Using GPT Agents For Content Creation (Tools & Predictions) ft Nicholas Holland

Do you want to know how work will be done in the future? The majority of what you're doing today will be done for you by these things called AI agents. That is the topic of this show today. We are going to get into some of the best examples of AI agents, what large companies are releasing over the coming months, what is an AI agent and how it is going to impact marketers and B2B software in general. Ankaran Flanagan, CMO over at Zapier now, Kip is living his best life. He has been in Tokyo, he is now in Singapore. So I have brought in Nicholas Holland. He's back on the show, product leader over at HubSpot, leader of the Marketing Hub business, to talk all about AI. Let's get into today's show. Welcome everyone to another episode of Marketing against the Grain. We're back with Nicholas Holland, product leader over at a HubSpot, the leader for the Marketing Hub, business leader for the Marketing Hub. And we are going to do a show all about the future of work and AI agents. If you really care about how work gets done in the future, how AI is going to be transformative for how all of us do work and what that means for you, you want to stay tuned for this episode. Nicholas, welcome back to two shows and three weeks.

Yeah, it feels like there's a lot going on. Obviously we have a lot to talk about.

This is going to be a fun show for you and I and hopefully people enjoy it because this is the stuff we like to nerd out on. We spent 15 minutes nerd night on other things before we got started here. But what we want to do for the listeners right there, we want to kind of talk about AI agents. We're going to get into some of the big announcements that OpenAI made, some of the other companies have made around AI agents. So there's a lot of like, press and a lot of discussions around AI agents and how they're going to transform work. But I don't really know if we know what AI agents are. Right. I don't even know if we know there's like one meaning for that term. And you and I kind of just started talking about this off mic. Do you want to talk a little bit about how you think about AI agents? Because you're doing a bunch of this work in HubSpot, and then I can get into some of the examples and the way I could explain it to listeners.

Sure. So. So the Wikipedia definition of an agent is basically like an autonomous entity that's kind of performing some sort of task in the background. And boy, there's a lot of interpretation whenever you say that. And the irony is, is that before anyone goes and reads that Wikipedia article, a lot of times they're thinking of like, you know, Ultron the robot in Avengers, right? But then Wikipedia quickly moves to giving an example that a thermostat is an agent, which is kind of a wonk wonk moment when people are talking to me.

Not as sexy as the Avengers example.

So we. A couple of things are happening that I think would help the listeners and kind of helped me. We're at this kind of interesting next S curve of. Of AI. AI effectively is building lots of foundational stuff, whether it's LLMs, these large language models, or these large action models. It's building these kind of technological entities that allow a computer or a technical entity to do stuff. And in the past, if you think about it, we had a computer and we said, if user does this, do this. And that was very much that. The developer would give a whole bunch of parameters, heuristics, we call it like rules, and then the system would do that. So HubSpot today, think about it. Most of HubSpot is you click on this nav item and then it refreshes this screen, right? Pretty straightforward. So what happens in a world where the programmer doesn't give it specific input, output rules? It gives it basically a goal. You know, if you think about the thermostat, I know this is silly, but I'll just start there. The thermostat, the goal is to keep it at 72. Or if you're in my house with my wife at 69, which is like.

Freezing all the time.

But if that's the goal, then the thermostat does that. And it takes all sorts of inputs. It takes the temperature and the humidity and the time of the day and are you on ego? And it works through all that, but it tries to keep it at that temperature. The ability now, because of what's happened with OpenAI leading the way in the LLMs, is that now you can start to move that. Well, be something simple like keep it at 72 degrees. And so we've started to think here at HubSpot, like, oh, man. So really what this is is like a fidelity issue. The problem, keep it at 72 degrees. Pretty discreet. What if the problem was something like make the house comfortable? Now it's a lot more subjective. What does comfortable mean? There has to be a lot of intuition behind that. And then you move on and on. And so at HubSpot, what we've started to do is first try to figure out what are the problems that we're trying to solve and so take it. In my world, I use one very straightforward is like social media. Like I am trying to be on brand, I'm trying to publish some posts, I'm trying to react to people appropriately, I'm trying to listen and monitor and then I'm trying to also tell the powers that be how we're doing. So it's like five discrete kind of problems that are wrapped up in this social media role. And so we've started asking ourselves like, could AI start to do that? And there's all sorts of debates as to what does autonomous mean. And so if any of you listeners are like having this like deep debate in your organization, we, we get wrapped around the axle at HubSpot because people really want this thing to be autonomous. And then on the flip side, we're trying to use some sort of practical knowledge and be like, well, you don't want it to turn into Skynet, so at some point it has to check in with a human and be like, do these social posts look good? And, and so that. There's a lot of interpretation under all of this. But I'll pause there because I think that really with agents now the concept is how do you combine a bunch of fundamental capabilities with the right fidelity of problem and it's goal oriented. So you're trying to say, hey, we're trying to get this done. And when you do that, it begins to look a lot. I know it sounds weird, it begins to look a lot like a human in terms of the role that's performing. And so I'll pause there. That's kind of how we're thinking about it.

Yeah, so you can like give it a goal, give it some, you know, you fine tune it towards that goal and it can complete that goal autonomously. Some examples might be training an AI agent on all of your internal customer support documentation, your policies, your internal training and then have it give it the goal to like get back to every single customer support ticket within a certain. Well, it should be kind of instantaneously because there's no, like, there's no resource constraints with AI agents, which is a scary thing for humans. But I guess like, yeah, this is funny, okay, there is resource constraints, maybe with gpu.

Well, I was gonna say this is funny to take it down the path we were debating on and this gets really nerdy really quick. But like we were taking down the path of like OpenAI they call their custom GPTs agents.

Right, right.

And if you think about it, what they're simply saying is, is that it wasn't programmed by anybody. It was given like a goal in the background. Like I want you to recite everything in poetry and the in the voice of Gordon Ramsay. Like that's like a goal. It's kind of a weird goal, but like that's enough for it to do that. So a question for you Karen, is right now when I go ask Chat GPT to write me a blog post, is that an agent in your mind?

I don't think that's an agent because it hasn't been fine. I think of agents as specialists. Like you actually use that word off mic and that's how I can you.

And I roll back the clock. Ten years from now, is it reasonable to go hire an agency to blog for you?

10 years ago?

Yeah, yeah. So why is it not applicable now?

But I think your example is Chat gbt right in the blog post. But chatgpt is like interface that does broad based set of tasks. Whereas I think of an AI agent as something that has been trained. Let's like one of the ones that I've seen time and time again is like in the medical sector where you have an agent trained on propriety data and then you can give it some sort of autonomous goal because it's been trained on that data and it's specialized to do that task. And so when you have these kind of large language models like ChatGPT, they haven't been like they're pre trained but they're not fine tuned. Right. Like they're pre trained, pre trained on the majority of the data on the Internet, but they're not fine tuned to complete any one specific task. Like they'll complete whatever task they can do based upon that data. And I could be wrong in this, but then like an AI agent to me is like something has been like really fine tuned to be autonomous to the point where it can autonomously complete a task based upon that data.

Two ideas are going to smash into each other. If you think about it, you're basically saying it's not an agent because it's too general.

Right.

But what happens when we hit AGI?

Yeah, I think all kind of artificial.

General intelligence where something can learn on its own. I mean think about it. Yeah, he is a human. I'm not just a product leader, I'm not just a writer, I'm not just a speaker. I am this kind of AGI into, you know, entity that has multiple skill sets. So can you imagine if someone said you're not really a human because you're not very like, you know, until you're in this particular specific role. And I think this is where people start to get kind of wrapped around the axle, which is, I do think personally, in my particular stance, that when you go to chat GBT and you're like, write me a blog, I do think that is an agent. Now I think we're well past that, which kind of blows people's minds. And I think that doesn't capture the, the essence of what we're really trying to talk about today, to your point. But I think if I say, hey, I'm writing for, you know, a particular entity or a particular outlet and this is my audience and this is the topic and I need you to help me write something that is goal oriented, it autonomously does it, it has the ability to, if you think about it, to talk back and forth. Just like you hired somebody 10 years ago and said, well, I don't like this, doesn't fit my tone. The problem is that we've just blasted past blogging as a use case so fast that now we're moving into stuff like, like you were saying, like the medical archival or the lawyer ones. But this was a debate that I had a while back and I'll leave it here, is that right now let's assume that Effectively this chat GPT agent, this next version 5 or whatever, let's assume that now they can instantly go grab data vector embeddings or kind of like stores of data elsewhere. If you had a brilliant genius chat GPT that could go grab all of the, you know, medical data somewhere, where does the agent begin and the data.

Yeah, so that's, it's basically when does it like it continues to not automate, but it continues to remove the human requirement from the loop. Right. Like the, the reason I actually can see why OpenAI describe custom GPTs as agents because you give them a goal and then you give them some sort of upload to fine tune them towards that goal.

Absolutely.

But over time, your point is that that is like a human having to like give you the goal. And that's right, like give you some additional data to fine tune you.

But I would say it's more simply the fidelity is too low. At the end of the day, the fidelity is too low. And so even in HubSpot, for example, we've got these things called assist assists. Like you write an article, you can highlight it and have it rewrite it, it comes up with ideas, for example, on what you should write. It will basically draft outlines. Like it's all of these different individual tasks. But we have started to move down the path of where it would say like do my, at the end of the day put together a content publishing schedule for me. So like now, if you think about it, if it puts together five or six posts that are on brand, that are basically tied to some relevant moment in time or something with your company, it then effectively submits those for approval from a human. The human looks at it, says this is good. It then publishes those. Is that an agent or not?

To us, we think that, yeah, I think the agent thing is confusing. Like one of the ways that I've been thinking through it or thinking about it, it. I've just been going by like how people like OpenAI describe it. So they did their agent launch. I'm sure you've seen that there's like two agents. They clearly describe two agents. One is going to learn how you use software and replicate how you use that software. And I have an example of that in action, not from OpenAI but I want to show you someone tweeted about it today. And then the other one is like this task based engine and we're going to get into a little bit on rabbit R1 because that is like a task based engine which is they want to be able to have you used any software through natural, natural language. And there was a company, I'm sure you know them at depth that were really first to like move in that direction and OpenAI looked like they want to move in that direction. And then the second one is there are, there are a lot of search that we do which is just search that we just want to query to the answer. But there's a lot of search that we do has an action past the search that we have to do ourselves and OpenAI, I think want to both build a web tool because they're reliant on Bing today to do the search and they've said they want to build their own web tool and then complete the action that is like two and two. Giant market. It is.

Watch this. Just playing ping pong with you for a minute. Do you know who Stephen Hawking is?

Yep.

He was a brilliant scientist, had a degenerative health issue that. That effectively locked him into a wheelchair for the rest of his life. What did he have at his disposal? He had thought and he had language. He had no action. It doesn't make him any less of a human or a Powerful entity. It just means that there's a body of things in the human world that he couldn't do. That's kind of how I think about where we are now on the agent standpoint, which is you've got this, you know, these large language models that are effectively this kind of bucket of, of intelligence or knowledge, whatever you want to call it, married with. The only action they can do is language.

Right.

And to your point, you can give it more knowledge and you can kind of tell it to be more specific and you can give it some knowledge instructions. Don't do this, do do this. But if the medium basically leaves the knowledge part or leaves the language part, that whole world is unavailable to them. And so to your point, what I think is interesting is that those are. Stephen Hawking is scientist and the human, but he's a lot different than somebody who can go physically manifest that into the world. And I think that's where the action models are coming in. So I see that all as part of this agent aspect, to your point. And I just go back to. I think of it as a set of like, capabilities. We're going to keep giving them capabilities, we're going to be able to keep giving them higher and higher fidelity of goals and then of course they will be able to put all that together and start doing it. But I just wanted to start there because I think when you are dealing with ChatGPT or a custom GPT that is an agent, it's just a very low fidelity, not a lot of capabilities versus where we're going.

Right. As someone who builds business software, what are your thoughts like on the. Because I've racked my brain around this, like, what are your thoughts on, let's say, OpenAI Adept? These kind of companies do win and the majority of people can use their software through a natural language interface layer and don't really have to go into the product. Like the best example actually is like one of the ones that are very applicable to HubSpot, which is the number one reason a CRM fails in any single company is because the sales reps do not want to go into the app and use it. Right. They want to use the tools and then magically have the data ingested and they want to have the data come back at them, but they never want to really log into the app itself. And HubSpot has been successful in that space because they've made it rep friendly. Like I was part of the team that launch that and like the thing that that team really wanted to do was Build it for the reps and make sure the reps liked using it. But if you could give sales reps a natural language tool and have it, you know, used through email or something else. And so it uses the CRM in a way where they never have to go in and physically use that tool, they would probably do it. But it puts software in a weird place because it cares what the back end is.

Weird place. Yeah, we. I think there's like a bunch of emotions I go through when I think through this. I think the first is, the first time somebody invented HTML and CSS and a little bit of JavaScript, they had no idea where we would end up with today's technologies. So I think that, you know, using the past to inform the future, we have no idea where we're going to end up. I think the second thing is, is that, as you know, Karen, I like a lot of philosophy. And so one of my philosophical phrases is around anxiety. And I like one of the stoic phrases, which is like, why suffer twice? You know, at the end of the day, if you're going to suffer, just suffer at the moment that it happens, why basically worry about it? So I basically continue to encourage my customers, the people in my sphere, our employees, like, why suffer twice? And what I mean by that is, is we don't know where it's going to end up. Have a good time along the way, and just continue to try to basically stay in, you know, stay focused on solving for the customer, do good work. What I think is fascinating about all of this is that I do think this is a move the cheese moment. I don't know if anybody's read that book. It's kind of an American based book, but like, there's a concept that effectively every day this rat goes in a maze, sees where its cheese is, and one day that cheese is moved. And this rat cannot, cannot fathom that the cheese has been moved. And so it basically dies in that spot waiting for the cheese. And this is a moment where the cheese has been moved. And if you don't move, you will die. I do believe that right now, how much you move, how far you move, that, that's the game here that a lot of people are playing. There'll be startups that are like so far ahead of everybody that they'll fail miserably because the world's not ready. I mean, you got to think about it right now. I laugh. We were just at our. I was up in Boston this last week and I'm asking some of the world's best marketers. You know, you gotta remember HubSpot is incredible. Like not just the software, but like the marketers at HubSpot are the best. And I'm like, what AI tools are you using? And we have a corporate version of ChatGPT and three of the six goes, I love our corporate version of Chat GPT. Think about that. They're still just using ChatGPT as their main thing. Some had moved on to using a little bit of search but like we're very early and so I think that there's a lot of innovation happening but we also have to make sure we meet people where they are. They have to be ready for. You've got someone like my wife that is like an absolute like rejecter of technology. You've got people who are basically feeling a little bit, you know, skeptical of it. You know, talk to somebody about agents and then make the mistake of equating that to Alexa and you will quickly see all of the emotion and excitement drain out of them because you can barely get Alexa to play the right song for you half the time. So I just bring all that up because I think that we are at the precipice of a big change. But to your point, I'm not too worried about it because right now I don't think anybody knows what's up. I think we're watching all sorts of really cool things happen but we don't know exactly what's going to catch the zeitgeist.

Yeah, I think it is interesting on the user level. I think a lot of AI apps have really struggled to get any type of usage and retention. I think ChatGPT has kind of swept the market in terms of being like for the, for the person who's not really in a day to day. ChatGPT is AI. Right. You don't really need to go beyond that. Most of the money being made is actually in the back end. The fine tuning of models, the LLMs, the pre training, all of that kind of stuff is where a lot of the investment and money is being made. But I think the big thing is what you kind of alluded to there is we don't know what the user interface is when you don't have to click on apps. Right. That is the big thing. I think the one of the launches that came out recently, I know Kip obviously bought one of these because he buys all the new gadgets. We were WhatsApp and I was like, I'm going to buy one. But then I didn't end up buying One because I'm always like can never. I'm just so indecisive with these things. But the rabbit R1 I actually, I'm going to get one because I read about it again today. So the rabbit R1 and for people who aren't aware it's this device and it's like a little orange box. I think they've cleverly made it like a Tamagotchi. Do you remember the Tamagotchis where the little pets hearing. It's like a, it's like a Tamagotchi S because it has a little rabbit as the, as the person, the kind of personality. But basically really what it is is it's a large action model. And the large action model means in basic terms it learns how people use websites, apps and mimics those actions for a voice prompt. So over time, the more you use your Rabbit, the more it can actually complete these tasks on your behalf. And interesting about that is there is no apps to click. It is all done through voice and you kind of pull that forward and you're. And you know, this is the open AI. Are they going to work with Johnny Ice and a new phone? What does a large language model operating system look like if you actually have that running your laptop versus something that is like app based? That's the, like that. That to me is the biggest unknown for all of us working in the click, click, click, moot, drag things around B2B software is. I don't know if any of that makes the cut in the next kind of user interface.

I think it will. I mean I'm an Android user, much to the chagrin of my peers. And I tell them after many years. I originally stuck with Android because I think Apple is not my favorite company just on some of their practices. But then I have subsequently stuck with Android honestly because of voice.

Right.

It's unassailable. Like the. And I talk to my phone all the time. I every text message, every email, every, you know, call my wife, all of that stuff is, is talk, talk, talk. And they've done a great job of that. On the flip side, they just brought Gemini to the phone for the first time and where some of this gets funny and you know it's going to get fixed. But like I will ask, you know, I can say H.G. you know, it's Hey G O. I'm not going to say it now my phone's here. But I can say that. And it's extremely responsive. It answers stuff. Gemini is awesome. It's like having kind of Like a chat GPT with perplexity, like, you know, in the, in the phone. But the funny part is to, to your point about the action model is I will ask it to call my wife and it still can't do that. So it literally got rid of Google, it got rid of Google Assistant, and it can't even call my wife. And so we've got to figure this stuff out. I think to your point, some things that have been interesting on the agent side, I'll give you some, some real brain busters that I've been thinking about. First, we've got a user adoption. And it's got to be this, it's got to be an iterative loop you just talked about. You have to train the rabbit to do stuff so effectively. These models have to work for the person. And these people are now trying to pull off personalized models, which we've never pulled that off in almost anything we do. Nothing is really that personalized across our whole, their whole life right now. So they're going for a very, very big challenge. That's number one. Number two, what I think is fascinating is you're asking humans to be part of this experiment. There will always be early adopters, but do you envision your grandmother doing a rabbit and saying, order me an Uber? Maybe not. So now we've got just timescale will matter over the next decades, you know, so we've got that. The third thing that I think is fascinating is there's just so much to be done that I don't know if it's going to be a very general model. So like a OpenAI, like a rabbit or. What I actually am starting to think will happen is I think there's two things. Efficiency and effectiveness are just the two kind of big levers that I think about with AI. So I think to myself, what can we do to make people more efficient? What can we do to make them more effective? And on the agent side, I think that's why we've seen such a blossoming of the copilot. Because if you think about an adoption model, I need to work with the human to get better. I basically now to work with the human, I have to present my way myself in a way that's not threatening to them. That is, I am a collaborative entity. I'm a co pilot with you. And we both now have a common goal. I'm this co pilot entity. I don't get paid, I don't have a salary. So my common goal is to help you human. And your goal is to basically do work for the company to get paid. So all of the incentives are aligned at this moment. And now the co pilot's job, and I don't think a lot of companies get this. The co pilot's job is not to replace the human, it's to basically make the human more valuable. So in this next phase, and I'm not saying at some point there, there might not be humans that are replaced. I mean, you got to look at the Luddites in the past with that, with the, the printing press, and you don't have stenographers anymore. So there will be some roles that catch the short end of the stick. But in general, what I think is awesome is just take a. Today's marketer, take their computer away and be like, do your job. They can't. They would be completely worthless because all of the technology brought to bear today gives them a certain level of capability and productivity. That's where I think this stuff is about to go with the co pilots and the agents, for example. And I'm pretty excited because I'm about to give a talk here in Nashville and I was talking about some of this stuff and the guy goes, should I be scared? And I was like, what do you mean? He goes, I do social media myself. Like, am I about to be out of a job? And I was like, if you suck at social media, yeah, you might be.

Out of a job.

On the flip side, at the end of the day, imagine if you had basically a college educated kind of intern that came to work for you. Could you put them to work in such a way that it actually helps you with your goal, your boss, your company? Of course you could help. They could do better research. They might have better time for you to basically do the menial tasks that unlock your creativity or strategic thinking. And so that's where I think next is coming with the agents, I think they'll do a lot of menial tasks. Rewriting a blog post, for example, is kind of a menial task nowadays. So I think they'll do menial tasks. And I think what's exciting about most of us is that that will allow us now to move higher up the value curve is what I really believe is going to happen. Last crazy concept is what happens when you buy software to your point today, you negotiate as hard as you can. Oh, I want HubSpot to give me a discount and I want it to be cheaper, cheaper, cheaper. And then you basically bang on all your employees. I bought this expensive software, Kieran, you better use this Those are your two motions. What happens for the first time whenever you buy HubSpot and you get extra capacity for the very first time you buy HubSpot and you have agents that you can now hire that do something like your. It's like a whole content marketing department in a box. It's like a whole social media department in a box. That is really going to change, I think how people buy software and think about stuff and I don't know where that's going to shake out, but I think that's going to be the interesting next step. Copilots and agents. That's how I think it's going to touch business.

Yeah, I think everyone gets a copilot and will have to learn how to outsource parts. They have to divide up task. Like all jobs are just array of tasks and you'll have to like figure out the tasks that you can give to the co pilot so you can actually get better at the other ones. The one that you touched on I actually have thought a lot about which is just the purchase of software because all marketing is really the entire. The entirety of B2B marketing is just trying to nudge you towards buying a piece of software. We like to think of it as more than that, but that's really what it is, right?

That's right.

So if you actually think about that, I was actually stepped back yesterday and I was thinking about just that process. No, no one. I don't know if you've ever met someone whose job it is to buy software enjoys buying software in the first place, right. It's not. You don't wake up in the morning and go oh, I'd love to get really great at buying software. It's just part of your job. And I think that it's is an unnecessary task when you actually have an agent because what an agent can actually do is they can take a goal. To your point, like here are the exact requirements in terms of what my company needs. It can scour the web and then build you a scorecard, right. It can actually summarize your website. It can sign up for the company's websites and instead of you getting drip, drip, drip, drip, drip, drip website, you can actually have that agent receive that website through an inbox and actually summarize those things for you. It can take a demo of the product, right. Because you can just give it some guidelines. And if you look at where we're going with OpenAI in the depths you can have things that sign into software and complete tasks and it can score Character tasks to say whether it was able to do those things or not. Similar within the free trial. And what I was thinking through was like, okay, well in the future, if a company can have an agent just build a scorecard of software and like where. Where that software landed in terms of their requirements and the agent did the majority of that. And then human makes a final decision based on those scorecards. Like, the other thing the agent can do is aggregate user reviews and based upon the exact things you care about, summarize, like where the user reviews landed. Positive, negative, whatever it may be. Well, they were. It. Actually you don't. You're not marketing to humans anymore because now your entire B2B sale has been reduced just to a scorecard.

But actually that doesn't bother me. This is like when people tell me they're like, oh, I'm so scared this is gonna just produce an enormous amount of terrible content. And I'm like, that's a solved problem. That's a solved.

I don't think it's gonna. I don't think. I'm not scared of that. I'm.

No, but I'm saying like what you just described, you just described like G2 crowd. It's a solved problem. Like I go there and I look at the reviews and I just. And so all that's going to happen is like energy is going to ebb and flow to different places. There's going to be, you know, either G2 crowd is going to lock it down and not let people scrape it because they want the humans to come there or reverse. What's going to happen is everybody's going to get really good at their Capterra and G2 crowd and their stuff like that. And so.

But I'm saying that you, Nicholas, have to buy a. Let's get like a Calvin we like tool, right in your company. You and you just have an agent, give it a set of requirements and it comes back to you individually. It's not like on a. It's not on the aggregate side. The thing that I'm actually more thinking through is not that it creates a bunch of content, but all of these marketing tricks that we use to put it towards the purchase don't matter anymore because.

No, no, no, no, no.

That's.

That's what I'm saying is I think that's a solved problem.

Okay, how is it at the end.

Of the day, here's what I'm saying is like, I think that you're right. This will be a new motion. So make no mistake. But what I'm saying is, is wherever that perplexity goes out or the Gemini goes out, there's like, it's like what marketers do. Like we will have to adjust. It will get its data from someplace and then we will begin to figure out how to get our data into that place. And the battle will no longer be for the first page of the search engine. It might be for some of these destination type sites like Capterra or G2 crowd like that's all they do nowadays.

But even your emails, right, like today, a lot of demand gen when you capture the lead, you create like really great copy and you try to segment your users into different part like they are in these different parts of the journey and I map the copy and the value. What I'm saying is that that is not going to be that impactful when it's not a human receiving that, it's an agent.

Just that's, that's a new. And if I don't know that that's happened yet. But yeah, you're right. If, if you get to a point where something can connect the dots and do an instantiated search followed by it knows how to sign up, followed by it knows how to then summarize. Yeah, then we've got a new game that we've got to figure out that is happening.

Let me just show you this because this was pretty, pretty timely because it actually is related to this. Someone was posting about the NHT that they created. So I caught this on Twitter earlier on. I thought it was really interesting because it's like how quickly we can train something to complete tasks on our behalf, which is someone like an agent where, you know, this person was looking for an apartment on Zillow and Gemini was able to code and replicate that task and describe everything they did step by step. And so the code opens a Chrome browser, navigates to Zillow, enters the area they want to look in for rent, enters a price range bedrooms, and then applies to condos that are within their wheelhouse. The thing that they want to do now this is a very simplistic thing. This to me will become how quickly it is to outsource what we've talked about. Outsource a bunch of tasks. Because where I see this working is to your point, like you don't want to, you don't want to put a ton of friction on the user. So this just sits in the background. Which is why I actually think OpenAI need a browser. Like if you actually look at where they're going, they actually need a browser to actually be able to see users do these things across the web.

Totally.

But this is a really good example of like Gemini doing this. Like Gemini was able to kind of just replicate this task for that person.

But this is why I said it's a solved problem. And maybe I'm, you know, too nostalgic for the past, but like, you had search engine optimization come out because of this big trend and it kicked off a whole new industry. To me, when I say it's a solved problem, it's like somebody will switch out to do something like Gemini optimization expert. There will be something. There will be a degree out there where marketers was. That was that Jurassic park phrase, like life will find a way. Marketers will find a way.

At the end of the day, we will find a way to saturate whatever you. That's right.

You know, make no mistake. And so I do think that, you know, marketers have to be aware of this changing the speed of which it's changing the degree. We have to be very flexible. The cheese will be moved. This like, as to when. I don't know if it's going to be as fast as some people think. So take like what you said right there. There's no net new value. It's just a neat science project. There's no net new value there because it's using all the structured data that you could have, you know, theoretically just gone to Zillow and searched. So what it did is that's almost like a poor person's API, what you just described. Platform.

Exactly.

And I think that's cool. So I think ultimately the next phase for me, I think will be to your point, how do you take some of these new capabilities and apply them towards kind of what we're kind of calling like a specialist type role. I think there'll be a specialist role, will be like a role that you might hire in a company. There's like different fidelities. I think there's a co pilot model which will be so like the specialist would be doing work in a particular genre. The co pilot will be probably helping you do your work. So now where this gets really wild, I don't know if you want to get. This is like, we're not at the point yet and this is the next paradigm I expect to happen is like what happens when a co pilot agent can go hire a specialist. So now what happens when agents can call on other agents? That's what I think is going to come next.

Oh, 100. 100. I like there, there was. I can't. We covered auto. Was it auto GPT that was kind of autonomously. This is way. It's. Well, I was gonna say way back when. It's literally like a year ago. That's how far we've moved or how quickly we've moved. But it was, it can autonomously call on other GBTs to complete tasks. Well, which makes.

We're thinking of that now because this is why I go back to the agent definition. If you don't think that like the blog agent is an agent, this might be tough to get to. But if you begin to think of like an agent is you just give it a goal and it does an autonomous thing, well then you begin to like open your mind a bit to the point where you're like, I need an image that matches the underlying source material. I'm going to generate an image. Pretty straightforward. Think about it like I'm a. I'm on Dolly. I'm like, I'm going to post some text. Can you pick the right image for it? That is effectively you post text, it does a summarization model, then creates a prompt and then post it to Dall E. If you're spending much time in AI, that's like the simplest of all things that are happening right now. But in my mind you could roll back the clock a few years and that used to be a photographer. That used to be basically a stock image selector. Used to be what you would call a design layout specialist in the, in the real world. Used to be that. So now you have. When you're writing a blog post, this is what gets interesting. If I'm like, write me a blog post. If it basically does some research, that's a research agent. It then basically writes it. That's a copy agent. It then applies a brand voice against it. That's an editorial agent. And then it goes and gets an image. That's an image agent. That's where I think the auto GPT or the crew AIs are going to go. Because in my mind those are four agents that were hired by an art director or a content director. And what's nice is that now you can begin to break up the image issue. So let's say the image. And this is where like I think HubSpot. I get pretty excited about HubSpot right now. Dolly is awesome, but imagine from Google is also awesome. And you get to mid journey is also awesome. And so you're going to find flavors of even these low level items that are going to be better at stuff. So now what happens? And this is where now it starts to become super meta. I might hire engineers from mit, I might hire marketers from Tulane, I might go and get this role from another college. So I'm grabbing these different specialists. Even at HubSpot we're hiring into these different places. So now if we at HubSpot can help a customer through a kind of a co pilot or a specialist model, say, hey, let us generate some content for you, it doesn't have to be drivel. And if anything, I have a belief that over the next 18 months, because of the breadth and the depth, like the overall product will be better than what people put out today. It'll be more research, it'll have more applicable, stylish imagery and things like that. It'll be better on brand anyways. So my point is when you have an agent that can basically help you now instantiate those four or five agents and let's say part of that sucks now we at HubSpot can go fix the image agent, we can go fix the brand voice agent, we can go fix that. And so now you've got, you're pushing on multiple skill sets at the same time. That didn't get rolled up in this. That's where I think things start to get interesting. Two other crazy concepts and this is for the what if as a counterpoint, content actually gets better. Let that sink in for a minute.

So I want to answer both those I want to the HubSpot one, there's real actual network effects in that because the More Someone uses HubSpot the better their agents will be trained. Like the brand voice one is a really good example where if I host all of my content in HubSpot, if I use HubSpot's blogging tool, if I use HubSpot's email tool, so I have more and more of my brand journey in there, then HubSpot can learn over time what my voice like you're, I'm talking to the person who's building it. But like I just, it just struck me there's actual real network effects which is like the more I use the product, the better my agents are trained.

That's right. And those agents now work for you. So think about when you're in a sales email, it can actually generate images that are still on brand that still basically do that. It knows your products. And so like now when you begin to think about these people working in the company, you get the network effects. You have basically this back and forth. The humans are kind of guiding it. So it's like having these virtual people work for you. And after a while, they're pretty good, Right?

I. I think the content. And I do agree, I think there will be more content, right? So if you ask, there will be more content. Let's say the distribution today is 80 of it is garbage. 10 is, like, pretty good, and 10 is excellent. I suspect that stays mostly the same. But there's just more.

Yes, I agree with that.

More of the 20. 20%. There's more of the 80%.

Yes.

But the people who are really good at content coming, like, what we're really describing here is they will be able to offload their repetitive tasks to actually concentrate on the more creative side of things. So they should actually do even better content. Right? They actually should be able to elevate their work. But the people who will. Content is just one of those things that you are not good at. Content. An AI tool doesn't help.

No, no. It's even one step further. This is why I think things are good. Like, I think about breadth and depth. So check this out. When I talk to an SMB business today, if I were to say, hey, real quick, do you have a podcast specialist, a videographer, a video editor? Do you have a sound guy? Do you have someone who can put that all together, an SEO person? They'd be like, no pound sand. Of course, I don't have all that. What I have is Sarah, who basically is an English lit major.