This is a YouTube interview (hosted by Patrick Biddiscombe) featuring Dharmesh (HubSpot co-founder) discussing AI agents. Dharmesh defines an AI agent as software that uses AI to accomplish multi-step goals by chaining LLMs and other tools, and positions agents as next-generation automation especially valuable for marketing tasks (e.g., CRM lead routing, templating ads). The hosts describe practical scenarios where agents are preferable to conversational chat tools.
Chapters
First of all, we're sharing a brand new platform that Dharmesh is working on that you're going to want to check out that has lots of free AI tools. Dharmesh is giving the marketing, it's a grain community, extra credits. So more usage of that tool. And I would love it if you all could harass Kieran in the comments.
To release the tool.
He's got an amazing AI agent that he is being selfish and not wanting to share. You need to hit him up in the comments and get him to share this tool because it is sick. That and some learnings on the future of AI, all from Dharmesh, all on today's show.
Dharmesh, thanks for so much for coming on the show. I've been teasing the show over the last couple of weeks on LinkedIn. I am an avid user of agent AI, so we are excited to have you on.
Excited to be on. This is a long time coming.
I know.
I feel like, I feel like we've been talking to Dharmesh about coming on the pod for like a year. Finally have like.
Yeah, but I wanted something like, worthy of your time, folks. It's like, okay, I had to go like build some real stuff.
I'm a user of the real stuff. And what we might want to do is just jump into like, what do we even mean by an agent? Because you were kind enough to give me some access. I've been like hammering away on Claude for a long time to improve my writing process. I hit a new goal on LinkedIn over the last week. Some of it is trolling, but like up to like over a million impressions a week. And actually one of the things that's really working is AI assistance to help me write. And you're agent unlocked a ton of creativity. So I want to like get in to show people, you know, what, what are these agents and how do you actually use them? Let me do this, Dharmesh, before I get specifically into this agent that we can kind of go back and forth on this. Do you want to just maybe give a little summary of like, what even is an AI agent? So you can kind of set the scene of like, what is an AI agent and why are you so excited to build this platform?
Sure. So let's, we'll, we'll take a step back. So there are varying definitions of what an AI agent actually is and I think people sometimes tend to overly complicate it. And so there's a spectrum of capabilities. You have very, very simple agents and you can have sophisticated agents that do kind of multi agent things and reasoning. But at the bare minimum, I think of an AI agent as a piece of software that uses AI to accomplish a multi step goal, a goal that requires multiple steps to get through. So if you think about it, the best comparison is that when, when you're using something like a chatgpt, a conversational tool, you're going back and forth, you're saying write me a blog post or do this and tweak that. But it's a very kind of synchronous back and forth model. Whereas with an agent you're giving it a higher order goal and it may need to invoke LLMs, it may need to invoke multiple LLMs and use other AI tools and some classic tools and bring it all together to pull together whatever goal you're trying to accomplish. So that's the difference between an agent and a regular kind of conversational ux. And the reason I think it's interesting, especially for the kind of marketers in the audience, the way I think about it is that back in the day, like way back in the day, you know, we used to do kind of what is now called marketing automation sort of by hand, right? We had developers in the back room working in B2C companies saying oh when this happens I want to send this email out and track this or whatever. And then along came the marketing automation category that says, hey, we're going to have non developers using like a workflow builder. HubSpot has one, there are others out there basically automate some number of these things, right? And so it's a script, a multi step thing that could do that. So a simple way to think about it is one class of agents can be these next generation automation. So I know what I need to do. But now that we have access to AI, the kinds of little steps that you can do are actually mind bogglingly powerful, right? Things you could never do before now can be part of that automation script. So that's one way to think about it.
Dharmesh building on that, you know, if somebody's watching the show today and they're like oh okay, what? When should I think about using kind of a traditional chat or conversational interface versus an agent? How should they think about that so that they don't go down the wrong path and waste a bunch of time or get frustrated.
Everyone should stay practical. The time to use an agent is when you have something where you need a human in the loop. But there are lots of steps that can be done without human intervention. So maybe the human kicks it off, maybe the human reviews the work before it goes out. So I'll give you an example. For instance, you could in theory say I want to build a. I not build, I want to do like lead routing in my CRM using ChatGPT. Like I'm a small business, I'm going to take the customer data, I'm going to paste it into ChatGPT or Claude and then it's going to tell me things, whatever. It's like, you could do that. That's sort of a manual process, but every interaction is sort of human based there. So you're not really automating a lot other than the kind of intermediate steps. But you could also imagine, hey, I want to build an agent that says, bring me the most recent company that was added to HubSpot CRM. Let's say I want to go do some research on that based on what's there. I'm going to invoke perplexity to answer some questions that I have. Then I'm going to write out my rules as far as how I want my lead routing to happen. If it's leads from Europe, I want it to do this. If it's in the tech business, I want it to do that and route it. And all that is doable with today's technology, with what I think is a high degree of accuracy and precision because you're reducing the numbers of degrees of freedom at each step of the process. Like, yep, AI is good at that, it can do that and you're just sort of chaining it all together and, and, or at least semi automating it. So that's where you want kind of agents to get built, where you sort of know what to do manually and you could do it manually, but you want to automate some of those steps instead of trying to do this kind of one off shot. With ChatGPT, we have a perfect example.
This is actually sets up the kind of agent that I got to build on Dharmesh's U platform really well. So one of the things that I've been obsessed about with AI in general is when, when can AI take some things and build you a template to replicate that thing? Because whenever it can do that, my mind starts to melt, right? Because I was like, hey, if I can take like all of these ads and templatize why they're good and just templatize them for the AI assistant to be able to replicate that, then everything can be templatized, like every singular.
