Perplexity Computer: The Super Agent Playbook (5 Real Workflows)

Okay, perplexity computer is just out and it is incredible. I just give it one prompt and it built me a live interactive website using polymarket data. Not just that, but we used it to grade HubSpot's entire product marketing strategy. It built us an incredible product marketing skill that you can use because we're going to give it away at the end of this show. All of that and more on this episode of Marketing Against the Green. Here's a quick word from HubSpot. HubSpot helped Tumblr solve a big problem. They needed to move fast to produce trending content, but their marketing team was stuck waiting on engineers to code every single email campaign. Now they use HubSpot's customer platform to email real-time trending content to millions of users in just seconds. The impact? Three times more engagement doubled the content creation. Want to move faster like Tumblr? Visit HubSpot.com. Here we are on a super agent bender on the show today. We've done a bunch of stuff with Manus. We've done some Claude code stuff with our friend James at Boring Marketer. Now we're all in on perplexity computer and all the cool stuff you can build with these really advanced agents. We're going to walk you through today some really awesome marketing growth use cases. But before we do that, off air we were talking about it seems like all these AI companies are converging on one core use case that we'll all be doing. It's just a question of where we'll be doing it. Maybe you break down what you mean by that and what you think that use case really is. I guess if you've just taken the last week or week and a half of all of our episodes and you're kind of watching them back from Manus from Claude code and even the super scale where they had a tall and this paid agent. What is happening? You have a super agent and that agent has connectors and skills. It's able to connect to whatever amount of tools you give it access to and then it has a bunch of skills to do things in those tools. It's breaking my brain is every single thing is converging on, hey we're going to be a super agent and you give us tools and tools or things like, hey you can access my email, you can access YouTube. This is what OpenClaw has got a ton of press about it. You can go back and watch the episode we did and that. OpenClaw is the same thing. An autonomous super agent connects to your tools, has a bunch of skills. The skill is like I can do content, I can do paid, I can do AEO, I can do some kind of pre-sales work. I've started to go what is going to differentiate all of these different apps from each other. They kind of all are converging on, hey where's this? A super agent can do things autonomously, we can enable tools and it has a bunch of skills that can access. I think you're going to show perplexity computer is in the ballpark of one of those super agent and it's pretty sweet. It's pretty sweet. I'm in this all the time. I spent the entire weekend in Cloud Code. I built myself a financial advisor on Saturday night whilst I was watching a movie called Giant by Princeton's Eamland, a huge Princeton's Eamfan. I'm just there. I'm in my life. I'm building a financial advisor here at Cloud Code and watching a movie which is a bad thing to do. I don't think you should. You just take some time away from your laptop, it's just like a mental option that you are. But I couldn't. Then I get phomo because you're like, hey I'm using perplexity computer. It's sweet. It's $200 a month. I don't care about it. I'll pay for it. And you show me this stuff. I'm like, oh, this is awesome. We're going to show you that some of that stuff. But I'm like, there's too much. There's too many super agents with tools and skills. I don't know which one to use. So I do think what we're saying here, we're about to get into perplexity computer, which is another one of these super agents. I think the advice you're giving folks is like, oh, this use case is getting pretty clear. And it's going to be kind of a primary where you interface with all your other software and data and tools. You're probably better off to focus on one of them. I think you're a better and really build and iterate and get right. And that list is essentially Cloud Code/Cowork perplexity computer, Manus, OpenClaw, or anything else you would put on that list just for people. You know what's fascinating? Only because of an acquisition, OpenAI wouldn't have been in your list. They're only in there because they bought OpenClaw, which I think is fascinating. And Google isn't in that list, yeah. So that's very fascinating. And by the way, OpenAI is in the list where I'd say the most difficult one. OpenClaw is the most technical, it's the least accessible. Most people watching this show probably will not go and use OpenClaw. Cloud, Cowork, Code, or perplexity computer is probably the best option for the average person watching this show. So I did see here in that Mac Minis are sold out in New York City because of OpenClaw. I think OpenClaw has caught the zeitgeist. I think they are the one that has the community being built around it. This is a great marketing lesson. Like the guy I watched him on the lecturing mend and he just wanted to do something different from everyone else and basically give it a personality, a quirky personality, which is why it had the lobster and make it feel fun, not make it feel like a tech-caron. And I think he is an example of a brand marketer. He branded himself against the curve. And people now think of that as the fun community doing the end of the stuff. And so it does show you that I think marketing is going to be the de facto skill to have because you're going to have all of these super agents. They do very similar things. And OpenClaw because of one reason or another is the one that has caught the imagination of people. Yeah. Hey guys, one of the most valuable things we shared in the show today was this amazing product marketing skill to audit the product marketing on your website. We're giving that away for you for free. So if you want that skill, click the link in the description below and we're going to give that to you. It has never been a better time to be a creative person with taste. Right. Never. And we've said that a lot on the show, but let's show you exactly what that means. Let's share a few computer examples. So Kirin, you and I write in a book. We've talked about it a little bit on the show before, right? And so this is Proplexity computer. When you sign into Proplexity, you have search and now you have computer. And when you start a task in computer, you describe a task and that task just runs in the background until it's done, like you were running it on Cloud Code or in a terminal on your local computer, right? Hence Proplexity computer. So I want to show you the tasks that I kicked off last night. I want you to build me a skill MD file that can design the world's best business book cover. We're currently writing a book about marketing AI. I want to be able to give a full brief on the book and I have the skill execute five remarkable design concepts. To do that, I need you to go and scrape and reverse engineer the covers at the top 100 business books sold over the last 24 months. So I need you to find the images. I need you to analyze the images. I need you to name the core components, design styles, how they interpret, how they're going to connect to the core content, message of the book, blah, blah. So there was some really complex stuff I asked it to do, right? It had to compile a ranking. It had to go and scrape all those images. It had to then go analyze each image and compile insights from each image, right? And then it had to take all of those insights and build them into a skill that you could use in Proplexity computer or Cloud Code anywhere. Remember, skills are one of the most important things we talk about on the show. And it just went and did it, Karen. I think that's what's really cool. Yeah. If you were doing this in Cloud Code, like you probably would need to install FireClaw, you would have had to do some more technical work. It might have been better quality in Cloud Code because you would have had much more granular control. But I do think that's one of the trade-offs here is ease versus like depth of control. And Proplexity computer, I do think, has a little bit more depth of control. So you can see that it builds a plan of what it needs to do. And you and I have been talking a lot about like planning, becoming a really big important part of how you work with AI. I'll probably do a show on that soon. So then it researches through all the best-selling lists what the books are, compiles the lists. And then what's interesting is once it has a text file of all the books, it collects the cover images, and then it spawns, Karen, four different batches to generate and review the covers of everything. Right. It's like creating agentic teams. Yeah, now let me launch four parallel subagents each analyzing 25 book covers in depth, which is pretty wild, right? And so look, the other thing about Proplexity computer is that it's model agnostic, which means it can use Google models, opening AI models, cloud models, whatever, right? You can see right here it's using Claude Saun at 4.6 to do this. And it's telling you, each of these 25 titles took 10 minutes. So basically if they hadn't been batched, it would probably take them like an hour, right, to do all of this. And so then it gives you all of the batches and the learnings. It writes the skill file that you can now use in everything, gives you, basically we now have a book/cover design file that I can send to you and we can both iterate on book covers together, right? It gives me all the learnings and then I then gave it a brief of the book that we're writing. And then it created a bunch of concepts, which are cool, but I needed to be able to share it with you in a more easy way. So then it built a website. I hadn't built a website so that I could just send you the link. Basically built to this beautiful website of rationale of why that cover is the way it is. And by the way, these are good mocks, man. These are really good mocks. I was really good. It's a really good box. And the reason they're so good is because to your point, it's calling the best tool for each task. Yes. And then I'm going to collect these image tool. Again, coming back to the opener of the show, the super agent that can use the best tools and skills to complete the task is going to be the primary way you use AI. And that means the results are going to be like best in class. I remember when we first talked about AI all the way a number of years ago, we talked about code becoming disposable. And there's so many use cases of it. Here's a website that I've just built for this five minutes. things so you can show me these and then never think about it. There's two things that I think we got really right, disposable web for sure, agents, yes. And then I didn't get a third. I would just like to go on record for the pod community, even if this gets edited out, that I was right about the X valuation, by the way. Oh, yeah. And I had a whole podcast on if Elon would be able to turn Twitter around. And I just saw that the valuation is 6X above where he bought it. Always investing the person. Exactly. All the way back. But I want to tell you about another podcast I love. The DTC pod hosted by Ramon Barrios and Blaine Bullis is brought to you by HubSpot Media. DTC pod is a podcast about all things direct to consumer. Ramon and Blaine cover everything from starting, growing, and optimizing e-commerce stores and direct to consumer brands. They talk with founders, marketers, platforms, creators, and marketing and growth agencies to cover topics like brand building, social media, influencer marketing, website conversion, paid media, Facebook ads, and much, much more. If you're interested in the stories behind your favorite consumer brands, this podcast is for you. They did an amazing show called MetaAdSecrets. How top DTC brands spend 300K monthly, profitably. You can listen to the DTC pod wherever you get your podcast. You had a cool thing that you wanted to show. A cool thing to go do to test this out is, they have this cool thing where they have live examples, which is great onboarding tool. They have a lot of live examples here that you can go click on, you can see at work and bring up the final asset. I hadn't thought about this. They did this state of US politics and the core source is polymarkets. Polymarket is a tool that where you can bet on outcomes, anything at all. That is such a killer thing. They have the entirety of old bets being made around US politics in a single website. So cool. You can see here that they have stuff around Trump's presidency, federal reserve, and it's just pulling in all of the info from polymarket. And so you can do that about anything. Like you can take any kind of topic and you can basically create a website all around the bets being made in polymarket on that topic. So like for sports, I could pull one together for everything that's happening around the premiership and send that to my friends to show like, these are the outcomes that people are bending on. I think this one is really killer because polymarket also tends to be right a lot of the time. So it's not a bad way to understand what's happening in the world because you can literally do everything. And by the way, I do a separate rev of this where you put a confidence score on each of these based on polymarkets success in each of these categories. And you put like a confidence score on them, which would be pretty cool. Yeah. So I think that's a really good use case. You can go basically ask it to just build something, use polymarket as a source, and it will actually go do that. There was another funny one actually. So this one's kind of cool, right? It has all of the AI leading companies, has the whole watch list built for them. But first of all, I thought this was like someone had just, you know, you can publish things publicly in-- Yeah. --clawed. And so I probably actually have the same thing where you can publish your work publicly. And I thought someone had accidentally done this, but then I realized it was an example. But pretty good, this is someone like, get in a offer at Stripe as a senior product manager and using perplexity to come back and give a counter. End of your year. Wow, it's really hard to deal with humans anymore, like negotiation because every human is so well-informed. Right? It's the AI just builds the counter offer. And so that's a pretty interesting one if you're going through a role at the moment and you need some help. This one was really well done. Like research the different base salaries, research what you should get in RSUs, and then came back with a counter offer. I thought that was a pretty cool one. That's actually a really sick use case. You know, I've been really impressed with the Kirin. Let me show a couple other things, and then we'll talk about some best practices also for how to use perplexity computer. I gave it my version of the marketing touring test, Kirin. This is the thing that I ask every new AI advancement to do, okay? Which I basically gave it a long instruction of, I want you to go find the fastest growing, not obvious companies out there and reverse engineer the unique marketing strategies that they've done, okay? Basically, I'm trying to create a mechanism to discover new hacks to grow. Right, yeah, yeah. And so it's like, I wanted them to go out, find a list of these high growing websites that aren't like the anthropic open AI's of the world. Like the next set down, what's driving their growth, what tactics are they doing to drive that growth, what could be learned from that? Okay? And so it's pretty complicated because they have to go and figure out, find enough data to be confident in a ranking of these companies, and then basically reverse engineer what those companies have done. And they've looked at data from similar web and lots of other purpose. They also like look at our organic traffic decline and they're doing this like relative to HubSpot. But what's interesting is that the end of it, Kirin, I've essentially have a full website report of what these companies are doing. And there's seven patterns driving, these are B2B only. And I think you and I will largely agree with these, but I'm interested in what you think. PLG, DOM, that's the did 15 companies. Next time we'll do a bigger sample. PLG, programmatic SEO, user generated content, adjacent category SEO, product loops, community driven distribution, and low ad spin. Yeah. Where's it going the ad spin from? I have to go back and do all that. I don't think the ad spin one is right. But what's interesting is like it's driving deep. So if you go into near zero ad spin, it's trying to find percent of traffic coming from ads. I think it's what it is. It's not doing actual ad spin. So because this is traffic, not customer, just customers, it probably look very different. But I wanted to do a breakdown here. So like linear, it has really high direct traffic. So you're seeing a bunch of interesting interactive ways to look at cool data. And if I am looking to grow, there are some tactics that I could consider. Yeah, reverse engineering companies is a super cool one. Research in general is just such a solid problem. You can base anything that's external you can get. I think this is a really good example. What I had perplexity computer do is I had to basically build a skill of what great product marketing looks like. Then I had to crawl the HubSpot website for all of our products. And then I had it use that skill to audit how we were doing. And basically it's like, hey, our big time product pages are doing really well. It's our feature kind of second tier product pages where we could have a lot of improvement. And like Kieran, for example, I think this is probably true for almost every company. Right? But it's something about seeing it in this way. Where you're not having the person who's doing the work, doing some audit, and you're going to have it by us. It's like, I think you and I am the entire product marketing could just sit around and look at this and be like, oh, yeah. It's kind of makes sense. And there's some obvious areas where it's like, oh, maybe some of these have been lower priority features. And maybe some, but like sales forecasting and analytics, like we should make those a lot better. Right? Let's go make those better. They should be a lot better because we should just automate the creation of them. Exactly. And so what I think is really cool is that I asked it to build an out of ranking algorithm and score everything. And so it was able to give us that ranking. Tell us what we're doing well. And I think, look at what our weaknesses are. Don't you agree with all of these? Yeah. I think this is really good because you just get this from another agent who actually goes off and fixes them based upon this. Correct. And it outlines the methodology. But this is to say, if you're running a team, especially now, you can have incredible insight into the work you're doing and your team's doing without having to spend like literally your entire weekend going through stuff manually, which I think was the alternative up until now. Yeah. Yeah. I think what's coming out in 2026 is like such a-- Step up. Jump up. Step up in terms of the tools we've had in the past. And so there's a few other things to Karen. One is that all of these tools, Cloud Code, Cloud Code Work, Manus, Proplexity, everything, they're about connecting data from lots of places. And so Proplexia, just like all the others, have all these awesome connectors. You can just connect all of your HubSpot data, right? And start going to town with how you want to visualize your email data. All of that-- that's pretty cool. That's pretty awesome. We have the tooling, right? There's just no more tooling I suspect the average person use. We have way more product than the world could use at this point. Yeah. I think the thing that is hard is how do you start to fit it into your workflows? Or how do you change the way that you work? Changing your habits is the harder problem to solve today than actually-- the tooling is amazing. Proplexity is amazing. I'm looking at it thinking why there's so much I could do here. I spend a lot of my time in Cloud Code. We love Manus, but it's hard to go back and continue to use it, even though I have it on my telegram, on my phone. OpenClaw, you have to come back and do an OpenClaw episode with some of the marketing use cases. Just crazy. What's interesting, Karen, is that we're starting at the point where the technology is no longer very-- the operation, the implementation, using the technology is becoming the problem. That is the problem. And I guess before we close out here, I think it'd be helpful to our audience if we try to pause it some ways to do this. I will tell you something I was doing last night. You can tell me if I'm a crazy person or not. We have a Glean here in HubSpot, which allows you to access all of your internal data with AI in a really semantic AI search kind of way. Karen, I was trying to back into all of the roles on the marketing team and the core skills in those roles so that we could then create skill MD files. for all of the skills for each of those roles. Like, I think one of the places you would start is like, if you're a team of two people, record the work that you do on a daily basis in video, put that in Gemini or get the transcript and put it in cloud or something, have it tell you what type of skills you need and then work to iterate and build those skills so that you can do the work you're doing much faster, much more efficient. That, to me, seems like one of the most obvious things that nobody's doing that we will all be doing like six months from now. - Yeah. - Do you agree with that? - I think actually loom in yourself doing work and using that as a transcript to build AI workflows from is actually great. We do that somewhat internally where we take looms, like literally just do work and narrate yourself doing work and you can turn that into skills and workflows. The other thing I would suggest that's kind of the following from yours is do one at a time because you could also have 50 skills available to you but then you have to get into the habit of integrating those into your workflow. And that's the hard part I think people struggle with is changing your behavior is default into using those skills when you have to do that task. And so for me, what I tell people is, do one workflow at a time. Build skills, build the things you need and then when you go to do that workflow, try to start to default to the AI way of doing it and just go workflow, workflow, workflow, workflow rather than I think what a lot of folks are doing is just building things like build, build, build, build, build. But they're not like using them in any meaningful way and I think that's going to be one of the outcomes we see for a period of time where it's so easy to build stuff. There's gonna be more things being passed around within companies than anyone can actually use. And so you have to start to like do one workflow at a time and make sure that you're using those things and they are helpful. For me, you have to use those things too care because you have to iterate on them and make them better. - Yeah, I think it's one of those things where it's like, cool, let me build this thing, right? Which is what you're saying. But now it's like, no, I need this thing to be really, really great. Let's have a product marketer and I want a skill to do my first draft a product copy, right? I probably need to work on that skill for like 20 to 40 hours. - Right. - I need to give it all the work, I need to question it, I need to give it my point of view, I need to do all of those things, then I need to test it. And basically that skill's not good enough until you can run it on a new product and basically have no changes. Yeah. And until you're like, oh my gosh, this is like 99% of the way there. It's not good enough. - Yeah. - And you shouldn't actually roll anything out beyond like one individual human until it gets to that point. Like you should have like a big repository of skills until it gets to that point, I don't think. - Yeah. One of the things you can look at is number of edits to see how good your skill is. - Yes. - And if you have a curve, you have to get the curve right down to like minimal amount of edits. And I still think a lot of it comes back to, there's gonna be this messy middle where it's really easy to do a first thing. And then you're like, oh, this is like great. Claw just built it all for me or perplexe built it all for me. I'll just start to pass it around. Whereas all of the value will become, I'll use this repeatedly and iterate and make it better with the AI. And I think that's gonna be the way to value is like, integrate into workflow, use it every single day. How do I make this much, much better and make that systematic way of working? - Yeah, maybe drop us a comment if you would like us to do, you know, a show on recording your workflow, turning that into a skill and what this process is. Like it might be valuable, but I don't know for sure. So if you think that'd be valuable, a show drops in comments that if it is, we'll do that and kind of show you kind of soup to nuts, what that would look like to do it. But I think your advice is right here in that it's about focusing and not spreading yourself too thin, using too many tools and not building too many low quality like, things that skills being stuff junk in the world, right? Which is AI Slop was one thing, AI clutter is the new thing. Right? It's like it's just so much clutter of AI in your life and you're like, what do I have to do? - It's a little less, yeah. Yeah, it all looks pretty great. Like I said, don't mean head, right? Like you got these dopamine heads, right? - Oh, I made that. No one's actually using it and I'm not using it, but that cover skill was like my fifth iteration of a book cover skill. - Yeah, yeah, yeah. - And like, and it still probably needs like three more iterations to like really, really be good. - We're fine with it. - And the last couple of things I would say before we close out the show here in, is that Proplexity Computer is only available in Proplexity Max, which is 200 bucks a month. So it is expensive. It is only available on desktop, cannot use it on mobile yet, which has been a real hindrance for me. I really want the mobile use case. And there are some tips that I asked around how to use it well. And it's basically like, hey, go and give it complex tasks and use the connectors, all the stuff we talked about. But I would ask Proplexity or Clawed how to use it specifically to the core problems you have in your role to get a little inspiration. If you're using it or co-work, BAN has anything else. I would be really focused on how you were using one of these super agents. - Yeah, like ask it to build you an onboard and plan. - Exactly. So good. All right, this is fun. Shout out to Proplexity. I think they've built a really good product. I'm excited for this super agent battle to keep going. And we'll see everybody real soon on the next episode of Marketing Instagram. (upbeat music) (upbeat music) - Hey everyone, you know, Kieran and I have been doing the podcast for a while now. We've been at this for a couple years. We love it. We could not be happier to be doing this. But we wanted to take things to the next level. We want to level up the impact we're having with Marketing Instagram. So the next step of our journey is something we're really, really excited about. We're gonna launch the Marketing Instagram newsletter. And Marketing Against the Green newsletter is going to be amazing. If you are a marketing leader, practitioner, you're in the trenches doing marketing every day. This is for you. We're gonna deliver right to your email inbox. And you're gonna get all the behind the scenes, frameworks, practices, tutorials. From us, from the guests we have on the show, and from people even beyond the podcast that we think are gonna be helpful and really have an impact on your day to day week to week doing marketing. You're going to love it. It is something we've been talking about for a while. We're really excited to have it out in the world. We've already got 100,000 marketers who are on this newsletter. Please join. It's completely free. We'd love to have you as part of the Marketing Against the Green Community. And it's easy. You can click the link in the description below or you can head to marketing against the grain.com/subscribe.