Eric Housh

Hello and welcome back to AI in Action. I’m Eric Housh, joined as always by the one and only Zack Terry. Zack, welcome.

Zack Terry

Hey Eric, how you doing, man?

Eric Housh

Man, it’s been a great week, it’s been a great week. And this week on the show, we’re going to talk about HubSpot. The question we often get from business leaders is, man, HubSpot’s talking a lot about AI, it seems like all the messaging coming from HubSpot these days is AI-centric. And how should I actually be using HubSpot AI in my business? So today we’re really going to cut to unpack that. We’re going to talk about HubSpot’s AI agents, what they have out there right now that you can turn on, what’s coming, and most importantly, how do you take those agents, connect them to your business processes, to actually save time and drive results. But as we like to do on the show, we like to take a snapshot of the world of AI this week. Let’s hit a few headlines. First of all, ChatGPT launches a safety mode. What’s up with that?

Zack Terry

They did. So this is supposed to be a dedicated experience for users that are under 18, and it’s supposed to do things like age prediction and ID checks. So I guess if it ascertains that you’re probably underage, it now has some safety mechanisms involved there. So it will block things like sexual or graphic content. It’ll try to limit self-harm discussions, which we know has been a big issue with ChatGPT. There’s a lot of news out there around just some really inappropriate things and some things that have led to some really bad outcomes. So definitely a response by them to start putting some of these safeguards in place, escalation pathways for parents or law enforcement in certain acute cases. So I think it’s really a new benchmark for safety, and I think it comes as a result of a lot of regulatory pressure. So overall, I think it’s a good thing. We’ll have to keep our eyes out and just kind of see how it works in practice, but I’m a fan of this idea, especially with so many of these misuses that have been going on. So anything we can do to kind of help limit that while also not affecting the freedom of the platform. There’s always kind of a fine line there, but I do think for underage users, this is a big thing.

Eric Housh

Yeah, and these tools becoming so increasingly popular with underage users, I think it was kind of an inevitable move. So good to see progress on that front, for sure. Next up, we’re talking Microsoft, they’re auto-installing Copilot on Windows devices. Why?

Zack Terry

Well, if you’re not a fan of Copilot and you don’t want it too bad, you’re going to have it now. Yeah, yeah, it’s like…

Eric Housh

Like that U2 album, remember when Apple…

Zack Terry

…Apple installed it on your iPod. Now you’ve got it on all your computers

Eric Housh

you

Zack Terry

and devices. So, beginning in October, Copilot will be auto-installed on all Windows PCs with Office apps, supposed to finish by mid-November. So it’s going to be sort of an October-through-November rollout, trying to hit all the devices. So there’s no opt-out for personal users, outside of just installing… uninstalling the software. So I’m assuming that there is an opt-out if you’re an enterprise user. But right now, if you’re a personal user, Copilot is there, right? You’ve got it. Those YouTube songs are downloaded and ready to play. So it’s interesting, it’s not really giving users a lot of choice, and I think they’re trying to force people to adopt this tool. And I’m not sure everybody wants it, but you’re going to get it.

Eric Housh

Yeah, forcing people to download, forcing people to install. Yeah, it seems like a great idea. Who greenlit that? People love that.

Zack Terry

Yeah, it’s really history repeating, really. People were a fan of that with Apple. So yeah, why not repeat history?

Eric Housh

Gosh. Speaking of, all right, so let’s talk China. China has asked technology firms to stop buying Nvidia chips.

Zack Terry

Crazy, right? The Cyberspace Administration of China has blocked purchases of Nvidia chips. So this is impacting companies like ByteDance or Alibaba, they are having to cancel pending orders that they had with Nvidia for these chips. It’s completely undercutting Nvidia’s business in China, and China is really trying to push for domestic alternatives. They, you know, they don’t want to have to outsource this or rely on Nvidia or rely on the United States. So I’m sure that it is a big, big, big impact to Nvidia’s overall business, considering that they are the market leader, and I’m sure China’s entire tech sector was heavily relying on these. And so, yeah, it’s a big story. I think, you know, I haven’t looked at the market in terms of the impact that it had, but I’d be surprised if it didn’t have a negative impact on Nvidia.

Eric Housh

This feels like a threat more than something serious. I mean, for the promise that they’re going to develop their own chip technology that can catch Nvidia, it just, I don’t know, it seems like a strategic chess move that they’ll probably backpedal on.

Zack Terry

Yeah, I don’t know if they have any truly competitive folks in the marketplace, at least that can stack up to what Nvidia is providing. But I mean, on the other hand, maybe that creates some innovation and they’re able to create something that’s even better, which I don’t know if we would want that from a geopolitical perspective, but from a technological perspective, that could be interesting to watch.

Eric Housh

Man, I heard today, speaking of Nvidia, that Nvidia is investing in Intel. And when you think about the history of these two companies, in 2022, we were talking three years ago, Intel is dwarfing Nvidia; in three short years, thanks to the GPU being so good at AI, Nvidia now is just towering past Intel. I think I worked at Intel back in the day, so that whole turn of events is just fascinating for me, and their headquarters are literally located about two miles apart in Santa Clara. Just really, really…

Zack Terry

I think, if you get… Intel was always providing chips for all of Apple’s devices, and then, years back, Apple decided they’re going to make their own silicon. And so now that’s a huge amount of business that they’ve lost there. Now they’re competing with Nvidia, and Nvidia wants to invest in them. So I wonder if this is a play to infuse some cash into the business, but it is definitely interesting.

Eric Housh

Yeah, for sure. All right, let’s talk about Google. Google launched a payment protocol,

Zack Terry

Google.

Eric Housh

AP2, for AI agents. What

Zack Terry

They did.

Eric Housh

is a payment protocol?

Zack Terry

So AP2 is what they’re calling the Agent Payments Protocol. And they developed

Eric Housh

Mm-hmm.

Zack Terry

this with more than 60 partners, including Mastercard, PayPal, Amex, and Coinbase. And the whole idea behind it is to allow AI agents to make secure purchases with mandates and audit trails. So theoretically, what it is designed to do is provide consent, fraud protection, and trust in “agentic commerce.”

That’s a very new sort of concept, right? So people are saying it’s like a foundational step toward AI-driven transactions. I think part of this came about with, is it Comet, the browser that Perplexity put out, where there were some issues where you could get prompt-injected and that could go out and complete a purchase for you automatically, or something crazy like that. So I think this is designed to kind of be an in-between, while still allowing these payments to be processed with agents. I personally would not trust this right now. I mean, this is brand new, and that’s a ton of trust that you’re putting — you’re giving your wallet, essentially you’re giving your bank account over to AI, and I’m not ready for that step. And I imagine there’s a lot of people that are not ready for that yet either. But if we’re ever going to be able to have some kind of system where that’s realistic, this is essentially the first step towards making that a trusted and protected way to interact and perform commerce online using agents.

But I think it is in its very early stages, and I’d be surprised if it gains a lot of traction, at least immediately.

Eric Housh

Yeah, it seems like we’re gonna kind of tiptoe into those waters. It will be cool though, a few months from now, when we can get AI to fill out our Instacart order or our DoorDash order, for sure.

Zack Terry

Yeah, and I still think even if it can — like, you could still get AI to kind of do that for you, and then you’re the one pressing the button. I still

Eric Housh

Yeah.

Zack Terry

want that control, right? Like, I want to be the one making the decision to actually drain that money out of my bank account. So that’s where I’m a little bit hesitant.

Eric Housh

Keep the human in the loop for now.

Zack Terry

Yeah, exactly.

Eric Housh

DeepMind’s Gemini 2.5, killing it at the ICPC World Finals.

Zack Terry

They earned a gold-medal performance, which — if you didn’t know that there are these essentially arenas for these tools and these benchmarks that are going on. Yeah, I…

Eric Housh

I didn’t know this until today, honestly.

Zack Terry

You know, I think about it kind of like your favorite beer or whiskey wins an international spirits competition. It’s like, well, they paid to be there, and,

Eric Housh

Yeah.

Zack Terry

you know, it’s not really like truly the best, that’s kind of how I think about it. But it is interesting, because I mean, Gemini is a great model. It’s been a great model. From this sort of contest, it solved this fluid network optimization problem that no human team had cracked. And so it solved an interesting, novel problem, which I guess is why it won first place, it got the gold medal. It fell short on two of the twelve challenges, however. So it does show that there are still limits, there are still some things that these models struggle to do. But it’s also showing the rapid progress in reasoning and programming with these systems. And I think it’s cool that Gemini is sort of leading in that way now, because for a long time I think Gemini was trailing behind some of the other models, but it’s super capable now, it’s a great model.

Eric Housh

Awesome. Well, that takes us through our headlines. I mean, obviously we put these headlines together this morning, there’s probably, while we’ve been recording this, more breaking news that we can talk about, but thanks for unpacking those for us. Let’s talk HubSpot now. And specifically, you’ve said before that HubSpot is really leaning into this concept of making AI easy. What do you mean by

Zack Terry

HubSpot? And actually, if you kind of zoom out of AI and you just look at HubSpot as a CRM provider and a marketing tool, they’ve kind of always led on the side of, “we make this stuff easy.” It’s always been something that’s relatively easy to configure and to access, and has a lot of power-user capability now that maybe wasn’t there a decade ago. So they’ve got both, they’ve got a lot of power and a lot of ease of use, and they’re really taking that ease-of-use thing as far as they can go with AI. And so it is very simple to set up AI in HubSpot, and every edition has it, including the free edition. So if you have a free version of HubSpot and you want to work with the Breeze assistant, you have access to that. And then if you wanted to do some more targeted actions, they have an entire suite of specific agents. And what’s interesting about those agents, it’s not like you have to go in and configure it with all of your sort of business logic and everything that you need for that agent to do the job. HubSpot has basically taken the approach of, let me identify specific problems that many businesses — not every business, but many businesses — run into, and they need a solution to solve for it, and let’s create an agent that targets exactly those things. And so the impact of that is, if you happen to align with some of those business problems, you have a very easy-to-use, easy-to-access agent for a number of those business problems. Now, the downside to that is you may not have the maximum flexibility to build a completely custom agent, at least not yet. HubSpot may come out with something like that, but right now it’s much more targeted, it solves specific problems, it’s very sort of out-of-the-box. You know, if you want a specific agent, you go to this marketplace and you enable it and you turn it on, and that’s kind of it. It just kind of becomes available in whatever hub you’re working in, and it does the thing. You can modify the prompt a little bit, so if you wanted to change a little bit about how it behaves, you can do that. So there is some flexibility, but overall, I mean, HubSpot’s philosophy here really seems to be AI should be seamlessly layered into the tools you’re already using, as opposed to bolted on in pieces later. And so I think this approach is awesome, I think it’s great. They’re making something that is very complicated, at least from the outset. I mean, if you’re a business leader and you’re looking at AI and you haven’t started to roll that out across the business, and you’re trying to think about how can I get started — well, if you’re a HubSpot customer, it’s never been easier to get started with some of these AI tools using the data that you already have. And if you’re not a HubSpot customer and you’re thinking about it, that could be a compelling reason to test it out, to get that free version, or maybe get on a plan if you’re ready to kind of jump in, because you get access to it right out of the box.

Eric Housh

Yeah, I think it’s worth reiterating that point, that this stuff is already baked in. So if you own HubSpot, you can turn these things on and start working with them. Well, let’s

Zack Terry

Yeah, absolutely.

Eric Housh

double-click there. Let’s actually talk about some of those agents that they can turn on. You know, we have these conversations daily, they’re saying, hey, what’s available now, what’s coming, how do I need to be thinking about these Breeze agents? So let’s break that down.

Let’s talk about the customer agent first, one?

Zack Terry

Customer agent, I think, is one of the easier AI concepts to grasp, because we’ve been talking about this for a long time in terms of the use case. This is like the obvious use case, it’s helping customers with questions that they might have. So leveraging your unstructured data, leveraging your CRM data, retrieving those answers autonomously, 24/7, without having to have a human involved the entire time, being able to escalate complex issues or sensitive issues to human reps. So maybe there are certain things that you don’t want an agent to handle, like if someone’s talking about a discount, and you don’t want an agent to be able to handle that, that goes straight to a human being. But the idea is it’s just resolving routine inquiries with fast and accurate responses grounded on your data. And this is very similar to what Salesforce has done with their service agent. It’s essentially the same use case, right, but I think if you’re looking at those two solutions, the Salesforce solution can be much more — you can use that service agent and customize it to kind of do anything that you want from a customer-facing perspective. This is gonna be much more specific, but it’s also easier to turn on, you basically just turn it on, and if you want to put it on, let’s say, a HubSpot landing page or on your website, you just add the code snippet to the website and it’s ready to go. And really, this is what HubSpot is telling people to start with, after you’ve worked with the assistant. So HubSpot really wants you to start with the assistant, just to get familiar with AI, and part of that is because everyone has access to it. And then, if you’re on a plan that supports agents, then start with that customer agent, because it’s the easiest one to understand, it works really well. And once you’ve sort of gone beyond that, or you’ve stood that agent up and you’re ready to look into other ones, then you go to the marketplace and you start looking for some of these other agents, which we’ll talk about here in a moment.

Eric Housh

And the customer agent, they can get that one right now, right? But the rest of these we’re gonna talk

Zack Terry

Yes.

Eric Housh

about, as of today, September 19th, 2025, these are still in beta.

Zack Terry

Correct, yes, the customer agent is available now. If you go — and we’ll link this in the comments and the show notes — but if you go to the landing page that HubSpot has for all their AI agents, you can actually go in and see, they call it the marketplace, but it’s not like you’re going in and you’re paying for these different agents, you’re just figuring out the agents that you wanna use. The cost of it is both — you have to have a certain plan, a certain tier of HubSpot, and then you also pay for them on a consumption basis. So you get a certain number of credits per month that you can allocate towards these agents, and you can do things like set some thresholds, or set a budget, you know, things like that. But some of the ones that are in beta that we expect to come soon include a prospecting agent. So, similar to the Salesforce SDR agent, this is acting like a digital BDR. That’s what HubSpot describes it as, and it’s able to do things like conduct research on your prospects. So it’s not going out and actually finding prospects for you, but once you have that data, it can go and enrich that information, it can go and research information about those prospects, it can write personalized outreach emails, leveraging your brand voice, which you can define across your entire HubSpot organization. And then your agents can essentially reference that, so that anything that they generate actually aligns with how your brand voice would speak to those things. And so it can also help with closing, and ultimately the value proposition is to spend more time closing and less time prospecting.

Eric Housh

Saw this one in the marketplace, and if you pair this with HubSpot — now at Inbound they announced free data enrichment. So essentially, for every record in your database, you can now augment it with things that are known from different data sources that may not have previously been in your database. So things like industry, and revenue band, and things like that. So when you think about the concept of, if I’ve got my existing customers in here and I know what my ICP is with this set of data, then we can go out and leverage the prospecting agent to go find companies that look like those. So I got really excited about that.

Zack Terry

Yeah, I think it’s a great idea, and it’s a great feature, and some other ones that are coming… there’s one called a closing agent, which is designed to give sales teams insights into buyer engagement. Imagine you have buyers that are asking questions, they can get answers from approved content without having to interact with a sales rep or with a person. It can also alert sellers inside of HubSpot when it thinks it’s time to step in with a human touch. So it kind of analyzes the sentiment and the information about that process, to try to get you in there at the right time. Then we’ve got a knowledge base agent, which is designed to surface gaps in your help content, and so it can help you draft new articles based on actual customer tickets inside of Service Hub. It can keep support documentation up to date without having to manually go through and constantly revise these articles, it can help you do that.

There’s a personalization agent which can identify segments that respond best to tailored content. And so you can kind of see where this is going. I can generate all of the content with a specific agent, and then I can have a personalization agent try to go out and find segments that are going to respond well to that particular messaging. And so not only are we helping generate the content, but then we’re also mining our database for insights, to figure out who that content will resonate best with, which I think is super interesting. So I’m excited for that one to come out of beta and to be available. And then one of the last ones we’ll talk about — and again, this is like, what, six of the 33 or so that they’ve announced that you can kind of browse in the marketplace — but one that they talked about at Inbound is the data agent. And so if you’re familiar with HubSpot and Operations Hub, Operations Hub is essentially now Data Hub. So it’s the evolution of Operations Hub to more of a comprehensive tool. And the Data Agent is designed to answer business questions around your CRM data, your unstructured conversations, documents that you might have in the system. And it can even do things like web research. So it’s combining a lot of different components there. And then it can provide decision-making insights from all of that data to help you make those decisions. And so it’s really treating the AI kind of like a business analyst, and not just a data cleanser, so to speak. So instead of just having this AI go in and try to find things that are wrong and fix it, it’s actually looking at the data sort of as a business analyst, and drawing insights from that, so that it can report that back to you, and then you can make decisions on it.

Eric Housh

Super cool stuff. Yeah, circling back, if you own HubSpot today, go play with the customer agent. These others are gonna be rolling out of beta soon enough. And again, a lot more in the marketplace you can go check out. I think a lot of business leaders get stuck, Zack — they ask, how do I use AI? But really, where in the process should I plug it in? How should they be thinking about that whole question there?

Zack Terry

It is a tough question. And I think that’s one of the hardest questions to answer when it comes to AI. The reason behind that, I think, is there’s many reasons, but one of the primary ones is we got this new technology and had the ability to apply it to business, but we didn’t necessarily identify problems beforehand, right? And that’s kind of the opposite of how we would recommend doing any kind of technology implementation or discovery, right? And so the first thing I would say is: think about any sort of technology implementation, or when you’re searching for tools, why are you searching for those tools? And what’s your reasoning behind that, right? That’s really starting off with a problem. And so, having some process discovery around, what are some problems that we have? Where are our teams spending hours on repeatable or low-value tasks? Where do handoffs between different departments or different teams break down? Why did they break down? And then a question that I really like asking, and that I encourage everyone to ask themselves and ask their team, is: what do you hate doing? What about your job and your day-to-day do you just dread, right? Like, if you can — everyone has them, right? We all have those things. The challenge is actually sitting down and identifying it. So for some of us, I think we’ll be like, “I have a thing,” immediately, it immediately comes to mind. The rest of us, I think we have to kind of think about it throughout our day-to-day, and then, “okay, now that I’m doing this, this is that thing,” right? And so, just thinking about those things — I use the word “hate” strongly, because I think if you want to find something that’s going to be effective, find the worst things, find the things that you just absolutely dread doing, and focus on those. And then you get the added benefit of adoption, beyond just solving a problem, because people don’t want to do those things. And if you give them a tool that helps them spend less time on those things, they are likely to use that tool, which I think is great. So, some common sort of high-value use cases that you can think of: from a sales perspective, think about what you’re doing in your coding process — can AI help you do that? Can you describe it in natural language instead of having to click in a bunch of different places and add a bunch of different products? Can you have a smoother lead-to-rep handoff? So, with something like a Salesforce SDR agent or a prospecting agent, can we have some of that top-of-funnel stuff get handled by AI, because that’s high volume, it’s generally low value, it only turns into high value once you’ve put them into the pipeline, right, and you’re turning that into an opportunity, you’re turning it into a deal. On the marketing side of things, you’ve got content generation, campaign optimization, things like looking at insights from your engagement data and figuring out what’s going on. Our team definitely uses it for content generation, as well as laying out sort of entire plans for different campaigns. And it’s not like our team isn’t spending time thinking through the strategy, they absolutely are, but now they’re not having to spend hours and hours actually hand-crafting all of these assets, right? They have a phenomenal head start by using AI to do that. On the service side, I mean, we’ve talked about the customer service agent, you can do things like auto-resolving tickets, routing cases to the right people, understanding plain language. So if you’ve got emails that are coming in and that’s what you’re using to generate your tickets, well, a lot of times a human has to go in and kind of read the plain text from that email, and then determine the information about the ticket, categorize it, and all that kind of stuff. Well, you could put AI on it and give it that task, and let it go in and figure out what it thinks the actual categorization information is, and allow it to offload that, right? And then it’s important to note that it’s never gonna be like a hundred percent accurate, but if you’re seeing 90, 95% accuracy, that’s a huge time save, right? So now you’re moving from the process of actually analyzing every single one of these cases on a one-off basis, to basically triaging exceptions. And so a lot of that time spent is now doing other things, and you’re checking the AI, right? It’s really important to audit it, keep that human in the loop. But even with the exception management, you’re still saving a ton of time. Anyway, it all kind of goes back to the scalable business framework, which is: first define the process, and then identify where automation and AI can fit in. So it’s always kind of start with that problem, start with that process, and then figure out the right tool for the job.

Eric Housh

Absolutely. And if you don’t have a copy of the Scalable Business Framework, we’ve got a link in the show notes. Big news on that: the audiobook version dropped today, so I’ll be taking a listen to that in the gym tomorrow morning, for sure. Wrapping it back up: HubSpot’s making AI something you don’t need to custom-build. Again, a little bit of a different approach from Salesforce, it’s baked into the platform, designed to solve real, everyday business problems. So if you’re on HubSpot, I mean, the next step is really easy. Go check them out, play around, ask for those leakages in time and energy in your processes, and then turn these tools on and test them out.

Zack Terry

That’s right, that’s right. And I’ll kind of echo something that we’ve heard from Dharmesh, who is a co-founder at HubSpot. And I think he would call himself the chief AI enthusiast, he’s super into AI, he built out this whole agentic AI platform, but he had a great message in a 30-minute presentation that he called “You to the Power of AI.” Really focusing in on this idea that we’re not in the business of trying to replace people, we are in the business of trying to empower people. And so if I think about myself, and then myself with AI tools, I have more capabilities with those tools than I do without them. And so it’s really about freeing up your team from repetitive work to focus on higher-value activities, and creating this sort of hybrid human-AI team, where the team is leveraging these tools to become better.

Eric Housh

Yeah, we talk about that concept of working on the business versus working in the business, and AI is making that a lot easier for our most valuable players. So for you out there listening, that, you know, maybe this is interesting to you, AI still may be a little bit intimidating, we’d love to have a conversation. If you hit us up, fastslowmotion.com/AI, we are into this stuff. We’ve got several experts on staff that Zack’s been working with. We can walk you through what’s available, we can talk about your processes, we can get you started. All of your competitors are doing this, so if you’re not doing it, the time is now to reach out and start that conversation. Thanks for listening. Zack, thanks, it’s been great as always, man.

Zack Terry

Thanks, Eric. Good to see you.

Eric Housh

See you next time.