← DTC Podcast
Loading summary
Transcript34 lines
- [00:00]
A
Claude Cowork has proven that it's like having an assistant that is capable of creating task lists for itself, asking follow up questions. It's becoming like a real co worker for me. And I think that's really the change. There's people who download Claude and you know, they'll try to use it and they'll just say like, it's just kind of like another chat. But if you take the time to really understand how it works, that's when I think you can start leveraging it. And so for me personally, I think I'm still in this building phase where I'm trying to figure out foreign.
- [00:35]
B
It's all killer, no filler. I'm Eric and this week we are back with our resident AI expert, Braden Germain, senior content manager over at our sister agency Pilothouse. Welcome to all killer, no filler. What's cracking in the AI world?
- [00:48]
A
Oh man, let's, I mean last time we talked, we talked, we kind of touched on Claude Cowork. But I feel like, I don't know if everyone else has noticed this, but it certainly has kind of blown up in the zeitgeist. I think a lot of people are using it and talking about it and yeah, within the last few weeks we've been starting to adopt it more and it's just super stoked on it.
- [01:07]
B
I think of the metaphor about the idea about humanoid robots and there may actually be only a short period of time where we used humanoid, where we're going to use humanoid robots because in the long run we're going to have purpose built robots. That's for instance, like the robot that builds the houses will have a spot that lives in the houses kind of thing. So that will become your robot of the house. It won't be a humanoid robot. And when looking at some of the topics you wanted to talk about today, it was kind of on the scale of that idea where you're just, just learning how to rethink what work is and how to work in general. And rather than thinking about how to do the things you do all the time with these robots, there's, there's a, a layer deeper maybe where you're actually talking about, you know, how to think about work itself.
- [01:50]
A
Yeah, I mean, I think for a long time people were just using ChatGPT and they were using these tools as just like a chat. I mean, that's how I was using it. Maybe I'd get into, you know, custom GPTs and things, but most of the time it was just the throwing questions at it. They had pretty good memory and you would just get answers back and then you would use that for your job just to kind of speed things up. But now it's really turned into this coworker. It's like having, I think Claude Cowork has proven that. It's like having like an assistant that is capable of creating task lists for itself, asking follow up questions. I love how Cowork kind of gives you these like follow up questions. You can like give it more context and you know, it will execute on a plan and the outputs just become, become better. And not to mention all the ways you can connect it to all your tools. It's becoming like a real coworker for, for me and I think that's really the change.
- [02:37]
B
Walk me through your day. Walk me like when you talk about it becoming more of a coworker and not just a tool that you query or whatever, like walk me through, you know, one of your workflows or like, do, you know, do you turn it on? Like when you sort of sign in in the day, do you sort of put it on and sort of have it be working alongside you? Is it still more sort of, okay, I gotta, you know, send all these invites, I gotta do all this and just use it.
- [02:58]
A
I think that it takes a lot of learning. Like I will, I will say, like, you know, it does take a lot of learning to kind of set everything up. And I think that's like, part of it is that there's, there's people who get in, you know, download Claude and you know, they'll try to use it and they'll just say like, it's just kind of like another chat. But if you take the time to really understand how it works, that's when I think you can start leveraging it. And so for me personally, I think I'm still in this building phase where I'm trying to figure out all the skills that I need to give it so that when the time comes, when I wake up in the morning, when I log into my computer, I can say like, hey, give me a brief on what I need to do today. And it can go into Slack, it can go into my calendar, it can look at my Gmail and it can give me like a rundown. But that's like a process, a skill that I need to build within it. So I'm still kind of in this like building phase of like figuring out what skills are really important to me and how to leverage them. But I think, you know, we could talk about this. Like, I'M really interested in breaking down what my day consists of from a really granular level. And so say I have, say I want to make an ad, saying, okay, Claude, make me an ad. It encompasses a lot of little micro tasks. And so I'm trying to break down all these micro tasks of what it consists of making an ad, getting that delivered to like a designer for them to execute on. And so then I'm looking at all the ways that I can say, oh, I can automate this process. I need to go into Google Drive, I need to create Google Drive links. Well, I can automate that process. I need to find example ads that maybe I could replicate from another brand or I need to come up with a new concept. So I'm trying to break all those things down, create skills for a lot of those so that then I can kind of only worry about the things that involve my taste or things like selecting the right assets or going into Nano Banana and creating assets from scratch where I get to kind of use my taste and creativity. But then I can kind of offboard all these structural things that it's just, it's always the same for me to do. You know, I got to copy and paste this here, I got to do this, this and this so I can try to off board that stuff.
- [05:06]
B
And whereas if you had just given it the instruction like, get me from point A to point B without all the taste interventions, you would get something terrible and you'd be like, oh, AI is terrible at this kind of thing. But it is by breaking it down and applying human in the loop where it's important, you can get great results.
- [05:21]
A
Yeah, yeah. And I think like YouTube is such a great resource for people to learn about this thing. Like, that's like what I've been doing. I've basically just been trying to train my YouTube algorithm to just feed me Claude Stu. Because a lot of times what I'm finding is a lot of the content online is not necessarily about DTC. It's more so about SaaS products, it's more so about entrepreneurs sort of thing like, you know, business owners who are just like one person. And from like a marketer's perspective, I kind of need to see what they're doing and try to internalize those mental models so that then I can understand the process, the way that they're thinking, so I can go start executing that on like my day to day. So that's kind of like, you know, that's like what I've been doing is just trying to learn as Much as I can from YouTube and be relentless with watching videos just so that I can, like I said, get those mental models down.
- [06:08]
B
Can you give me an example of a topic that you've sort of, like, integrated, that you sort of saw and was like, okay, this is something I gotta do.
- [06:14]
A
I guess the one thing that comes to mind is like, standardizing things across client accounts. So I think for us on the pilot house side, it's just really important that, like, if there's multiple people who are writing, for instance, copywriting, you know, if there's multiple teams, like the email team, the meta copywriters, the people who are making the ad creative themselves, landing pages, if there's multiple people who are engaged in copy, that all those people are speaking the same language. And when you have four or five clients, it's difficult to really digest their brand tone and voice. It's doable, but it's difficult. And so what I've been really trying to solve for a couple of our clients is how do I create skills that are built for clients so that, like, I can share them across multiple teams? And, you know, that means like a base layer of like, defining what good copy is. So having a few, like, reference files within the skill to say, like, this is what really good copy is. The next level is like, this is how the brand speaks. And then the next level is like, here's some really good examples. And so creating this really strong Claude skill that then I can share across teams and like, then that way, you know, you're going to get way more client trust because they're going to say, wow, they're just, just crush and copy every time. And that's really the name of the game. So that's like one way that I'm actively building something out right now.
- [07:36]
B
I actually haven't. I've worked with Claude Korc a lot. But how is the training? Is it just as easy as it is to do everything on Claude?
- [07:43]
A
Well, the cool thing with Claude is like, there's a skill called Claude Skill Builder, basically. So it's a skill that Claude has. So you can basically just call on this skill and say, hey, use your skill. Backslash skill. You can say, use backslash skill to help me build a skill for X, Y and Z. It knows exactly what to do. It will ask you questions, it will ask for resources, you can feed the information to it. It will basically build the skill for you so that then you can kind of create this workflow or create this, this function and then run it, run it consistently and so that's effectively what I've been doing is using the Claude skill builder skill.
- [08:18]
B
Very cool.
- [08:18]
A
And it's. And it's easy. And it's easy.
- [08:20]
B
Yeah. Like everything is with Claude, especially when you break it down into small pieces.
- [08:24]
A
Right.
- [08:25]
B
The part that really spoke to me here was something because I'm, I'm going through this. We're trying to really optimize the DTC website, optimize our content there and, and just sort of like with it about not only how I could optimize the website, but how, how you do it in an ongoing way and sort of in an, in an automatic way. You have a point here about Andrej Karpathy and auto research, slash auto optimization. Talk to me about that.
- [08:45]
A
Yeah, so I kind of stumbled upon this. I was listening to another podcast and they were talking about their landing page and they were talking about how amazing their team is, how they ran this landing page through auto research. And so I kind of was like, well, what is Auto research? I got to look into this. So I looked into it and it's on GitHub and it's a skill that Andrej Karpathy wrote that basically allows you to treat anything that has like a KPI so that you can connect with an API as like, basically like a self generating auto research. So it will take like, you could take like a landing page for example, and then if you have an API that you can connect to it, that kind of gives it the like click through rate data or the conversion rate data and then you can connect it to this skill and you can let it run for months and it will just continuously improve self, improve the website over time so that it's optimizing for conversion rate. You could do that for ad copy, you could do that for even ad creatives, email taglines, email subject lines. If you're doing cold email outreach, like you can basically feed the API into this skill, this auto research skill and optimize things over time. And people have talked amazing things. I wouldn't say I'm an expert at it. It does take a little bit of coding expertise. But I mean Claude is good at explaining things and, and it's definitely worth looking into if you're trying to solve those sorts of problems.
- [10:05]
B
Have you heard. I've heard a few horror stories about Claude integrating or you know, crawling Ads Manager and having it be a problem for the account that Claude is integrating. Have you any, any guardrails around that?
- [10:17]
A
I haven't used Claude with Ads Manager and I have heard from People that they've had their issues with Meta, like, Meta will shut down their ad account if they're using Claude, integrating Claude with their Meta Ads account. So I think that's a big flag there. But I have heard people getting a lot of success with Manus because Meta bought Manus, and so that might be a better option there. But then there is also, like, I think other companies like Triple Whale. I'm pretty sure you can connect APIs to Triple Whale as well to get that data out as well. So there's opportunities there too. But definitely worth looking into the guardrails and being careful with Claude and integrating it with Ads Manager.
- [10:59]
B
And then the other piece that you mentioned here, it kind of completes my weird metaphor that I started with with. It's a different way of working, a way that humans aren't always capable, especially when you work in teams. You have sprints where you go through big things and then people don't always, you know, follow through and. But where with AI, you can have always on execution. So this is the sort of the final piece of these systems you want to create is thinking about how can they be always on and hold. Hold the humans accountable even.
- [11:26]
A
Yeah, I listened to this great video and the guy on the video was talking about how you kind of want to win an AI. You want to almost like AI yourself out of a job. It's a scary road to go down. But I think that learning these tools, learning how to automate, learning how to, you know, have a computer that's running that you can call on anytime to run and execute on tasks, is going to serve us, serve us really well. And I think that, like, you'll be able to bring your agents or your ideas to new jobs or new opportunities, which I think is really going to be awesome. So that's kind of the mindset I'm kind of taking is like, hey, what can I automate? And now with these new tools, Claude has this new dispatch tool where you can set up a Mac and you can call on Claude from anywhere on your phone. You could text Claude and say, hey, I found this great ad idea while I was scrolling Facebook. Here's a screenshot of it. Can you execute on the skill that I have? And it will go and it can, you know, you have a computer that's always running and it can go and it can run those skills for you. So you all of a sudden become this kind of like skill builder. You're taking your expertise and then you're kind of driving the machines as opposed to like doing the tasks themselves. And I think that's really kind of what the future is and what I'm trying to kind of get to with how we're operating here at Pile House.
- [12:44]
B
Very cool. Any other sort of starters on like Claude skills that you think would be really valuable for people in your position, either media buyers or on the creative side?
- [12:54]
A
Yeah, I think so. GitHub is a great resource. You can go on GitHub. You can also go and like just Google and just search Claude skill libraries and people have generated, you know, all these different skills that you can just download and then upload into your Claude. So worth looking into. There's some great ones for like ad copywriting. There's like a full marketing suite of one where it's. It's great for like SaaS companies and it basically has all these skills built into it. So you can, you know, call upon say the cold email skill and get it to check your work or you can say call upon the copy editing skill. So go and check out these GitHub libraries for Claude skills and then what you can do is you can actually look under the hood, which is what I like to do is I like to go on GitHub, I like to find these like, kind of like marketing skills. And then I like to see what is the reference files. They have all these markdown reference files inside and they're just plain English. So you can just read them. What do they say? And all of a sudden you can start to see, okay, well this reference file is talking about what makes great copywriting. This one's talking about what are the, you know, the best practices for a number of characters in meta ad copywriting for like ad copy. This one is about. And so then all of a sudden you can kind of start understanding a little bit better dissecting. So it's a great place to get skills as well as understand how skills are written from a bunch of other people. So definitely recommend doing that.
- [14:13]
B
It's such an interesting movement. The. The news this past week of Claude's code base being leaked.
- [14:18]
A
Yes.
- [14:18]
B
Was. Was an interesting one. Like it is inherently it trends towards being open source in a way. Right. Just with the ability, with the, the fact with AI models that you don't need huge computational ability in order to run them, I think ultimately may lend themselves to be to it being a more decentralized technology. Tell that to Claude and OpenAI that are the fastest software companies and you know, growing software companies ever. So they're centralizing pretty good. But wild. Wild times.
- [14:49]
A
It is wild times. So much has happened in the last, like, three weeks. It seems like every week it's something completely new is coming up and it's across every axis.
- [14:57]
B
We just on the agency community joinagency co. If you're an agency out there, we just a call. We just met someone that we're bringing in to speak to the community that just specializes in building whole systems for all sorts of difference for the medical industry, for the legal industry. And it's just this, like, proliferation of people that are going to be able to sell the tools and to make. To, you know, help make organizations, you know, superhuman, essentially. What a fun time to be in business.
- [15:23]
A
Yeah. Like I said, learn the tools, go on YouTube, learn some stuff. And it's pretty fun once you start kind of understanding things and understanding the possibilities, so totally.
- [15:32]
B
All right, thanks, Braden. Talk to you again soon.
- [15:34]
A
Thanks, Eric.
- [15:41]
B
Thanks for listening to today's episode. If you're not getting the DTC newsletter, you can subscribe for free at directtoconsumer. Co. And if you want to learn more about Pilothouse's all killer no filler services, take off to Pilothouse Co. I'm Eric Dick and this has been the DTC podcast. We'll see you next time.