for you and you can put it in and it's more easy.

Oh my God, I was just doing that like an hour ago, right? I was trying to— I have two monitors.

Yeah.

So, um, and I have two machines. One of the machines is for a client and it is extremely secure. Like there's nothing can come in and out of it. But I would like to sometimes be able to have that displayed on one monitor while on the other monitor, my other one, because I work on both of them instead of having to switch back and forth. So I was thinking to myself, every time I switch, I have an MX Keys, this one. So there's a button that switches the keyboard and the mouse. So I just push one button and boom, I'm on the other machine. But I'm looking at a laptop and I'm thinking, I would love to be able to see this display on my big blind person display, right? And so I thought, if I do that, I can switch the mouse and the keyboard, but I can't switch the display. Well, I have a Stream Deck. You know what?

Yeah, yeah, yeah.

Yeah. And I use it for productivity. Yeah, definitely. So I thought there has to be a way that you can push a button and it would switch from mirror display to extend display so that I can go to the other machine and have them both at the same time or have them extend my main desktop? So extend my main desktop or have one of each. Well, guess what? There is not. There really is not a very easy way to switch display from mirror to extend in a UI-friendly way. There's no button. There's no— right? So therefore, you can't really—

It should be a script.

You can't pro— right, so there's no way to do it with a button, but with Stream Deck you can do it as a script. So I was racking my mind trying to figure out how to write a bash script that would switch the displays. I'm trying to like get the resolution and the monitoring, and then I don't know why I didn't think of this before, I went to ChatGPT and I said, how do I write a script that changes from mirror to extend. That's literally all I asked. And it was like, well, there's no easy way to do it. And then it starts going through this, explains me all this stuff, and I go, uh, can you just write the script? And it wrote the script, and it was a few lines. And I'm like, okay, I can take this script and I can put it into a home folder and make it into an executable so that whenever I push the button on the Stream Deck, it runs that code and switches the screens. So it created the script for me, but there's a joke. I put the script in, I program it into the Stream Deck, I push the button, and I fully expect the monitors to switch. Instead, the monitor that is my second monitor goes blank.

Oh my God.

And not only does it go blank, it also turns off. And now I can't see it. I can't recognize it. I can't— It's like it basically just disappeared. The machine doesn't even remember that it ever existed. So I have to go back and write a script that says, how do I force recognize? And then it gives me this process to reset. And then I went back in and I had to clean the script. So the lesson is I could have just started by asking ChatGPT how to write the script for this because I didn't even think about it. But when I did get the script for it, It broke my machine, so I still had to fix it.

Sometimes this happens. Yeah, this awful— it's also hallucinating a lot. If you're not a coder but you can read some code, sometimes you said, okay, that's going not the right direction, I want to change this. Did you talk about this, this, and this? Oh, good point. Yeah, let me change.

That's a good point. I love how they flatter you. That's a very good point.

You're such a smart person.

Good boy, good boy. It's like, no, I just want the answer, please. So, in terms of— I got it fixed and it did take me about an hour to figure all this out after I crashed the monitor. So, now I have a button that I can push. And let me explain to you how often this happens. So, throughout the day, I probably switch between my dual displays and my single display probably a dozen times a day. And so, each time that I do that, it takes me about 30 seconds. But I do that 12 times a day every single day. That's 120 times in 100 days. So, I'm saving a significant amount of time by not twiddling around with this stuff. And that's just one simple thing. That's obvious to me, but when it's not obvious, How can you tell what is the measurable, like, how do you decide this is worth me spending X amount of hours in fixing this stupid script because I'm gonna get this outcome versus, oh, I'm, you know, trying to do something that's just fancy and for flair, but doesn't really actually produce anything.

Yeah. What I'm doing with a lot of, most of the times recruitment agencies, because there are a lot of recruitment agencies in Europe. And they're way ahead of—

and speed is very important.

Yeah. And they have corporates, corporates, uh, they're way ahead of corporates. So corporates falls behind. Maybe they're doing some ChatGPT or whatever, but they don't do automation for now. So what I see when I'm doing a lot of roadmap sessions with agencies in Europe and also abroad is that we always start with the process. Yeah, stop at the process.

What are you doing now?

Right, what are you doing now? Where is it leaking? And what you want to automate? So, and make sense what you want to automate. So, um, one, one agency, um, has to deliver very fast because they are in a BPO environment, so they have to deliver very fast and they have to cut their recruitment. So time to deliver a good candidate to the BPO portal from 2 days to 2 to 4 hours because they have a lot of competition over there. And they also have, if a candidate is stacked by another, company, they will get first. So I was talking with them, how can we automate those processes? And that's really easy. That's not very easy to do, but it's really easy to grasp that idea because they're doing it every single day.

Time-consuming.

Time-consuming and losing a lot of time in the selection and and so on. And this is an easy, easy thing. Other customer has a couple of hundred people working for them as contractors. So they contract people out to their customers. And a couple of times per year they evaluate the candidates. So they evaluate with a short sort of test and this test outcome, they then have made a custom GPT for it. But then the custom GPT output has to be cut and paste into a conf file because they want to make it real pretty. So for every single report, which takes about an hour per report, so copy-pasting this, do that, and then make it a little bit more into Canva. So they take about an hour. So an hour times 100 people, 200 people, times 4 times per year is 800 hours. So that's really easy to— there's an automation. So look at your process, see where the time is leaking, and if it's leaking for 1 person or for 10 persons or 20 persons, There's also a difference. Makes sense?

Okay. Yep. So, automated for one person, not a bad idea. Maybe time wasting. Automated for 20 people, yeah. Now it's—

Yeah.

Right. Because now as an entity, as an enterprise, you might be saving only 10 seconds per transaction, but that's 200 seconds of company time, even though it's only Shelly's 10 seconds. Yeah, okay. So that's it. So number 1, is it something a lot of people have to do? Okay, or more, you know, more— does it affect more than one person? Okay, correct. All right, good. What's the next one, or do you have another one?

Um, what's your business model? Uh, so if, if it's— if it's— if the, the fastest, uh, fast, fast time is very important for you That's one thing you have to automate a lot of things because if you're not the fastest in debt, in debt in your business, then you have to try to automate a lot of stuff. So that's also one thing. Another thing is take away the boring stuff from recruiters.

So, so that one may not even be— it might not save you a lot of time, but it's boring, right? So maybe it takes you 2 seconds to do a keystroke, but less than a second to do a button. No, you're only saving 1 second, but you're doing that 500 times and it's time-consuming or boring. Like I'm thinking like Excel, you know, when— so I used to see recruiters that had to like split the name and the email. Oh yeah, yeah. And they're like doing it one by one. It's like, oh my God, that's gonna take you forever, right? Versus you could nowadays, you could just do a fill series, but whatever, you know what I'm saying? That kind of thing.

Yeah.

Yeah. Okay. That's a good one.

Okay.

Well, that's, that's 3 really good things. Yeah.

Notetakers is also very easy.

Oh yeah. Right.

That's also a huge time saver for a lot of people because, uh, A, you have, uh, your, you can really focus on the, on the, on the conversation with the candidate or with your, with your hiring manager, whatever. You have raw data out of this. You have, if you have this configured correctly, you have also an internal or a job, um, um, uh, because a candidate profile out of the, the interview you have, uh, you can have a lot of technology behind it to, to get more information for your interviewees. So for your interviewers, so, uh, hey, look at this, or people want to, uh, people, you don't, you speak too much or you don't ask too many open questions, whatever. But the main thing is also that you have raw data in the candidate database in your ATS. So you can also search those, um, those, those things, uh, if you have a good search engine in your ATS.

Okay. Yeah. So you have the notes inside of the record, CRM or whatever. Correct.

CRM or ATS, whatever.

Have you found this? Again, one of the things that I kind of laugh internally about, the whole concept of text-to-speech, speech-to-text. There was a company, I don't know if they're still around, called Dragon Dictate that was really good at this.

Mostly from blind people, I guess.

Yeah, exactly. And there's been a lot of technology around speech-to-text and Macs, for example, Macintosh has had a lot of stuff built in both from the read it to me and from I talk to you and you put it in and everything. That's been around for a very long time since before AI. And then AI comes around and all of a sudden everybody's aware of this and it's now like what you're saying, let's translate interviews. It used to be you could record an interview and send it to a machine machine to type it out. So, that's not that new. But what is new is what I found this to be, and this is the funny part to me anyway. It used to be that you would get a transcript and there would be all kinds of weird typos or things that were not spelled correctly or words that were similar, but that's probably what the machine heard. However, it's not what you meant.

Right? Correct.

Right? Now, that still exists, but what I've noticed is that the LLMs are actually really good at ignoring—

Contextualizing. Contextualizing.

Yeah, exactly. So if you see the transcript, it's still awful.

Yeah, correct.

But the machine language translation or whatever, when you get the notes and the outline, it gets the idea across. So you don't have to worry about spelling-proofing, spell— spell checking or whatever the document, because for the most part, I really am interested in the outline, not the actual word for like in the document. Uh, yeah. Uh-huh. Yes. That doesn't, that's not useful. Right. So what I've noticed is that that's really what's gotten really, really good is the summarization. Where else do you find summarization besides interview or dictation? Where else do you find summarization? I can give you my idea if you have another one. Other than interviews or calls, right?

Uh, interviews or calls, summarization of YouTube videos or websites or whole websites or profiles of people, LinkedIn, GitHub, whatever. Uh, you can have so much. Every time it's too long to read, summarize it.

TL;DR, basically.

Yeah, TL;DR is that too long, do not read.

Too long, didn't read it.

Yeah, can't read it, can't be bothered. Yeah. I've Comet as my browser, so I have the assistant in there and I can ask to summarize a page for me. It's really handy. It's not as clever as people say, so you can't do searching or whatever with it, but it's getting better and better.

Well, you would love my browser. Oh, really? Yeah, it is. It does all of that. And when you said Comet, I was thinking, is that Perplexity? Who's Comet? Yeah, yeah, Perplexity. Yeah, the one that I use is made by the browser company. And it's called Dia for day in Latin, Dia. And basically, so it does a multi-tab. So if you have 15 LinkedIn profiles, it opens different tabs, you can actually interact with all of them through the chat. It's basically a chat-based browser. So what changed is that it's really not a browser, it's a chat that opens pages basically, right? So you have all these tabs open and you can ask it to collect information, interact with it, stuff like that. You can also create custom scripts like a prompt, like you could with Claude or GPT or Copilot or gems in Gemini that is a set of instructions that you can invoke by a single command, a backslash or whatever, which do things like, for example, let's say that you normally need to make an outline of this stuff to put into a PowerPoint. You can say outline this, right? And that it knows what you mean by that. Translate this or, 'Extract this,' whatever. Getting emails from a page, or if there's— are there any emails in these profiles? You know, extract email addresses, those kinds of things. But it also does other things. It remembers on your machine locally, it remembers things like how you write, and after a few days of using it, it starts to write for you. So if you're going to reply to a message, it actually fills out what it thinks you're about to say. And that is taking advantage of one of the major things that these LLMs do, which is token prediction, which is correct. What is the next token that usually comes after this? That's really what they're doing is based on everything we've ever seen, there's a whatever percentage that the next word is going to be this. Yeah, well, that's great when it's general generic stuff, but this is looking at my emails, my documents, my So it starts to write with my same kind of weird style. And of course it does it in advance of your cursor. So it's not writing it and then saying, go back and delete. It's like you type and it puts it forward progressively. And then if you want it, you just hit space and it, you know, and then you can also backslash correct and stuff like that. So it's really useful at reading the page. Like for example, LinkedIn comments. right? Yeah. It used to be I had to actually type in the whole thing, but now I can actually, from the sidebar, I can say comment slash, and it will write the comment. Oh, okay. Under that box, it opens the reply and it writes it under the box. I can still backslash and edit it, but it's a lot faster for me to make a comment on a page by starting to type comment and then start to type something and then it sort of knows what I'm going to say, right? Yeah. So it's automating every little tiny thing like that. So it learns from me, it automates daily things that I am normally doing, and it works across pages. Now that's just the beginning. What I'm starting to do now is actually take control of the browser through the bot. And that's a little bit different. That's where I'm actually telling the bot to do something and then it goes and does it on my machine. Yeah, with, you know, including like logging into websites. For example, if you want to look something up in a password-protected website, you have to log in, assuming you're not logged in, go to the page, click on the box, type things into the box. And then you and I used to do that through URL injection, remember back in the Opera days where you would have like a slash command that it injects the variable? This is doing that, but on anything. It's, it's a lot more multi-purpose. Yeah, fast and multi-purpose. Yeah. So those are some good examples, really good examples of saving time without getting sidetracked into, you know, oh, this is just bright shiny object. Yeah. When you walk into a company— oh, sorry.

Yeah, yeah. I've also made some automation for myself. What One of them is my email because I hate replying to my email. So I made an automation for my email. So my AI reads every email I have. It automatically tags my email if it's a newsletter, is it a partnership, or is it a customer inquiry. And then based on the email, it will draft me a response in my tone of voice. So I only have to check the drafts or redo them and send them away. And the next step is that I'm doing it automatically. And I also did it from my LinkedIn, so my LinkedIn inbox, because I get a lot of inquiries. You have a guide for me? Okay. I always did a bit magical text responder, but it's also taking time. So I made also an automation for that. Because it's the same question over and over again, right? Oh, same question over and over again. So 400 times I'm on the guides. Okay, you can have a guide. But now I've automated that part. And also the questions I get in my LinkedIn inbox also with a little human in the loop thing in it because sometimes it hallucinates or I want to edit a little bit, but it's getting better and better.

It reads your LinkedIn from a browser or?

No, it's in a done and edit in. I have an API with LinkedIn.

Got it. Yeah. Right. So it's actually opening a LinkedIn page behind the scenes. Correct. Headless browser or whatever they call that.

Yeah, it's the API call.

Yeah. So the question I was going to ask is when you walk into a company and this may be a little bit of the secret sauce, but when you walk into the company that has said we want to do AI, They bought in, right? What's the first sign that you look for that tells you that their AI adoption or their AI wants are superficial rather than actually significantly transformational?

Good question. Good question. What kind of signals am I looking for if there's—

if they're really just like looking for hype?

There is a lot of AI fluff about AI agents. I want agents for everybody, for everything. There are no real AI agents now. There are agents lookalikes, but not real AI agents. That's right.

So if they say we want agents, you let you know right away that, okay, I'm going to have to do some education.

Yeah, you have to do some education. So one of the things I'm also doing is making a context book. So what does an AI agent actually is and what is an LLM. And what is— so I'm making this also available for people. So because there is a lot of noise around, so people also say that a custom GPT is an agent. I'll build an agent. Okay, let me show it. And they open a custom GPT. I said, there's no other agents. That's a custom GPT. So that's one. Um, so the— and also what they're trying to achieve with it.

So there's also a good goal, like what is a big thing that they say that they want to achieve that you're just like, oh, come on, please.

Yeah, yeah, yeah, yeah. So, um, uh, they want to get rid of all the people, but, uh, for example, I never heard it, but, uh, that could be one of the, one of the red flags for me. Because you don't want to get rid of people, you want to replace tasks. And the tasks can be replaced by automation and some automation with AI, but it's not always AI that's running the game. It's a lot of the times, I think 90%, 89%, no, let's do the Pareto 80/20 rule. 80% is automation, 20% is AI.

Okay, that's a, that's a really good way to apply that formula, the 80/20 rule, or the Pareto principle by the Italian economist Pareto. I'm explaining for, for the audience, um, it basically, it was an economic theory when it started. It was 80% of the wealth is owned by 20% of the people, but it became sort of a golden ratio of the multiplier effect. So 20% of your activity produces 80% of your results, for example. So what you're saying is 20% of it is actually AI and 80% of it is the automation. That makes a lot of sense. That does make a lot of sense. Yeah. And of course, that lines up with what my experience, which is really not a whole lot of AI. It's just a lot of automating tasks. And in my case, it's being able to push a button that initiates the task, but the task itself, is automation. It's not really AI. It's just that I now have a button that I can push. That was the AI part, as in the ChatGPT that could just spit out the code, right? It spit out the code. I still had to fix it, but at least it did that where I was spending too much time trying to write it. So that makes a whole lot of sense. Yeah. Um, when they say we want to replace everybody with agents though, what, what is that? What does that really mean?

They don't know what they are talking about because—

but yeah, I know that, but like, is it— are they saying we want to save money, or are they saying—

yeah, they want to save money, or they want to be a part of the AI hype, or, um, they want to do more with less people. And that's all— that's also fine if you want to do more with less people. That's— that's—

we've always wanted to do more with less people.

That's why machines were invented.

I'm pretty sure the wheel was invented because somebody wanted to do more with less people. Correct. Um, and in my opinion, this may be controversial, but I think that what we call AI now, this, this whole automation and, you know, large language models that are, and machine learning, which is what this really is, is, is really not that different than when the internet came around. You know, business was business until the internet, and then when the internet came, there were a lot of people that said The internet's going to take all the jobs. The internet's going to destroy the economy. The internet's going to take over. So many things that were fantastically good and also horrible were assigned to the internet, right? And eventually, it's now basically the same as electricity or water. The internet is something that we use in order to do our job. It's not what used to be sensationalized. This is the same thing. It's to say, oh my God, AI is going to replace. What's going to happen is eventually AI will be like the internet or electricity or water. These will be things that are built into everything that help you do stuff faster because it's more efficient. It's literally, that's all it is. It used to be you had to put a letter on a pony in order to horseback ride all the way through the desert to get a letter from one place to the other. And then somebody said, what if we go with— and then now it's, you know, Telegram. Morse code. Morse code. So, did that revolutionize things? Not really, because we were still writing letters. They're just shorter and faster. It just made the world smaller. And then the internet made the world smaller again. And now this is making the world maybe bigger, but it's the same kind of change. It's not the same kind of change.

The only thing I'm not worried about, but it's more the fast pace it's going right now. So it's replacing the task more faster than we can replace it with new jobs.

So you're saying that it would be like if the internet took more jobs than what it— than the— because the internet also generated a lot of value in the economy.

Of course, of course. It generated a lot of value in the economy, but it came more gradually.

Ah, so this is just faster.

This is faster. It's really fast.

Well, there's a, there's a theory of economic evolution that says that anytime a particular cycle or jump in advancement of some kind, in economical advancement, for example, going from a hunter-gatherer society to an agrarian society, every time there's a change like that, whoever does it next, so the next society, or economy to do that change, for example, here in the US, going from an agrarian society to an industrial society, let's say that that takes culture A 100 years to go from agriculture to industry. This theory says that the next time another culture, another society does that will be 50% reduced in time. So, if it took country A 100 years to go from farm to factory, It will take country B 50 years to go from farm to factory, and country C 25 years to go from farm to factory. And this has been proven many times because that change was gradual for that country, but for the other country, they learned from a lot of those mistakes and improvements and did it faster. So, the argument could be it took 20 years for the internet to change the world. Maybe it'll take 10 years. Literally just making up numbers for ChatGPT. So it will be faster than the internet. My question though is, it's not really a— I guess I'm asking for you to predict the future. Yeah. Is this— Look at my glass ball. Exactly. Yes. Magic ball. Is this right now, this is the change, or is this, in my opinion, this is the fetus that will grow up into the change?

This is the change for certain jobs. So, um, any white-collar jobs, for example, consultants, accountants, but also customer services, translators, writers, content makers, content engineers— sorry, developers, software developers. Um, uh, what do you have here?

More So for them, this is the change.

This is the change for them.

There's not another one in a few years. No. Okay. No, I don't. But for other industries, there's a bigger change coming.

Yeah, of course, because this is now for text or for image or whatever. Right. Context, text generation, it's really a text generator. But also, the humanoid robots are also replacing some jobs within the factories. And sometimes it's good because the factory is too unsafe, unsafe, or in a nuclear reactor or whatever, you want to have humanoid robots instead of actual people. So but it's also going to change a lot of jobs. So, and the, the fastness where it's going with— and maybe I'm explaining myself not correctly because English is not my nature language— but, um, the fastness where it's going, that's, that's— the internet was a little bit more slow, disruptive, but more slow. And this is changing more tasks by automation and AI or large language models.

Got it. More tasks, broader influence, faster than the internet, but it's the same kind of principle of change. Correct. Correct. What do you think is— I think that the hype of the internet was tempered a little bit because some people were like, you know, that's just a lot of hype. And they didn't get absorbed into it. But in this case, I feel like we have as a society been hypnotized. And the hysteria, you know, hysteria. Yeah, yeah, I know. Yeah, the hysteria and the rapidness, the easiness in which people are just accepting these big claims is so bad. If you told me in 1997 that there was going to be an internet that basically created an entire industry, the Silicon Valley dot com, I would have been like, I don't know, he should, yeah, but I'm not going to just believe that, right? Yeah. But now if you told anybody on the street that there's an AI that is already curing every cancer in the world, they'll go, oh really? Yeah. They don't even think There's like not even a, hold on a minute, that's not possible, right? So people are just believing universally any bullshit that people say. Yeah. But with the internet, some people are like, oh, cool. And other people are like, I don't think so. Let's wait and see. With AI, everybody's like, oh my God, I can't believe it. So the other day I saw a post that said AI is dead. It's not artificial intelligence. It's now sentient intelligence. We have a new tool that's better than AI because it's sentient. I'm like, listen, dude, AI doesn't even exist. So you're replacing it with something that also doesn't exist. Correct. It's not really AI, right? But people are just seeing that and they go, wow, look, this is revolutionary. Why are we so hypnotized? I mean, to me, it feels like this.

It feels like look deeply into my eyes, you know, and everybody says, oh, there's a good YouTube documentary about, I think it's in Dutch, by the way, but it's a Chinese— I have an API that can translate. Yeah, yeah, you can translate. A Chinese professor who also predicted, he also worked in America, by the way, so he's not not real Chinese, but he also predicted the financial crisis and some other things. So just to give some context around it, he says that it's more red rice, more red rice to get more money, to get more money. Got it. So eventually there are some big tech companies here, most of them in the US, but also in In China, they are in a rat race to get more money for this. There's no AGI. There won't be any AGI for the next 5 or 10 years. I agree with that.

Not what they call an AGI, artificial general intelligence. That's for folks that are listening. Not everybody is in on the lingo. No, that's okay. I was just going to explain just briefly, an artificial general intelligence is different from what we call AI now, which is a chatbot. In that the AI that we're selling nowadays is basically the equivalent of a Roomba that can go around your house and sweep everything on the carpet because it's learned your house. It knows where everything is and it knows how to navigate and it's smart by learning. But an artificial general intelligence is a Roomba that can do your homework, take the kids to school, do the laundry, and write a dissertation. Correct. That doesn't exist. It may someday, but not right now. So anyway, just for folks that were— Can we talk about money? I like money. I don't have it.

I like also money, yeah.

Yeah. Unfortunately, I wish I would have more, and I can't really figure out ways to make that happen, but I do like money. What are the— hidden costs of this trend, this adoption that people ignore? For example, they say we want to automate and then you start coming up with real solutions that really automate and they realize it's getting expensive and now they're like, ooh, that's a very— do we really want to do that? So what are the hidden costs that people just don't even consider? Because they say, well, we're going to automate and we're going to eliminate 3 jobs because we only need 1 person to do this now because of automation. Yes, but how much did it cost you? Correct. Can you tell us what those are?

Or not all of them, but like, uh, the automation tool you use, uh, the subscription.

Yeah, there's more than just the purchase price.

Yes, more than the purchase price. You have to maintain it, you have to, to feed it, you have to, uh, if it's broken, you have to put it back to life. And you have to pay somebody else to do it. Yep, correct. So that's one of the costs, the tokens you have to pay. The tokens is the price you pay for every word you say back and forward with the large language model. So if you put something to ChatGPT or OpenAI, you have to pay some price for that. That's not a huge price, but if you do large pieces, it's going to be more costly. What else? Voice agents. We didn't speak about voice agents. There's also new voice— not voice agents, falling in the trap myself. It's large language models with a voice to it. So they can interpret that voice and they can speak back to you at a real natural language with even, uh, and everything. So there is a huge potential for barbershops, restaurants, uh, what you have. Uh, so, uh, also for hospitals, by the way, because hospitals have a very, uh, high rate of not showing up. Uh, I didn't know how you call this. But people don't show up with the appointment. It's costing a lot of money for the hospital. Yep. But also for a barbershop and for a restaurant. Yeah.

You have an appointment, that person's not going to do their job that they get paid for. Correct. Show up.

Yeah. And you can make a real nice automation around that. So the day before, you call the person up, you say, hey, this is the AI assistant of this and this barbershop. 'You have an appointment tomorrow at 9 AM. Are you still coming?' 'Yeah, I'm still coming.' And then you have more certainty that it will come. And there's automation with some large language model which can speak with the lab or whatever. So that's also a hidden cost, but it saves you money.

If it's a cost that turns out to be a savings, then you're really just investing. I'm thinking about the cost that you you don't really see that end up costing more. Like, for example, they want to implement this new AI software that's doing stuff, whatever it is, you know, posting jobs or sorting resumes. But, and they think, "Oh, we're going to save time and money and, you know, do faster. Maybe we need fewer recruiters, whatever." But somebody has to program that. Somebody has to maintain it. Somebody has to supervise it. Somebody has to fix it. And somebody has to train the people that are still there on how to use it. And they're going to have to be training those people over and over and over and over again. Correct. Just those people, but the new people and even the same people that forget. So there's all of those costs. And then there's also the costs of how do you connect this to the other things that you have to do? Because it might be great that your AI can do this, but what if it doesn't talk to the ATS? Now you have to pay somebody to make it connect to the ATS. So people just don't think about this and it adds up.

Yeah, that's also what I'm doing in the roadmap. So it's not about, uh, okay, we can do this, but what's— what are the costs and what's the actual cost? So, uh, so, uh, what is to maintain, right?

Uh, sure, yeah, we can do that, we can automate everything, uh, but it'll cost you $75 million. Oh, okay, maybe I don't want to do that.

Uh, but I also see a lot of small SaaS solutions where they pay a lot of money for this and connecting everything to their ATS. They're not looking at their core ATS or CRM, what they can do with their core CRM or ATS. They don't really use that well their own software and they just plugging in some other stuff to get something work. But eventually they have to change their process sometimes because they want to have a specialized process because they are special. But that's not always the case because the ATSs make a process for the majority, and most of the time it will work. But sometimes you have to, to look at your own process and maybe you have to change the process a little so you don't have to spend that amount of money to do additional tools or whatever. And you have also a lot of AI tools which can solve some little tiny thing, and before you know it, you have a whole AI or application and AI landscape that you who don't know how it works, they have a really good big problem.

You know what, that just— I just thought about that. You like cars, we've talked about this before. Yeah, right. And, uh, it's, it's like the— what, when I was, when I was still a teenager, I actually built my own motorcycle. Um, I took the frame from— yeah, it was fun, it was a fun project, and it was a small one. But I built— I borrowed the— not borrowed, I recovered the frame from a wrecked motorcycle. So someone had a motorcycle that they wrecked and everything was broken, but the frame was actually in perfect condition. So I just basically— they were throwing it away and I'm like, yeah, I'll take the frame. So I got the frame and then I bought an engine. The engine was from a different frame, but I refitted it. So I put it all together and I put all the pieces together. And it ran. And every time that that motorcycle broke down or something was wrong with it, I was able to fix it while I was on the motorcycle. You know, like, I would be driving around, it would stop working, I could just literally just fix it. And, you know, because I knew everything, because I put it there, right? Yeah. The last motor— that was the first motorcycle I had. The last motorcycle that I had was a BMW Cruiser. Everything in it was completely sealed. I couldn't even get in. And if I tried to open it, it would void the warranty. So that's what you just described. When we used to do VBScript on Excel, we knew how it worked, and if it broke, we could fix it. But now it's a black box. We don't even know where to even start fixing it. Correct.

That's scary. Could be scary, but if you make the right decisions and make the right automation with the right automation tools, then you know how to— it's not always about fixing things, but it's also about documenting, updating. Oh yeah, right. And the documentation here in the European law is also very important. So if you use AI, you have to document everything.

But it— right, because that way you can go back and figure out where it went wrong. Correct. Right. That's smart. That should be a requirement anywhere. You know, I remember when it used to be hiring software engineers in, in the early 2000s, and one of the, one of the interview questions had to do with, with REM statements, because a lot of people just didn't even comment their code. And so the code did the stuff that it was supposed to do, but the next engineer couldn't even figure out what it was doing. So it was like it started becoming a requirement that you have to explain your code in the code so that future people can read it. It's kind of what you're saying is we need to do that with AI. We need to explain the AI while you're doing it so that somebody else can figure out. Correct. Yeah. That's a requirement, man. But I don't think a lot of people do that. I think they just turn it on. And they go, "Hey, we rock!" "It works, it works!" Right? 20 minutes later, "It stopped working, why?" "Oh, LinkedIn changed their interface." "Oh no, not again." 5 minutes later, "They changed it again." So anyway, I did have one more question. This has always been— talking to you is always fun, it's always very expanding my mind. But I had one more question. Thank you. Thank you. What pattern do you see that distinguishes a team or an individual that can take AI into actual placements versus people that just, you know, they end up using the tool, but it doesn't really— I mean, maybe it helps them do business faster. Maybe it saves them money. It does a lot of things. It helps you it's more efficient, et cetera, but it doesn't actually turn into more revenue. Like, what is the difference between someone who takes AI and makes more revenue versus someone who takes AI and just works faster?

Is there a difference? Two different questions, I guess, because first you said which companies or which—

I guess some people just do AI because they're lazy and they don't want to do the work versus other people do AI. That's fine.

Yeah, it was always good.

That's it. I'm raising my hand, uh, you know, very, very lazy here. But what I'm asking is, what's the difference between me, the lazy person, and someone who says, I'm gonna do this so I can double my revenue?

Uh, that's a different, different angle. Yeah, so there's no difference.

I mean, in the person. Okay, okay.

Is it different in the person? Because sometimes, uh, CEO or director of an agency wants to double the revenue or want to be the smartest recruitment agency in their space, or they have some, some sort of ambition and they want to use AI to reach that ambition. Then always downplay it because, okay, let's look at your process. Why we have to look at a process? Let's look at your process and then we're going to figure out which we do the first 90 days, then for the next couple of months, and then afterwards. But we can have that ambition. Okay, I want to have a hiring assistant next to my people and they want to solve everything for them. Okay, that's a good ambition, but that's a good ambition for maybe 2027 or 2028, but not for right now. First you have to fix this and you have to keep that in mind that this is your biggest goal. Are you big, heavy goal? Okay. And then you have to downplay it. What is the first 90 days? And also use an action priority matrix. That's the Eisenhower matrix, but that's the action priority matrix. Urgent, important, slow effort, high effort, high effort, low effort. And then low gains, high gains. Yep. So, and the high gains with low effort are your quick wins.

So I think I understand better my question now, and basically it translates to it's not necessarily the difference in the person but the difference in the objective. So Sometimes you have an objective that is really— you think it's about saving money, but it's really saving you time. Obviously, that may save you money, but you're really looking at things to automate because it's less work, not necessarily by profit. And what you're saying is first, before you decide on what to automate first, or what the first thing you do when you decide to automate, the first thing you do is figure out which automations or which improvements are going to be the most impactful in the shortest amount of time. Correct. And then you have some wins. And then, yeah, so, but if you focus on that long-term one, you might be building the castle for 100 years before it's made. And no, nobody— Correct.

Yeah, you're gonna be the, uh, Sagrada Familia, uh, in Barcelona. There's a huge, uh, church over there which is built already 100 years and still building it, right?

Nobody can use it until it's done. Not in that case. So the, the reason I asked the question— so that changed my question, but I got the answer I wanted. The reason that I asked the question was because I was thinking about the scooters. Amsterdam is— I know you don't live there, but you and I have been there before. So, it's a city where a lot of people choose to walk and a lot of people choose to ride bicycles. More people choose to do human transportation than cars there than in a lot of other cities. Maybe it's the design of the city for whatever the reason. But in cities in the United States, there was this like boom, which is starting to now fade finally, of these scooters. They're like little platforms with two wheels and a handle. And they put those everywhere in downtown and people were using this. And the reason I'm asking the question is most people use that because they didn't want to walk. Not because they would be making more money if they got there faster. So, it became this luxury that lazy people who didn't want to walk were using just for the laziness when people who actually needed to get somewhere quickly couldn't use it because the lazy ones were using it. And it also became the lazy ones would leave them laying around on the floor, in the entrance of a building and stuff like that, where the people who really needed them to get somewhere quickly, they would be the ones that actually would park it, make sure it gets charged. So, it just became a problem where so much of the laziness happened that the city even had to start passing laws against it because they were getting in the way of wheelchair access, they were getting in the way of cars entering driveways, they were tripping people, causing accidents. And so, That kind of automation really fed the worst part of us instead of the best part, which is I could go to a lunch and come back in 10 minutes instead of half an hour because I can get there faster.

You see what I mean?

Yeah, that's why I was asking.

It's all about the objective, and most of the time If you speak with the directors or with the senior management within an agency, they have a pretty good idea where they want to go, but they don't know how to do this. Right.

That's what we were talking at the beginning. They don't have the idea how. Well, we have arrived at the end of our time and I don't want to keep you past, but there's always a question I ask all my guests every time, and that is, who should be our next guest?

Oh, good question, good question.

Somebody you would like to see, um, doesn't mean that they will accept the invitation, you know, but whoever you think is going to be a good addition to the show, um, I will reach out and ask them because, you know, um, I love to see Clengeti again.

Okay, I know he was on the show before, I guess. Yeah, I got you.

Yeah, I'll write that down.

But, uh, anybody else?

Uh, you don't have to give me 2 people, 1 is more than enough. But yeah, yeah, okay, excellent. Well, thank you so much for showing up at your Friday night. You have better things to do on a Friday night, so thank you for being here.

We all appreciate it. Nurture my wife now.

Yeah, that's right. Close the business and make some dollars. All right, my friend. Well, I will see you hopefully. I was going to see you in Amsterdam for SOSUR, but it doesn't look like it's going to happen. So maybe we'll find another way to meet or we'll see you at ERE or something.

Yeah, I'd love to come again. Yeah.

Well, thanks everybody for being here. Come back next Friday. We have Ben Solomon as our guest. Also very good. Yeah. Thank you. Yeah. And until then, you know, be kind to yourselves.