OpenAI Agents 2.0… Is This The End Of Google Search?

All right, welcome to the show Today, everybody. We have a hot topic because we have some breaking news from OpenAI. They are doubling down on agents, and we're going to tell you what that means for your business and how to.

Take advantage of it.

In the early days of AI agents. Kieran, you are here. You're in the house. You are a new parent. Congratulations. I've been doing some solo shows, We've had some guest hosts, but you are kind of making some appearances when it works for you. And this is your first appearance back post being a parent. How's parenthood going?

I have a whole story to tell, but I thought. I thought I would save it until we were going to do an episode, I think, on how to build a demand gen engine. And I have a great story to tell that speaks to AI, of course, because I have a really interesting AI use case that I use during five days in hospital, which. Which some of that was me being in hospital. So there's the cliffhanger, and I thought we could spend some time on our demand gen episode going through that. But parenthood is equal amounts of awesomeness and equal amounts of hardness.

Equal melts. Awesome and hardness. You heard it here.

Like, there's equal amounts of. Wow, this is super amazing. And then, God, like, everything is not on your time. I mentioned this, remember I mentioned this in our last episode, that your time is not your own. So, like, I sat down to write something this morning, right, And I couldn't because Hope, who's our baby girl, she wanted to wake up and I. I had to, like, look after her and calm her down. So that's the. That is a big adjustment that your time is not really your own anymore. Right. You have something else that actually controls your time in a lot of ways.

And when you do have your time, you have to be very focused and, like, get it in.

Right, right. Totally. But good. I survived it. It's an interesting week. You know, the week I. I had, it was, like, the most incredibly draining. Weak. But I will tell the story in the next episode.

Stay tuned and tune in for some of the next couple episodes that Kieran's gonna be on because he's got some wild stories from his first couple weeks of parenting. But today we're gonna do you what's happened over the weekend. And I was like, oh, I think we should do a show about this. And you're like, oh, I'm gonna join if I can. And so we're here. And largely, I'll pull up what you originally shared, Kieran which is the email from Ben's Bytes, our friend Ben Tossel. A great AI email newsletter, Ben's Bytes. If you don't subscribe, you should check it out. And Ben broke down that OpenAI's got a new project and he cites an article from the information that OpenAI is building, two types of AI agents. And these agents are going to be able to perform tasks basically on your computer for you, right? And we're probably going to see them show up in lots of different forms. But I wanted. Kieran, now that I've got you here, when you first read this, like, what was your take? Like, what's just the immediate? You're like, oh, I think this is interesting. I think all of our listeners should know this.

So there's two really interesting things going on. First of all, so there's two types of agents, right? There's one that will sit in the background, learn how you do things on the laptop and then on your desktop or on your maybe at some point Vision Pro, and learn how you do tasks and they can actually do those tasks on your behalf, on your path. And I want to come back to that because there's something really interesting going on there. And then the second is like an agent that's going to be able to not just search for things like flights, but actually complete those tasks for you, like book those flights. And this is the thing people are really getting wrong around the evolution of AI chat versus search engines in that what we think, I think we're trying to compare is ChatGPT and these kind of things a good replacement for search, where they're actually something very different. We actually may not need search anymore, right? Like actually we. This may actually usurp that experience. Let's start with the AI agents being able to automate work, because this is like really interesting for I think all of us, all of us knowledge workers could potentially have these agents and it would learn how we do things day to day and be able to complete those tasks. Now what's really interesting is this is Adept. Remember when the very first companies that you and I looked at and thought was super interesting, way back when was Adept, the founder of Adept, used to lead engineering at OpenAI, right? So there's like some, some real similarities between or some connection between those two things. And I think this is the evil, the natural evolution for OpenAI is to be able to just like ChatGPT is really incredible that I can create text and create all these things within the app, but I can't do anything with it, right? Like, it doesn't, it doesn't go on and actually do anything but the things that it creates. The next evolution for OpenAI is to say, well, we'll just do those things for you, right? The demo they showed, or what showed up in the press release was taking data from one sheet and parsing it and putting it somewhere else, right? Just like some basic automation work. And to me, that is an incredible evolution of what we expect from something like a chatgpt, right? That it doesn't just give us information back, that it actually will execute work on our behalf. Now we. I'll pause there, pass it over to you. But like, there's a whole other segment we should do here. But like, like, because. Because what? What? Like, let's say I say, hey, like, go take this data, add it into my hotspot instance and then do this thing. Now, what wouldn't you want to happen that does happen today? You would not want that bot or agent to hallucinate where it takes data and says, oh, I'll just email that to your contact list, right? Like randomly do this thing that you were not expecting, right? So the actual privacy here and the ability to do things a hundred percent right, 100 of the times, that to me is going to be one of the blockers. But this is a game changer, right? This is huge change. This changes how we do work, right? That's how big this thing is.

Well, right, so, like, I love Adept's positioning on this, right, which is an AI teammate for everyone. Adept is building an entirely new way to get things done. It takes your goals in plain language and turns them into actions on the software you use every day. So if you're a knowledge worker, what a depth goal is, is like, oh, we're just going to do all the things you do in HubSpot and Zapier and all the different tools you use. We're going to do it for you, we're going to do it better, we're going to save you time, right? And that's a pretty valuable proposition. And what we're seeing is that OpenAI themselves are going to get into this game. And part of the reason for this, and I think what's important is that we talk about the why here. Large language models, which we've talked about a lot in the past, are going to keep getting commoditized. And I think, Kieran, you and I, for a long time, and we're not in the minority perspective. I've always thought large language models are largely going to Be probably more open source long term, more commoditized, lower differentiated and really the data and the use cases and the agents are around them will be what you really have high level of differentiation on. Right, right.

I actually think the large language model, there is no economic value long term. Like if you were a value in OpenAI, it cannot be based, I think on GPT. GPT4GB5 will come in, maybe they're like six months ahead. But over time the costs and the economical upside there just erodes. And so I think building these consumer apps like AI agents is a really good reason, is a reason to build these kind of consumer apps that sit on top of these large adword models now, how defensible they are, like how easy it's going to be for other companies to build them. I will tell you this exact use case because I do a bunch of angel invest and I know you do too. Yeah, this is like the number one use case that I've seen pop up over the past six months in terms of like pitch decks that I've seen which is hey, like we'll automate work based upon monitoring things that you do in your browser and be able to like easily build, easily build workflows. But I think we've seen it time and time again. I think you pull out a stat for GPT 3.5 in terms of the costs. But the costs here are going to go to zero over time. Like when you add Google, Microsoft, Amazon, they're all invested in these large language models. I just don't think there's any upside in having the best model. And open source, I've made a bunch of investments in open source companies that are.

You love open source.

Yeah. Going to make it really easy to like take a model, fine tune it and use it for specific use cases. Right. Like fine tuning the model using proprietary data and these other things for a specific use case makes it much more valuable than the base model. And so I think that OpenAI is probably going to go down the path of building some really great consumer apps on top of that model.

I think that's really right. And so what I'm showing here now is the. If you're watching on YouTube, you're watching the API documentation for how assistants work from OpenAI. A couple of things I would say, Kieran, is that it feels like the world hasn't fully adopted one name for this technology yet. Some people call it agents, some people call it assistants. I would just say if you're, you're watching those are used interchangeably like the like. There probably is some subtle nuance, but generally I think in the world now, people are using those two terms fairly interchangeably. Right. The other thing Kieran is kind of picking up where you were just talking about, if you're somebody out there who wants to start preparing for this technology, start using this technology. The early iterations of this technology are going to likely come through the form of browser extensions. Right, Karen, and can you maybe break down for people what the heck a browser extension is why they should probably get up to speed on using browser extensions because that's where agents are going to likely live, at least in the short run.

Yeah, I suspect I actually tried to dig into this when I first looked at Adapt because I thought they were such a killer company. How is it going to be able to understand how to complete these tasks? And again, how does it understand how to complete these tasks in a way where you can be very confident it's not doing something with. Cause these are, these are enterprise use cases, right? So like work based data primarily it's going to be something that sits in the background, monitors how you do work and then builds like, builds these workflows in the background, right? Like, so it's monitoring what you're doing through the day, learning how you do your work. And in the background it's building these customizable work workflows that can be templatized and say, hey, like you want to do this thing again, right? And you can do then through natural language and you'll probably do that through browser extensions where it's actually able to control the clicky clicking and click clicking and moving around of your browser to be able to complete all these different tasks. That's how I think the first one goes down. And we haven't even got into the second one, which is like the consumer version of this. This is really the enterprise version. Right. Like when you look at what they're launching the two forms of agents, there's one for consumers, which I think is an extension of ChatGPT. Maybe it can be a premium feature of ChatGPT, like actually have this feature that allows you to search and do work. And then the first one we're talking about really is the enterprise use case. And this is the use case. We did an episode really way way back when Adept first launched and this is where we started to discuss, well, what happens if this is the way you use.

It got dark.

I still haven't really figured that part out. Which is like what happens when, like if you take an app like Calendly, which is a great app. But like I, I, if I never have to go into that app, I just say, hey, like book a, book a meeting. Whenever suits would kip on my calendar, right? I just do that through my AI assistant. I never have to go into any of these apps. I could be using any app. Right. That's the thing is like how does software gets moved one layer deeper? That's problematic.

Let me run this by you because there's an outcome of this which is not super awesome. The best window of this is probably to look at the life of a sales rep. Right. What happens if you're a sales rep right now in most modern companies is that you have an inbox of tasks that you go through. Right. Prospects to reach out, automation sequences to check, all of those things. Right. And we all might live that life.

Right?

Because you're like, hey, I have all these agents, these agents are essentially doing jobs, but I'm going to review the queue and like, are they doing the right thing? Was that actually what I meant? Basically the approval and fine tuning of like, oh, this is or isn't what I wanted. And based on that, the model will get better and the agent will get smarter and be able to do it better the next day Time.

Right, right.

But there's, there's less building and there's more like reviewing and tuning, which I don't know if I like, we are.

Training ourselves out of a job, like totally very possible. We're going to teach AI assistants how to do our job. For us now, it still isn't like strategy, creativity, all these different things are going to be still unique for humans. And I actually, if I get to work on those things more than I have to like figure out how to use tooling, then wow, I'm a much happier person. And actually just one clarifying point, people may be like, is this not just like custom GPTs? The difference between something like this and custom GPTs is when you're, you can do some of this using custom GPTs. If the enterprise or the software app has an API. Right. It's done through APIs, this is done through the browser. So you don't need to. Not all software apps have APIs. So this is, this is different because you don't need that app to have an API. It's done through learning how these things work through the browser.

Yeah. If you're familiar with like Selenium or any of these like kind of dev testing tools, like they're doing a lot of this now. Right. Where you're writing programming for them to go and test things in a browser environment. We're talking about an agent that's going to go and do things in the browser, not via API.

The second one is the evolution of the natural evolution of search. So there was a demo back, I do remember, like, the Google Duplex demo back in 2018, where they would actually have an agent go and book a appointment with you through a hair salon, and it actually would call the hair salon and it would book the meeting for you. And the whole experience was done through AI, inclusive of the voice. Right. And we. I don't know. I can cut here real quick. Do you want to play that demo or not? Because I have it up.

You can play it.

So this is the. This is the demo back in 2018. I just want to play a wee bit of this. Hi, I'm calling to book a woman's haircut for our clients. I'm looking for something on May 3rd. Sure, give me one second. So, like, the entire thing is done through AI. So why did Google not release this in 2018? Or, like, they had some of these agent things working back then. They were really afraid of PR backlash. Right. That's one of the problems that Google have always had, is they have to adhere to, like, what the general public think about them. But this, to me, is the evolution of search. Right. I think Google are now starting to integrate this stuff because they've seen OpenAI are going to do this into search. So we don't really have a search engine. Like, I think the search engine naturally becomes a kind of task execution engine or a, like, task rabbit engine where you can search for something and have it do that thing for you and never have to, like, go do it yourself.

Yeah. You're basically saying it's what Kieran, you're advocating is, like, it's just gonna be a completely different search experience because you're gonna have all of these very niche agents that are specialized in something. Right. And they're gonna go do that, and they're going to be the searchers, not you. And they're going to search in whatever way that they're programmed and built to do. Right. Which is a very different way to approach search. I don't know if you saw this, Karen. I wanted to bring this up too, since we're talking about.

Yeah, I have this. I actually have this on my screen. I was.

Of course.

This was awesome.

Well, so, yeah, so we got a lot of episodes coming up about the future of search. First of all, this is more like agents and we just wanted to spend a couple minutes around like how agents are impacting search. But over the last couple of days, one of the things that went around was internal emails from Google in 2018 around how basically what then was machine learning. What became large language models were going to disrupt the core Google search business. One of those core Google disruptors is this company called Perplexity AI. This is their CEO Arvind and I thought his commentary on this was really interesting. In the first few weeks of launching Perplexity, an early OG Google employee told me the same thing as this. Don't worry too much about getting lots of users to compete with Google. You're living in an age where unsupervised learning from raw Internet text works. You don't need as much clickstream data to build a good index and ranking and therein lies your opportunity.

Exactly.

Explain to everybody what that actually means.

So Google's like, why, why did Google win search? Because they built an incredible feedback loop through page rank algorithm, right where you could actually have signal from what people link to and other things to determine whether that was the best result or not. And it truly was a huge evolution in search experience and no one ever really came close to them. I think what the engineer is saying is that clickstream data and that kind of feedback loop is no longer necessary when you can actually have an AI parse all of this text and probably give you just as good an answer. And that means you have removed a bunch of things that Google made that made Google unique. Right? That was Google's priority way of building an incredible search engine. Now that's no oversimplification because there's such evolution in the algorithm that there's many different factors and how that works. But, and they still, they, they use things like did the person click, did they bounce back like all these different things to determine whether that's the best page today. I think in the future what we're saying or what the competitors would say is you don't need as much of that because AI can actually parse and give you the answer that you want. And probably actually over time these assistants can ingest data, they're learning about you and give you the best answer for you. And, and that removes the competitive advantage Google had. And that's put, that put Google in a, in a competitive environment where it's going to be hard to see how they can differentiate themselves from these other competitors who have less economical overheads, have the freedom of not being a public company to like, fuck up a lot more and to make, like big swings a lot more. And I think that's what the engineer. The engineer basically used the experience that they had seen from translation and how translation completely changed overnight through AI and extrapolated that to say the same thing is going to happen to search and Google's advantage is going to overrode. And it was actually pretty incredible. They really showed the listen, because this was 2018 and he said, over five years and we're 2024. And that person, he or she was very, very right.

Well, they were very right. First of all, the second part of this is that the whole search category as we're talking about is changing.

So.

So Google happens to be the leader, but anybody in search, even if you're looking at something like Bing Search, right, Its competitive advantage is decaying. Right. Because the way search is just going to work and the accessibility of information has just changed so dramatically over the last few years. And so you're going to have a lot of different search experiences. And what we're talking about on today's show to kind of get back and start closing things out is we're talking about agents and their role in discovery. Because the second you have an agent start doing tasks for you, they're going to have to search and access information as part of that work. Right. And so instead of a giant search engine, you might be searching across 5, 10, 15 different agents. Right. And that is a very, very different thing as a company, as a marketer. Kieran, how do you think like, that just is going to impact how your product and services are going to be discovered in the future?

I don't know. I'm still curious about how the agents will recommend one product over another. Actually, I had a pretty interesting experience last night where I was trying to get it to recommend a health product to me based upon, like, very specific needs that I give it. And it actually wouldn't recommend any health product because it's. The OpenAI is actually having to adhere to, like, laws a lot more. Right?

Yes.

And so it's like, I can't do that. That's against my guidelines to like, recommend one health product over another. And it's not all products, but it's still not like it's trained on the Internet. Right. And so it's trained on the amount of content and what people are saying about your brand. And that's how it kind of distills the answers and gives you back something. But I still don't really understand. I don't think we really understand how it recommends one product over another. And so all you can do is continue to like build your brand, build the amount of great things that people say about you online. But for a marketeer, I think it's going to be a curious time to understand if people are using LLMs or these kind of AI chat products to figure out what products to recommend. I think it's going to be pretty difficult to figure out why your brand is not on like the top five that you ask. At the polar top five. It's not as easy to understand like why you're in or out of the.

List and monitoring and just seeing where you're at across all of these models and the agents that access them is going to be much harder.

Yeah, yeah, you can pull them like I do think it's good to start to pull them every month and keep a record like just like a search.

I agree with that. That's a good, just like go do from the show.

Yeah, just pull them every month. Look to see like have a series of questions. I do wonder, like one other thing we should get into at some point is one thing I've been thinking through actually someone from Zapier brought it up in terms of like how we actually do brand analysis. Right. Like how you actually, you know, we have the traditional ways of trying to figure out how people perceive our brand and we do them through these kind of ongoing trackers and studies that third party agencies usually do on behalf of companies. Does the. Can you pull the LM in the same way? Like could you have a series of questions and ask the LLM and if the LLM is trained on the Internet, could you say, well that is the perception of my brand or not. Interesting things.

True.

Yeah. Like so you can pull it to see not just are you in the top X percent of products that it recommends for certain queries, but also pull it to see like how is my brand perceived? Right. The Perception 1 I am not certain enough right now. I actually have a large question mark over that. Because can you distill perception from what exists online about you? It's kind of hard, right? It's pretty hard. And that's why you have these focus groups and these brand studies to actually tell you one way or the other. But there's something there like over time where maybe the way that we can measure the sentiment around our brand AI can actually help us do that.

Well, first of all, I think that that should be like two future shows or maybe one combined. We should do a future of like how you could use AI for brand sentiment, maybe do some testing and a little research there. The other thing is, like, just how you actually measure brand, I think is a topic that we haven't talked in depth about this. We've got, like a new modeling and methodology we're about to roll out. Kieran. So at HubSpot. So, like, once that's finalized, I could take everybody through kind of the methodology and the rationale behind it so that anybody could go and adopt on it or give us feedback of where we think we're off. But I think there's huge opportunity in brand, regardless of anything else over the next few years, as so much of online marketing as we've known it is going to get disrupted.

Yeah, we should definitely do that. I've spent so many time Talking to different CMOs and things about how they measure brands. So I think it'll be a great. I think a lot of people would, mostly founders would like to understand how their marketers should measure the impact of brand work.

All right, Perfect. All right. Anything else on agents before we close out today's show? Karen, anything?

No, I think. I think there's.

We're saying you go use some Chrome extensions, maybe read the OpenAI documentation. What are our other go dos?

I think start to look to see how these LLMs actually think of your brand. Like, pull it, have questions, see if you appear in those lists. I don't have any actionable ways that you can change the answers for those right now, but I think that's an interesting way. Just like, how does the LLM think of you as a company is a pretty interesting exercise to go and do.

I love it. Okay. All right. Kieran, it was awesome to see you. Like, we've missed you on the show, my friend. We're glad to have you back. And we'll see everybody real soon on another episode of Marketing Instagram. This data is wrong every freaking time.

Have you heard of HubSpot?

HubSpot is a CRM platform where everything is fully integrated.

Whoa. I can see the client's whole history.

Calls, support, tickets, emails.

And here's a task from three days ago. I totally missed HubSpot.

Grow better.