← Tech Brew Ride Home
Tech Brew Ride Home
(BNS) Brian's Vibecoding Journey
01:25:01
Chris is back! Hear about how he uses AI personally, and how I've been up at 3 am every day for the last month churning out my own vibecoding experiment.
Loading summary
Transcript298 lines
- [00:01]
Brian
So good, so good, so good. Everything you want for summer is at Nordstrom Rack stores now and up to 60% off.
- [00:08]
Chris Messina
Stock up and save on the brands
- [00:10]
Brian
you love, like Vince, Sam, Edelman, Frame and Free People.
- [00:13]
Chris Messina
Join the nordiclub to unlock exclusive discounts.
- [00:16]
Brian
Shop new arrivals first and more.
- [00:18]
Chris Messina
Plus, buy online and pick up at
- [00:20]
Brian
your favorite Rack store for free. Great brands, great prices. That's why you rack. Welcome to the first episode of How I AI, which is a bonus episode right now for the Tech Brew Rio podcast. But I don't know, like I've said for several weeks now, I'm working out a new concept here about talking about how people are using AI. So we're calling it How I AI. And to help me work out the kinks on this format, we've got our old friend Chris Messina here.
- [00:56]
Chris Messina
Chris, hello. Hello. Glad to be back?
- [00:59]
Brian
Yeah, it's been a while. I feel like we're on maybe every six months Cadence at this point, but anytime you're available, it's like five years
- [01:08]
Chris Messina
of things happen, and then we get together in six months. That's right.
- [01:12]
Brian
Yeah. And I just saw you a few days ago, so when it rains, it pours. So, as I said offline, I'm still working out the format. And I think what I'll always do is start by asking people how they got AI pilled or AI religion. What was the thing that made them realize you could use AI in your life before we get to the project, but what we need to do is. We need you to ask me that first.
- [01:38]
Chris Messina
Exactly. I was going to say. I mean, so that's. That's where this conversation starts. So.
- [01:42]
Brian
Right.
- [01:43]
Chris Messina
To set this up, like, we're sort of play acting. Play testing the format by starting with me being you. Uh, and you're turning me into your agent. I suppose in. In this context, that's a good way to.
- [01:55]
Brian
That's a good way to put it.
- [01:56]
Chris Messina
I'm. I'm. I'm a human version of Claude, where, you know, we can. We can roll with it. And it's a little bit more dynamic than you just being like, all right, Claude, like, I want you to interview me. Like, you know, how did I get AI pilled? Which would be not terribly compelling content. Uh, so, I mean, let's start there, you know, because I think it's. It's. It's probably even worth revisiting a little bit of, like, our backstory and why it's sort of like, shocking in a way that you got religion.
- [02:20]
Brian
You mean because we deployed millions of dollars in capital at an AI fund.
- [02:27]
Chris Messina
Correct. What I find so funny about this, and we'll just stand this for, like, 30 seconds, is this, like, I remember very clearly when we had the conversation in March of 24 or whenever it was essentially, and I pitched you on the idea of, like, AI varietals and the idea that vertical AI was going to be way more powerful in many respects than AGI, God mode, whatever. And the thinking and the thesis was that there's so many. It's almost like a fractal problem to get super deep into every use case and application, whether it's regulation or just, like, the feel of the software that we use. And so my thesis was that generative AI was going to introduce this new medium. This new medium is, as you've discovered, it's software, but it's software that anyone can write. And so, you know, you've had this. This resume writer's business forever that employed humans like, you know, from the 1800s or whatever. And, you know, you're bringing that into, like, the modern world. Just as, like, you know, obviously the job world is also changing. So it feels like something happened for you. I don't know if you just like, one day were like, you know, screw it, I got to do something. But bring us into that moment where, you know, you took the blue pill and suddenly on the other side, you realized that maybe vibe coding was something for you after all.
- [03:46]
Brian
Right. This is going to be specifically a vibe coding example, but Paul Ford was on Peter Kafka's podcast recently, and he was making the pitch to Peter about whether AI can work in Peter's daily life. And Paul said, I'm a guy that looks at problems to solve with software. That's how I think about the world. And if you don't think in that way, or we'll get to this. More importantly, if you don't think that you're qualified to look at the world that way, that's the sort of divide
- [04:15]
Chris Messina
for a very long time, well up until even, like last December, that was predominantly true for most people. Solving problems with software. You know, it's sort of like, you know, the person that has a hammer and everything is a nail, like, that's great if you, you know, can wield a hammer, but if you are clumsy,
- [04:31]
Brian
it works for you. I did not think until, like, a month ago that I could wield a hammer.
- [04:36]
Chris Messina
Correct.
- [04:37]
Brian
And now I feel like the pilling has been. I mean, not anybody can wield a hammer, but if you're willing, I mean,
- [04:44]
Chris Messina
you're not gonna, like, build a house, but you can probably start with a shed.
- [04:47]
Brian
So. Okay, real quick.
- [04:48]
Chris Messina
As.
- [04:48]
Brian
As quick as I can. Because you said that of course, everybody got transformed, even the people that are clued in. And around December, and when I was at CES in January, like, this was the heat of that moment, and I started screwing around with downloading models on my laptop and seeing what I can do locally and stuff like that. And that led to even that.
- [05:11]
Chris Messina
When you say downloading models, what were you using for that? Because even the gateway, you can't just download a model and something happens.
- [05:18]
Brian
By the way, I feel like maybe we'll do episodes on this sort of stuff, but things like Comfy AI or LLM. What do you call it?
- [05:26]
Chris Messina
LLM Studio.
- [05:27]
Brian
Yeah, LLM Studio. And just seeing what I could do with it. And that led to. The first thing that I realized I got good at was actually prompting. Because hacking away at stuff, just downloading a model and not knowing what to do with it. Like, I started to realize that all the prompting that I had done for other things had been, like, be as descriptive as possible and blah, blah, blah, blah, blah, blah. And that's actually not, for most things, the right way to do it.
- [05:59]
Chris Messina
Yeah. What did you change?
- [06:01]
Brian
I learned about. Well, when understanding how memory works in AI, so then using markdown files as reference for creating schematics for the project that you want to do, that's the biggest unlock. Because then the other thing is that then you're constantly referencing back and forth to. And you can. It's harnessing essentially in the most basic way. So I did little things like, you know, I said this to the other day that the dumb thing about this will be for the audience. Here's something tangible at the end of the show. The very last thing that I do because of the way publishing the show works, is create the links for the show notes, which is a dumb. I mean, it's dumb, but you have to go get the link cut and paste it in. Go get the link cut and paste it in. Go get the link cut and paste it in. And, you know, at the end of a show, like, I'm kind of cognitively and emotionally burned. And so, like, even though that's only 10 minutes or really five minutes or whatever, like, taking that off my plate, that was amazing. So, like, I created a script and a whole system for doing that, where every day now I just go into Claude and I say, show notes. It looks at my script that I did that day that's sitting in my Google Drive file and does it all automatically. Just a dumb, dumb little thing. Right.
- [07:23]
Chris Messina
So you basically have given Claude access to Google Drive and then you've created like a skill. Is it a skill?
- [07:29]
Brian
I mean, I've done it several different ways. And like, we, we can get into like the, the back and forth of. Or the differences in that. And again, that could be a whole nother episode. Just that.
- [07:38]
Chris Messina
Well, I guess, like, I think some of these details are important because they help to set up whether you want to think about it as scaffolding or essentially rungs in a ladder.
- [07:46]
Brian
Right.
- [07:47]
Chris Messina
Because I. Yeah, you know, we can, we can jump around. And I feel like one of the things that has also been like an unlock for me is to learn some of the basics in different contexts and then to apply them to other things that I'm trying to do. And so, for example, you know, maybe I would have started with like a basic skill to like, write the show notes, you know, for after a podcast is done. But you'd also want to get like, the transcripts. You also want to maybe like, have some key, you know, clips or like, you know, topics or, you know, like, posts that come out of that. And so you can create a number of skills and then you can apply that in another context.
- [08:22]
Brian
And if you're. Again, this is. I've been using Claude mostly, but so like, if you do the Claude work cowork bit like, like, so then you can set up entire systems around your files. Systems and stuff like that. Okay, let me, let me jump to the resume writing thing. So again, yeah, resumewriters.com was the first company I ever founded when I was 19. I actually still own it. It's 25, 26 years old. At its height, we had 90 writers around the world that were contracted with us. We have probably serviced more than 200,000 job seekers over those years.
- [09:00]
Chris Messina
And explain the products and the offering. Right, because essentially you can just get a resume and do it yourself or you can Download something from LinkedIn. But what's special about what you are?
- [09:09]
Brian
There's an entire, or there has been, until recent years, an entire industry of people that are certified resume writers. There's certification agencies and things like that. It's like a real sort of specialized writing profession. And I've always enjoyed the fact that I've made most of my money helping people find jobs and improve their lives, but then also helping writers make a living full time as writers. So essentially what resumewriters.com has always been is, hey, come fill out a Form in this industry, you're looking for a job, maybe you have this job opening that you're specifically fine. We ingest that. We have all these writers that specialize in various industries. Essentially, we're a sorting mechanism. And then the client works one on one with the writer. And then we also monitor to make sure that the customer service happens, that things get delivered. All that stuff, like I said, been a good company for 25 years, 26 years. But as soon as the AI stuff started, I was aware that, well, this is coming right for that. Right. And the first couple years try. So I've always screwed around with that. And the first couple years, it wasn't anywhere close to being able to do competent things. And then after January, after ces, as I'm screwing around with stuff, I got incentivized again to, like, play around. And, oh, yeah, it has gotten a lot better with a couple of caveats. So, yes, you could go to ChatGPT or Cloud right now, paste in a job description. You can even add your existing resume and say, you know, do this for me, which is what I always wanted. Which, by the way, side note, I have for three years been engaging various software developers to try to. That was my dream is. I want. I want to. I want to be able to throw a job description at the AI and tweak the. So that instead of having one resume that's one size fits all, you could have, like, customized a unique, customized resume for every job application that you do. And I've paid people here and there. It's never quite worked out. It's never been as good as I wanted it to do. And then. So when I'm screwing around with Claude now, so this would be January into February. It's good, but not good in the way that I want it because, I mean, obviously Claude is getting better and better. AI is getting better and better.
- [11:31]
Chris Messina
Which models were you using at that time?
- [11:33]
Brian
All of them. I've done Gemini. I've done Gemini, for me, is like.
- [11:37]
Chris Messina
Like Opus or like Sonnet?
- [11:38]
Brian
Oh, I always did Opus. Yeah, I've been paying the. Yeah, yeah. There's two things that I didn't like about it is that even though the AI is getting better, it still is a generic tool. It's not a specialized tool. And then also because one thing you got to keep in mind is I, by running this company for the first couple years, I wrote every resume. So I must have written over a thousand resumes the first couple years of the company. And then there were many years when I was managing it and managing the writers and stuff like that. So I have seen through my hands what a good resume is to the tune of tens of thousands of times. Right. So I have an opinion on this that is very specific. And so, A, I didn't like the output of the resume and B, as I'm learning to prompt, okay, after several iterations and again making markdown files and tweaking and tweaking and tweaking, I'm getting the result that I want, that I feel like is a well written resume, not a generically written clawed resume. It's a resume that I feel like a human from resume writers would create. Right. So now I have that. It's a skill. But I, you know, I. How do I go out and sell that? Well, oh, you'd have to create a website and a program and all that stuff. Okay. What actually triggered me into this was asking the users for their projects.
- [13:05]
Chris Messina
Listeners.
- [13:06]
Brian
Yeah, listeners. Sorry, Listen, I've been talking with Cloud so much about users getting listeners to say, well, what did you do? What did you use it for? And so many of the answers were, I didn't think I could code anything. But I just kept trying to go as far as I could until I couldn't go any further and I never hit the wall. And so then that made me think, well, all right, if this is the thing, if I'm sitting here with what I believe is a better way to use AI to write a resume, because I have prompt engineered it to be what I think is quality. And I also want to deliver it in a way that. Because again, the wrestling of the prompt engineer wrestling with Claude back and forth to get it to do what you want is so annoying. So what if I created a system that eliminated all that you and I have always talked about knobs and dials as opposed to just brute force talking into the chat, typing into the command line. And so I sat down one morning, it was a Thursday morning at 7:30 in the morning. And two and a half hours later I went back and forth with Claude. And Claude was like, all right, let's do this. Now I want to come back to that because I want to tell you what I learned.
- [14:19]
Chris Messina
What happened in the two and a half hours.
- [14:21]
Brian
The first thing I said was, hey, I have the domain resumewriting.com. i have this idea for how to do an AI resume writing product. Scope it out for me. What would it take? What would I need to set up? So it's saying, well, you'd need to get an account on Vercel and Supabase, and you'd need to be. I recommend we do it in all the things I keep saying to Claude, like, okay, that's great. I don't know what any of that means, but do you think that you can work with me to do that? Claude, of course, is always like, of course. And by the way, you're a genius and very good looking. So then by like 10 o' clock that morning, I'm like, all right, I'll give this a go. And I have. The first thing that Claude does is come up with a spec. Like a long detailed. Like if you printed it out, it's like at least a 30 to 40 page spec. Right. Sits in the cloud memory on a project. So it's a specific project separate from all your other conversations.
- [15:21]
Chris Messina
Sorry, is this in Cowork or where is this where you work?
- [15:25]
Brian
It sits. It kind of straddles both places. All right. For reasons.
- [15:30]
Chris Messina
Because projects are a little bit weird with Cowork versus just Claude. Because you can have kind of like a folder and you can have some instructions in the folder. Cowork allows you to have access to like the file system.
- [15:41]
Brian
Yes, actually Projects does too. Well, no, Projects does if you're through your. I'm coding in Terminal like a barbarian. But wait, what do you mean?
- [15:53]
Chris Messina
Are you using Claude code?
- [15:55]
Brian
Yeah, yeah, yeah, yeah. But like on, on the terminal, like you can install Claude code in the terminal. So what I'm saying is, because I'm using Terminal, right? Terminal has access to my file system. So that's not a problem. I see what you're saying, that in, in a project, in Claude chat.
- [16:12]
Chris Messina
Yes.
- [16:12]
Brian
You wouldn't have access to the file system.
- [16:14]
Chris Messina
Okay, let's step back because my confusion might be other people's confusion, or other people might just be using Claude in
- [16:20]
Brian
a browser or I'm using it as a dumb person, which is very likely.
- [16:25]
Chris Messina
The fact that you even are in a terminal sort of, I think, sets you aside for many people. Like, many people accidentally open Terminal and instantly freak out because like, you know, black background, white or green text, they're like, this is the matrix. This is the. I'm going to hurt my computer. I need to like, run away. So you decided to say, I'm willing to hurt my computer. And so you, you're in Terminal, you installed Claude code.
- [16:48]
Brian
Okay, let me caveat this. Listen, I'm. I'm an old enough person that I remember hacking around in DOS to get the illegally.
- [16:57]
Chris Messina
So you're not Command line illegally acquired
- [17:00]
Brian
video games onto My computer. I'm not afraid of that.
- [17:03]
Chris Messina
Okay.
- [17:04]
Brian
Yeah.
- [17:04]
Chris Messina
So I'm just again, delineating the differences between perhaps you and perhaps, you know, a certain type of average listener versus a listener that maybe is like, super smart and awesome, but has never ventured into the terminal. So we're now placing you in the terminal. I presume that you are in some directory that's on your hard drive and from in that directory, which let's call it resume writing.
- [17:25]
Brian
Yeah. Is my.
- [17:26]
Chris Messina
It's my repo Claude code.
- [17:28]
Brian
Yes.
- [17:29]
Chris Messina
And. Sorry, you said repo. Do you. Do you literally mean that you're using like, Git and GitHub for this?
- [17:33]
Brian
Oh, yeah.
- [17:33]
Chris Messina
Okay.
- [17:34]
Brian
All of these things. Remember how I.
- [17:36]
Chris Messina
Did you know these things before or was this new?
- [17:38]
Brian
Okay. This is also why I'm different. I know all of these terms. I have done startups. I have worked with software developers for 25 years, so I am familiar with these. That's an interesting thing that we should circle back to is having a greater understanding for software developers versus, like, I've always been the person on the other side of the table being like, I don't care your excuses. We need this done in two weeks.
- [18:00]
Chris Messina
Yeah, see, I think this is why we're both, like, dangerous and also, like, I'm trying to navigate how relatable we are because we have had all this exposure and I've been on developer relations teams even when I couldn't write any code, which, you know, obviously somehow I must have had a good resume. But my point is you still have a vernacular that separates you from many, even if you don't know about how to figure out Supabase and Vercel. And like these new Johnny Completely's, I do understand the components, navigate the command line, you can move between the directory and folder structure. You can understand how Git works.
- [18:37]
Brian
I do want to come back to that because I have a pushback on that in a second. Long story short, here's key lesson number one. Any project that you want to do, Talk back and forth with the thing, push back. Don't just say, write something in and be like, all right, make this for me. Go back and forth collaboratively with your AI and whittle down the scope of what it is that you want to
- [19:08]
Chris Messina
even a more specific thing. And this was mentioned at Code with Claude, which is the developer event I just went to in San Francisco. The Claude team says, have Claude interview you. So essentially it's like, here's the thing that I'm trying to build and here's Roughly what I think I want it to do, interview me about all the different details and I will walk through them and then sometimes you'll run up against something like Supabase or whatever. You're like, I don't know, I'll go with whatever you want.
- [19:37]
Brian
Let's skip ahead to that part because. Okay, so I'm starting to do this. I'm getting all the accounts, but even setting up a database or setting up, it'll assume that I know things like, well, now you're going to want to configure this thing or that thing every single time. And I know that this is brute forcing it and that I'm burning tokens that you wouldn't have to burn, but I'm just, I'm just screenshotting and screenshotting and screenshotting. So in a single session of working with Claude, I could have 300 screenshots down to the fine grain of like,
- [20:09]
Chris Messina
wait, what is the purpose of these screenshots?
- [20:11]
Brian
I'm like, I don't know where you told me to click to, like you're taking screenshots of.
- [20:19]
Chris Messina
So, okay, basically this.
- [20:20]
Brian
Imagine I'm trying to.
- [20:21]
Chris Messina
Slowest form of like computer use.
- [20:22]
Brian
Yes.
- [20:22]
Chris Messina
So you could just have Claude control your computer, but instead that's maybe too hard or something. And so you're like, here, Claude, I'm looking at this thing, what should I click?
- [20:31]
Brian
Okay, you know what this is? I want to see if this is what you're. Because I haven't made the leap to have IT control, like my computer, right?
- [20:38]
Chris Messina
Yes. Yeah.
- [20:39]
Brian
But again, for someone that has, let's say, less knowledge of things like databases and things like that than we do,
- [20:48]
Chris Messina
are you finding that you're learning as you go through this, or are you just kind of like tossing it over to Claude and being like, claude, tell me what to push and I'll push it.
- [20:54]
Brian
And you're like, I mean, what's that
- [20:56]
Chris Messina
could be in Japanese?
- [20:57]
Brian
I am, I am learning, but I kind of don't want to learn. The thing that I hate more than anything is smoke testing. So, like, after Claude runs something, it's like, hey, can you go back, Brian, and double check if this is actually built correctly? And I hate that shit. So the point I'm making is we're talking about creating a software product or a website, right?
- [21:22]
Chris Messina
Well, specifically a web app, right?
- [21:24]
Brian
And I'm saying I got through some difficult stuff by just taking screenshots. Screenshot, screenshots. Being like, you told me that I can click on the left sidebar and configure. I don't know what you're talking about or if it's like, I want you to do X, Y and Z. I'm like, and specifically tell me what to paste into the thing. Okay, so we're talking about software right now, but I'm saying that you could use this in any way that it's something that you feel like is beyond your Ken. And like, Lisa has done it recently for things involving the kids education, for getting IEPs done and stuff like that, where it's not just asking, like, what do you think these test results mean? Okay, but you need to apply for this school. Okay, how do I do it? Okay, where like, so again, I've been like, screenshot, screenshot, screenshot. Once Claude has an understanding or your AI has an understanding of what you want it to do, if you get lost any step of the way, just force it to guide you.
- [22:22]
Chris Messina
Yeah, I mean, the multimodality, I think, is really incredible. And another example of this, actually, just last night I was working on a project and I was arguing with Claude about how to handle authentication. You know, and I've. I've worked on authentication for 15 years. And it's like, I really, I want to do this this other way. And Claude's like, no, you should really do it this other way. And so I had to go into Sketch and I actually created some UI mockups to say, actually this is how I want it to flow. And I was able to essentially make my case better. Now, granted, you know, Claude's going to be sycophantic and, you know, nice, nice design, whatever. But nonetheless, once I was able to actually show what I wanted, then that was another way to actually advance the conversation as opposed to staying in, like, verbalese and even in cloud code. It'll create diagrams and all sort of ASCII art and things like that. And it's close, but it's never quite.
- [23:10]
Brian
Well, by the way, I also got the timing really right because if you use Claude design at all yet study and play come together on a Windows 11 PC. And for a limited time, college students get the best of both worlds. Get the unreal college deal. Everything you need. Study and play with select Windows 11 PCs. Eligible students get a year of Microsoft 365 Premium and a year of Xbox game. Pass ultimate with a custom color Xbox wireless controller. Learn more@windows.com studentoffer while supplies last ends June 30 terms@akams.collegepc with everything you balance at work wouldn't it be nice to have your own personal chief of staff. Something that can capture meetings, calls in person and online conversations without forcing anyone to use their phone or break eye contact. Well that's a exactly what PLAUD is a dedicated AI note taking device designed to capture in person conversations so you can stay in the moment. PLOD devices are small, secure and can automatically deliver accurate transcripts, clear summaries with action items right in your phone. And with over 20 hours of continuous recording, you can be more present without losing key details or phone battery life. For a limited time, you can get 10% off any plaud device when you check out with Code tbrh. Visit Plaud AI to learn more. That's P L A U D A I TBRH Code TBRH.
- [24:44]
Chris Messina
Ready to soundtrack your summer with Red
- [24:46]
Brian
Bull Summer All Day Play.
- [24:47]
Chris Messina
You choose a playlist that fits your
- [24:49]
Brian
summer vibe the best.
- [24:50]
Chris Messina
Are you a festival fanatic, a deep
- [24:52]
Brian
end dj, a road dog, or a trail mixer?
- [24:56]
Chris Messina
Just add a song to your chosen playlist and put your summer on track.
- [25:00]
Brian
Red Bull Summer All Day Play. Red Bull gives you wings. Visit Red Bull.com BrightSummerAhead to learn more.
- [25:07]
Chris Messina
See you this summer.
- [25:10]
Brian
My God, that has been like a leapt over stuff that would have taken a lot of. A lot of time. Okay, let me try to get this out as quickly as possible. So here's the bottom line. I'm not there yet. Actually, by the time hopefully this episode comes out, it will be launched and final whatever. Like I had a.
- [25:29]
Chris Messina
What's keeping you from launching it?
- [25:30]
Brian
I'm not done yet. I actually had to.
- [25:32]
Chris Messina
What does done mean?
- [25:34]
Brian
It's not exactly usable yet. I had to revert like a good two days of work this morning because I realized I had gone down a cul de sac of design that.
- [25:45]
Chris Messina
So let me actually this is another important point and I think this is actually so like given how you and I came up and how expensive things were to do, right. You'd spend, I don't know, $200,000 on a software project or something. You'd be like, okay, it's done. Don't touch. You know, I spent. I mean this will put it into some perspective. I built the first functioning version of this like identity verification app on a train on my way to Vancouver to the App Proto Blue Sky Developer Conference. So like in 45 minutes I built like the first version of this app, just like vibe coding and it worked. And that made a number of assumptions about how it would work. And so I went down a path and I thought that was going to be good. And in recent weeks I've expanded what I'm trying to do and what I've come to the realization of is that it's just better to just start actually from scratch, like completely start over and start with a new architecture and throw away v1 and you'll actually get much further, faster. And so whatever, you know, I don't know what exactly you're describing, but if you're like, oh, I need to do some refactoring, it's actually possible now to be like, maybe not refactor, but actually rebuild, tear it down and start over.
- [26:56]
Brian
Now this gets to you saying, are you learning as you're going? One of the things that I wanted to do was I got it to work. Essentially you upload your resume to my thing and then you cut and paste the job description and it tweaks. I'm going to show you a screenshot in a second. But what was happening was it would take a good 90 seconds to work, which we're used to that with AI. But if this is a consumer facing product, where I'm saying, hey, normal people, I'm trying to take the scary or the, the complication out of AI, I wanted to get that down and Claude claimed they could do it. And like I said, I'm talking about this is maybe 10 hours of consistent coding work. We got it down, but it kept breaking, breaking, breaking to the point where I realized, hey, remember how we had a functional system that you could cut and paste? It may have taken 90 seconds to work, which sucked, but let's get back to that. But now it doesn't work at all. So I literally said, stop, Claude. When was the last time this actually worked? And it gave me an answer. I was like, let's revert to that point. And it pushed back on me. I was like, motherfucker, we're doing this. And Claude was like, well, but look at all the sunk costs that you put into it. I was like, there's no point of a sunk cost of something that's non functional. Okay? So I want to bring what the product is to an end because I really hope people don't think that I'm trying to promote this as a product. By the way, I'm going to, I have a separate project where Claude is going to help me with the entire marketing campaign doing, of course, launching with Google Ads. And then. And again I did the same thing, like, here's what we want to achieve, here's the CAC that I want to achieve. Here's the blah, blah, blah. So create a schematic.
- [28:43]
Chris Messina
And so before you go ahead though, in terms of going back, was Claude checking in its progress as it was proceeding using git. Okay, yeah, yeah, yeah.
- [28:53]
Brian
Oh, yeah, no, I've been trying to. Claude is forcing me to do actual
- [28:58]
Chris Messina
better software engineering techniques, legitimate software engineering
- [29:01]
Brian
techniques where there are specific sprints. There are. Right. And so like there's. I understand now the difference between a commit and a push and things like that. Yeah, or what did I say the other day? I understand what environment. What is it?
- [29:17]
Chris Messina
What's environmental variables.
- [29:18]
Brian
Yeah, environment variables. I understand what those mean. Okay, so here's where we are right now. And like I said, by the time I post this episode, hopefully like anyone will be able to go to resumewriting.com and see it. But I'm going to share the screen with you right now because I want to show you this is not just a webpage. This is something that, I mean, it's not the most complicated thing in the world, but I didn't think I would.
- [29:42]
Chris Messina
Well, it's your baby, so of course it's going to be beautiful.
- [29:44]
Brian
Okay, so look, so this is after someone has ingested their resume, right. And then put in also a job description. And there are legitimate knobs and dials here where. Here are some blocking things that I need you to clarify before.
- [30:00]
Chris Messina
Let me explain this to the listeners, what we're seeing. Right, so this is resumewriting.com, it's actually quite like a, like a dark theme. And I don't know if that's specifically a choice.
- [30:09]
Brian
Yeah, I might go back on that too. I might revert that. But when you're logged in, it's dark. When you're not logged out and it's like anyway.
- [30:16]
Chris Messina
Yeah, okay, well, light and dark mode, that's anyways. And so on the left hand side we've got a couple menu options, tailor a job, tailorings my resume templates, preferences and account. And then on the right hand side, we're seeing a number of cards in an interface. And at the very top there's effectively what looks to be sort of a blocking issue that needs to be resolved that provides you a couple different options to take action on. I presume the suggestions that resume writing has provided. And then down below is the resume itself. And then a number of things that have been changed by the AI.
- [30:50]
Brian
Right, so now that blocking thing, why is that there? Because I kept running into problems where it's like, well, if the person hasn't given me enough information, then I'm just going to return gobbledygook anyway.
- [31:00]
Chris Messina
So this is a validation step.
- [31:02]
Brian
Validation. Then what you can also see is there is an actual image of the resume generated with the actual resume language in it, which again, I can do things now like on the fly, create PDFs and Word docx files in design and stuff like that. But here's the key part. This is what I love. So as opposed to it just being a chatbot where you have to wrestle with it, what I have over here is what the AI changed. A summary of what they changed.
- [31:35]
Chris Messina
Sorry. So what they changed is the delta between what someone uploaded and then what
- [31:39]
Brian
is shown here, someone's existing resume and then what they changed to tailor it to that specific new job. Okay. Then the situation read, which is like, here's what I think you were going to want one to two pages. You're trying to make a lateral move. So this is the AI telling you what it thinks you're trying to do. Right? Including goals. Now this is where it gets interesting. So this is where my prompt engineering is like, okay, it's weak right now because of these things. So confirm yes or no.
- [32:09]
Chris Messina
Okay, sorry. So for a listener, basically there are a number of either assumptions or kind of not quibbles, but like, in order to make the resume stronger, resume writing, I'll give you. I'll give you an exact a number of things and you can confirm yes or no, which will then change the shape of the resume. So, yeah, give us an example.
- [32:32]
Brian
This job is in Rosslyn, New York, but the resume was for someone living in Portland, Oregon. So it says, are you willing to signal that you're open to move for this position? Yes or no. Now, when you click yes or no, that goes into the prompt cache. And so now when we're going to do another iteration that's going to affect what happens. There's also a scoring for. It says 62 out of 100. And it gives you the reasons why the score is good or bad. I don't know that I love that I'm still.
- [33:04]
Chris Messina
That specifically is an ATS word match. And so essentially for automated systems, they're looking for specific keywords. And so if you want to basically put your resume into the meat grinder of atss, then you need to match on certain keywords in order to be flagged.
- [33:22]
Brian
Yeah, either. I bet I don't launch with that. I mean, I'll save it and then like launch it.
- [33:27]
Chris Messina
Post, you know, I mean, I suppose this is a completely different conversation or topic, but like from a pricing perspective, those are the types of things that could be, you know, upgradable.
- [33:36]
Brian
Oh, hey, that's part of. That's part of Claude's whole spec for this, man. We're going to.
- [33:40]
Chris Messina
I see.
- [33:40]
Brian
We're just launching the mvp and then we'll have like, we'll add on cover
- [33:44]
Chris Messina
letters to make this free. Or is it.
- [33:46]
Brian
Oh, no, no, no. If you look on the screen, 50 credits for $24.95 a month. Yep.
- [33:53]
Chris Messina
So it's like 50 cents for each credit.
- [33:56]
Brian
Yes, essentially. So, okay, last thing.
- [33:59]
Chris Messina
Do one resume is how many credits or.
- [34:01]
Brian
Yeah, okay, wait, I'll explain that.
- [34:04]
Chris Messina
Sorry. Okay.
- [34:04]
Brian
Ok. Thing is questions and gaps. So I've identified the AI or I. Again, because this, this is my prompt engineering that spent weeks and weeks figuring this out. These are the things that I'm still missing to make a good resume in the right way. Right. Which Claude is not going to necessarily catch because again, Claude is the one that has generated this first version of the resume. And I am saying, yeah, but it is, it is a blunt instrument. And here's how you can get it better. So there's clarifying things, eight different ones of them. There's specific questions. If you click into it, you can answer that question again. It gets cached into the next prompt iteration. So then final block here is, would you. Is there anything else you would like me to change before we do another iteration? Okay, so now if you hit retailer, it will now take everything that you just answered and do another run at it. And then I've even. Although this isn't live yet, you can come up here and show the changes, show what was rewritten, show what was prompted, and on the second iteration, it'll tell you again the changes that it made. So again, now, this is a glorified wrapper around AI. I get that. But it's also trying to guide people. It's number one. It's my take. You know, there's those startups out there that are like, hey, talk to this scientist or that therapist, because we've trained it on their writing or their philosophy or whatever. And so when you're talking to them, in theory, you're getting their expertise via AI. Okay, this is my expertise on what I think makes a good resume.
- [35:39]
Chris Messina
Well, you've productized your expertise through an interface.
- [35:43]
Brian
Exactly. And then the second part of the interface is, you know, again, someone that's not good at wrestling with the prompting and with the chatbots and things like that. I'm guiding you through the messy Steps to get a better result than just saying, hey, can you just make this better?
- [36:04]
Chris Messina
Right. Yeah, well, you are providing substance to the question of better.
- [36:13]
Brian
All right, I've stopped sharing the screen. Right? Good.
- [36:14]
Chris Messina
Yes.
- [36:15]
Brian
All right, so everything that I just described to you is something that I thought would be impossible for someone like me, right? And I have paid people tens of thousands of dollars to do. And no offense, I'm not saying these people were bad, but they weren't able to achieve what I wanted. That's another thing that I've learned is that again, as opposed to saying to somebody, hey, in your side opinion, your side free time, I'll give you $10,000 if you try to hack this together for me. And then they bring it back to me and I'm like, it's no good or whatever, but by me doing the actual steps myself, I'm able to make these pivots where it's like, hey, that's not working the way that I want it. As opposed to them delivering the product and me being like, yeah, it's not quite there, but not knowing how to tell them how to do it, you
- [37:02]
Chris Messina
know, I mean, I think it's a really important. The point, right? Because, you know, just throwing $10,000, you know, at a software project without a lot of direction and, like, fine grain controlled and like, writing the PRDs and like, getting super in the weeds, right, Is probably going to produce something, you know, that's like, pretty milquetoast and that's similar to how 100% used to be with prompting, right? You'd be like, oh, you know, make my resume better. And it's like, actually there's a lot of detail in each of these things and that you've honed over, you know, building an entire business around this.
- [37:34]
Brian
All right, well, think about. Actually, the interesting part is because I have this expertise in what makes a good resume, I was far away from the metal that would make the good product because I had to go to, like, I can't expect a software developer to know what a good resume is, even if I try. I write out tons of descriptions and. And have interviews and here's samples and stuff like that. I had the expertise, but I was several layers away from it because I couldn't do the coding well, I mean,
- [38:09]
Chris Messina
even putting a sort of different way, right, like the service that you provided was to write resumes for people that don't know the ins and outs of how to write a good resume for whatever reason. I mean, it's not something. I mean, Maybe you'll do a resume in high school or something, but it's not whatever. Maybe you find, you Google it and you get some tips. But it's like to your point, a lot of these super fine details, you know, like even when I look at some of my friends resumes and I look at like the job description, I'm like, it says right here, there's like a few things in here that are nowhere in your resume like what's going on. Right. Even that from just a human to human perspective is valuable.
- [38:43]
Brian
Listen, we've over the years written resumes for and I'm including Silicon Valley Amongst this CEOs whose names you know because they're even if you are the smartest, you know, richest, highest level person, like just getting a second opinion on what is and is not important in your career, that's half of what a resume writing.
- [39:03]
Chris Messina
But I guess this is what I'm pointing to though is like there's a question about what the software was meant to do when you hired these outside people. Right. Was just about, you know, making the flow better. Was it about scaling the process?
- [39:15]
Brian
Originally, like I said, I just wanted to be able to have a different resume for every job description because three years ago that didn't exist. That was an impossible thing to do. And what resumewriters.com did for people for 26 years was create a one size fits all resume as best we could.
- [39:32]
Chris Messina
I see. And so what you were hoping to do with AI was essentially kind of go deeper into personalizing and customizing each resume for each job description that would be done at a scale that would make sense for humans to do.
- [39:46]
Brian
Every time you apply to a new job, it should be tweaked to that specific job description. Yeah, right, yeah.
- [39:53]
Chris Messina
And so now the software both allows that, but then also puts more control in the individual's hands. You know, you can also get quite detailed in terms of the feedback that you're getting either on the prose or on the structure of the resume that is guiding the human and where the resume writer doesn't have to like go through and interview the person a million times. Right. You know, and I guess like maybe there's a meta commentary here which is just about the job market itself and how much more either specific it's gotten or more competitive. At least, you know, amongst my friends it seems to be much more competitive. And so therefore you have to tailor all of your submissions, you know, to the job description or else you're not even going to be considered. And so because of that that creates a different pressure on your business that you can't really have like a one size fits all resume and hope to do as well as before.
- [40:41]
Brian
Okay, I know that you're a product guy, but let's pull back from the product because again, I want to come back. I mean, if it looks like to you this is something that is kind of useful, then I'm happy because, okay, I could not have achieved a dream that I could achieve. So once you're done, you can change Designs. There's only 10 of them right now, but I'm going to add more and that'll be another upgrade and all that stuff. And then you can get it as a PDF or you can get it as a. The fact that I could code up a system that can create on the fly different designs in PDF or Docx files based on the text that also my program generated, I would never have dreamed that that was something that. I believe I could create a more complicated web app, but it would be Surface. Right.
- [41:39]
Chris Messina
This is an entirely different workflow.
- [41:41]
Brian
Yeah, this is deeper where we're producing things that are on a different so 1.
- [41:47]
Chris Messina
1. I think this is an important question because, like, when you have a business, you know, doing this, you know, whether it's a side hustle or. Or, you know, it's meaningful income, when it comes to deploying this, I think there's going to be, you know, a question where the rubber meets the road and then you have like real user issues and challenges. And so I guess, like, my. And we don't really, you know, know this for you yet, but the question really is, can you go from this, like, Vibe coded product, which I believe is net new. Right. You're not like taking an existing product and like updating or modernizing it.
- [42:16]
Brian
Yeah, totally.
- [42:17]
Chris Messina
When there are issues or bugs that come up, is it a matter of just kind of like shooting it over to Cloud and being like, hey, there's reporting this, like, how are you going to handle testing?
- [42:24]
Brian
There are things like Sentry and stuff like that and observability, you know, like, so I can get reports when like the thing.
- [42:32]
Chris Messina
Like, like we invested in companies that actually, like, build these products.
- [42:36]
Brian
But Chris, I had a conceptual understanding of what those things do. I didn't have a working.
- [42:42]
Chris Messina
I mean, I understand. Yeah, obviously once you start to touch it, you start to realize what this is and it demystifies it, like how it works.
- [42:50]
Brian
So this could be very wrong. But, like, I have also set up a separate project where it's like, Claude, I want to get a morning report every day. I want to know, like, I need the century reports for when, you know, people get stuck and they can't upload their resumes or like, the iteration on the resume breaks after try number three or something. I need all of that stuff.
- [43:12]
Chris Messina
Did you have Sentry for the existing platform?
- [43:15]
Brian
No. Oh, all those things are new.
- [43:17]
Chris Messina
Yep. So are there ever. There's no bugs in the current product because you're not looking for them.
- [43:23]
Brian
Are you talking about resume writers? The human product?
- [43:25]
Chris Messina
Yes, the existing.
- [43:26]
Brian
There are humans there because I have managers who manage the writers. And so like.
- [43:31]
Chris Messina
Yeah, but that's the human side. I'm asking if technically something fails every
- [43:35]
Brian
couple years, resume writers goes down in some funky way and it ruins my day because I have to be like, oh, God, I gotta fix this somehow.
- [43:42]
Chris Messina
So it could have been broken for many, many months before that, but then it just. At some point, it just craps up.
- [43:47]
Brian
Listen, the very first resume writer's website, I bought the software package Front Page produced by Microsoft in 1999.
- [43:56]
Chris Messina
I was a dreamweaver kid.
- [43:57]
Brian
Okay, so look, it's probably really not gotten much more sophisticated than that in 25 years.
- [44:04]
Chris Messina
Oh, my God. It's probably still running front page on, like, Windows.
- [44:07]
Brian
No, it's not. It's still running front page. But, yeah, long story short, it's all running on Shopify now and shit like that.
- [44:13]
Chris Messina
I see.
- [44:13]
Brian
Okay, okay. Back to the idea of. And I keep running them problems, but I keep fixing them. So what I'm saying is. So I said to my wife, your office, they're going to come to you, the industry you work in. They might be behind the eight ball by 18 months, but someday they're gonna get the religion and they're gonna come to you and they're gonna be like, we're gonna change how the office operates in this, that, and the other way. I saw the news that now Claude plugs into AutoCAD.
- [44:48]
Chris Messina
Oh, yeah, yeah, yeah.
- [44:50]
Brian
So I said. And so imagine I'm saying this to my wife, but I'm saying this to any listener out there. Okay? You can wait for that to happen for 18 months, or you could right now start to tinker with this stuff. And again, before they say, we need to revamp this entire office and how it operates using AI, what if you did that just for yourself? And then you could keep it a secret from your boss or whatever, but, like, just find a way to make your process again. What is the ideal software for the job that you do? That has never been Made because it's not made up.
- [45:28]
Chris Messina
There's an even simpler way to approach this, which literally is to ask Claude or your preferred AI provider to interview about your job.
- [45:37]
Brian
Where are the pain points? What could be better faster of what
- [45:42]
Chris Messina
are repetitive things that happen? And does it make sense to either build a little mini app, does it make sense to build a script, does it make sense to turn on some
- [45:51]
Brian
sort of automation right now? The benefits of that are maybe you succeed enough that you get a little more efficient and you write the show notes links automatically and save yourself 10 minutes every day. Better case scenario, you kind of like improve your processes for yourself. Next best case scenario, you take this to your boss and you say, look at what I did. And so you're the one that six months early is like, hey, do you realize what we can do here? As opposed to waiting for it to happen. So there's lots of, again, if you just try it and just see how far you can get before you get blocked, there's no real downside to it. And if the worst case scenario is you learn how to prompt better, then the next time you have a vacation,
- [46:35]
Chris Messina
the meme going around is like, you know, AI is not going to take your job. It's going to be someone who knows how to use AI. And especially in these vertical contexts. I think that's, that's true. And so you're like, the point that I was getting to before about your domain knowledge is effectively, now maybe in some ATS systems, the data exists that matches resumes to job descriptions. And that is amazing training data for someone that wants to build, you know, the replacement for resume writing. But, but the alternative is for you, Brian, to go and take all of your knowledge in a specific domain and say, actually, I've seen all these resumes and I've seen the outcomes and therefore I can, in conversation with Claude or my AI, come up with a set of heuristics that will walk people through my process. And they're essentially, I mean, you are right. So imagine allowing your knowledge to be essentially turned into a kind of raw material or gold and to be sold accordingly.
- [47:27]
Brian
There could be, there could be listeners out there that are life coaches or things like that. Like that could again, you could do something similar. And again, I have to credit the listeners with this because when I was specifically saying, if you're not software, you're not a software developer, but you have done something interesting, what was it? And it kept coming back and I can't use any of them now because I haven't gotten approval for anybody to talk about it or whatever. But it was like that. It's like, you know what, I do my job 30% better now because I fixed this problem that had always bothered me and it wasn't always suffer. Maybe it is just, you know, using Claude Cowork to, you know, manage your calendar better or something like that. Like I want to do segments on stuff, like if it's something that simple, like.
- [48:16]
Chris Messina
But I think I might have, I don't know if I told you about this experience that I had. I mean, because.
- [48:22]
Brian
Okay, let's get to you because we've talked about me long enough.
- [48:24]
Chris Messina
Yeah, well, I'll give you the. So. So I think there's, there's two thoughts here that I think are quite relevant. You know, at the beginning of the year I decided that, you know, for, for a lot of last year I was doing a lot of hourly coaching and I still do hourly coaching mostly on, on launch and go to market. And that is how I make basically all my income these days. And so, you know, what I was finding is that I was really reluctant to pare down the hours that I was available so you could kind of like book my time. Anyone from like 9am to like 5pm, like five, sometimes six days a week. I was just like, I never want to miss someone that wants to come and book time with me. And the result was that my calendar was super chaotic and it was like I never had time for focus work and things like that. The Wired newsroom is known for award winning reporting on how technology shapes our world. On WIRED's Uncanny Valley, we take that curiosity even further. Each week, journalists from Wired break down the biggest stories in tech while speaking directly with the people. Building challenging and reshaping the future. Is the AI boom sustainable? How do you protect your privacy in an age of constant surveillance? Uncanny Valley tackles the questions driving today's tech debates and lighting up your group chats.
- [49:39]
Brian
Listen to new episodes every Thursday wherever you get your podcasts.
- [49:44]
Chris Messina
So I decided coming into this year I wanted to take back my time. I wanted to spend more time able to work out. I wanted to spend more time actually doing these vibe Cody projects. And for that you actually need some, you know, chunky time. And so I worked with at the time ChatGPT to basically take my entire year of bookings from last year. I had it build a, you know, heuristic and a visual that showed me a heat map of what times of which days were the most essentially lucrative for me or the most productive and then I strip back all of my availability to just those times. And I said, we'll run Q1 with that availability and we'll see how that impacts my income. And it turned out that it didn't change it at all. I cut back my hours by probably 70%. So the time that I make the
- [50:29]
Brian
same amount of money, same amount of
- [50:31]
Chris Messina
money, but I have 70% more time to work on things that I want to work on as a result of this.
- [50:36]
Brian
Let me stop you there, because again, I've been talking about software. I've been talking. Yes, I have a job. I have the domain name resumewriting.com, right? Okay. I have this 25 years of a specific set of skills, to quote Liam Neeson. But what you just described, if you're a salesperson, if you're a therapist or whatever, like that sort of ability to change just the way you work, even if you don't touch software at all. Sure, yeah.
- [51:06]
Chris Messina
So I think this leads to the second point, and I think this is actually gonna be something that I'm gonna challenge you to do after the show, which is so what I did. And because I had the data, right, because I had a CSV of all my bookings and I had all my stripe transactions and I was able to, like, look at the actual story of my business, ChatGPT was able to say, okay, based on what the data shows, here's how you could go forward with this. Similarly, for a lot of the vibe coding projects that I've done recently, and I think for you, this would be super interesting, would be to come up with some kind of self analysis prompt that essentially asks, imagine that you're my hiring manager or you're head of hr, or you're the CTO of Resume Writing Inc. And you see, how is it austrosperous? I don't know, essentially entrepreneurial. I've been in building this, this product, I want to be able to do more compelling software projects, do an analysis of our process. How could I improve? What would I do if I were actually like 10x better? And what are the questions that I didn't ask? Well, and how could I improve my process so that next time when we work on something, it goes better? And I've actually done this on a couple of my projects and it's really pointed out a lot of gaps in how I think about things. Or for example, sometimes like. And I think a lot of people have this experience with vibe coding where you maybe you just don't quite know what you're trying to do. And you're just like, go build something, show me something. And there's a lot of back and forth and a lot of iteration and it can be, you know, kind of a wasted time. As opposed to what Claude was telling me is like, actually, why don't you ask me again to do this interviewing you so we can get clearer on what your goals are. Why don't you set like, it was like, upload some examples of things that you like, show me some things that you don't like. And as a result, Claude's like, I will be 40% more efficient in this process just because of that one little tweak.
- [52:56]
Brian
And I'll give you an example. I won't go into the details now for the time, but I told you one time about a startup idea that I had and I was like, how come no one's ever done this and you weren't really hot on it? And I asked Claude about it after doing this, and Claude shot me down and was like, here, X, Y and Z. This is already served by these products. And also the reason these products do it but no one else has is because blah, blah, blah, blah. So like, you can, again, you're talking about like getting better at stuff, but it can also save you from mistakes.
- [53:26]
Chris Messina
Yeah, so the other one is, and I saw this going around recently, you can do pre mortems. So I'm using this actually on my current project. I'm like, claude, I want you to go six months in the future and presume that this project has failed. Go through all the reasons based on where we are now.
- [53:40]
Brian
Wow, that's cool.
- [53:41]
Chris Messina
Why it failed.
- [53:42]
Brian
That's cool.
- [53:43]
Chris Messina
Because you asked Clyde to like do these projections and things like that. Of course it's going to be super optimistic and you're going to be making $6 million a day. You should be like, I want you to go in and basically tear this down. The project failed. It's, you know.
- [53:55]
Brian
No, I don't know that it's ever done that for me, but I have had it do things like, you know, I need to. What's our margins like? And so around designing it, like based on how the prompt is entering.
- [54:05]
Chris Messina
What I'm saying is it's great to have that in the positive.
- [54:07]
Brian
Sure, sure, sure.
- [54:08]
Chris Messina
What I'm saying is now have the contrarian take right and act as if it has failed. Well, that's just to bolster your strategy.
- [54:16]
Brian
But what I'm pointing out is that any kind of project, like budgeting, you can save yourself from things because you can be like, hey, if I do this change, will it. How will that affect the economics of this? Right. And then, well, it would change it this way and we change it that way. And then it's also interesting. And again, I don't know if this was the thing that happened over the last six months, but it will proactively suggest things to you. Right. So it's not always. I mean, yes, it's still very like, hey, you're great and beautiful and smart, but at the same time it'll also say, yeah, I see what you're doing here, but have you thought of this? And again, it's not always the most brilliant idea. Have you thought of this? But sometimes it's like, oh, I hadn't thought of that. Let's explore that.
- [55:04]
Chris Messina
I think it's like, yeah, I think what you're pointing to is the way in which the models have really changed both to be more proactive and to be thought partners. Meaning that it's no longer the case where you just kind of like try to be super declarative in your prompt so you get like a very specific outcome. It's sort of like, you know, a very bad way to program, which is
- [55:21]
Brian
how I said how I was familiar with it a year ago. Yeah.
- [55:24]
Chris Messina
And that's what I'm saying is that the difference now is that it really is kind of a bi directional, meaning kind of builder, where you're working together to come up with kind of a plan or a way forward as opposed to thinking, you know, best.
- [55:36]
Brian
That's the other thing. I don't know that I should specify this, and this is specifically for coding, but I don't know, maybe this is applicable to other things. I am running Claude against himself because. Right. I have Claude open over here and then I have Claude in terminal over here. So Claude code will return something. I will paste that back into Claude and say, code said this. And CLAUDE will be like, yeah, I don't think they got that right.
- [55:59]
Chris Messina
I do that with codecs and I do it with CLAUDE code. So essentially, like you said, if you're writing a bunch of plan files as markdown files, I will essentially have Codex go in and be a critiquer and it'll provide a bunch of like, evals about whether or not the plan is good. And then I'll take whatever it outputs and I'll bring it back to Claude and I'll be like, okay, you need to like, work on these things.
- [56:19]
Brian
So I'm like the. I'm like the
- [56:22]
Chris Messina
coordinator.
- [56:23]
Brian
I'm the child in the middle that is getting mom and dad to do what I want them to do by playing them off against each other. Your mom said this, your dad said this. Yeah, yeah, yeah. Okay, so have you gone down the route and if this isn't interesting, I want to know what you have been doing. Have you gone down the route of having OpenClaw and agents running on your
- [56:49]
Chris Messina
sort of. I've dabbled in it. You know, I was quite conservative and more reluctant to bring open claw like into my local machine. I don't have a spare Mac Mini running around and I think that while it sounds super powerful, I also just can't afford that level of vulnerability. However, the folks at every have a sort of an Early Access hosted OpenClaw products that you can access through Slack. It's called Plus One. And so I got into their beta and was using that for a little while and you know, it was very interesting on a number of fronts. One of the more interesting things was that there was a shared Slack channel with all the beta users and so everyone's ones were kind of in the same Slack channel and they could like talk to each other and then every has a product called Proof which is an agentic text editor that's on the web. And so essentially you can be like, hey, what did I call mine? I called mine Old Bay was the name of my plus one, you know, crab seasoning.
- [57:56]
Brian
I thought you were referring to an old girlfriend, but yep. Okay.
- [58:00]
Chris Messina
Not B E Y B A Y. The spice, the seasoning. Yeah. And I'd be like, hey Old Bay, you know, go write a proof doc about this thing. And it would go like generate something and it would send me the link and then we could like kind of collaborate on a real time. I went in and I did a bunch of like config stuff, you know, in the cloud hosted thing and it was, it was fine. I mean it's super cool that it does a bunch of things, but there was just like these failure cases that it felt like this is not my software, this is hosted. Like I can get a bunch of this stuff done using cloud code effectively and running locally. And so I didn't really stick with it. I think what is really interesting about it is how accessible it is, especially like on mobile or through like messaging apps.
- [58:41]
Brian
Yeah.
- [58:41]
Chris Messina
And then how you can like set it up to do things for you. And I know like Anthropic is moving forward with adding some of these features. You know, again this, this code conference that I just went to, they were Showing off their cloud agents and routines and memories and dreams and all these different things. And so they're getting there. But no, I haven't gone super deep on it.
- [59:05]
Brian
So you've already given us one thing. What have you done with AI that has tangibly improved your life, which was the calendar thing that you're mentioning, where you're working less but making the same amount of money. Do you got another example? And it can be as. I also created an app that's on my phone that I know exactly, exactly how long it takes me to walk to the subway stop that's nearest my house. I plugged into the MTA's data and I can plug in an address in Manhattan and I can say, let me know when I need to leave my house to hit the next F train to get into this. Like, and like, that's just an app on my phone now. And that was like, trivial.
- [59:44]
Chris Messina
You know, it's interesting to me because I feel like I've been living this life for many years, but, you know, without the ability to, you know, it's sort of like, I don't know, driving a go kart without the ability to like, open up the hood and like, replace the battery or something. And so. Yeah, that's.
- [59:57]
Brian
That's exactly right. Like, you can drive the car, but if it breaks, like, you actually don't know how it's functioning.
- [60:04]
Chris Messina
Correct.
- [60:04]
Brian
You open up the hood and it looks like a bunch of weird, scary stuff. And if it breaks down, you wouldn't know what to do with it.
- [60:11]
Chris Messina
Yeah. Or if it's a car, you know, the only, the only two things that you could really do is like, you know, put gasoline in it or like put some windshield wiper fluid in. Exactly.
- [60:18]
Brian
Turn it on and turn it off and try to get it to work again.
- [60:21]
Chris Messina
And then you take it to the experts, you know, and you paid them $10,000 to build the next version of it.
- [60:24]
Brian
Yeah, yeah.
- [60:27]
Chris Messina
I mean, I mean that I was doing this before because I was a long term Alfred user and so I would build workflows and Alfred had like a visual workflow builder where you could go in and, you know, connect different services. So I'd done that for Twitter and I'd done it for a bunch of other apps. Raycast is very, very similar. And I've talked about Raycast quite a bit in the past and how I've built extensions, you know, for myself. And that's just like blown up in terms of my productivity on that platform. So I've been down this path before, but I would say that your experience lines up with. I mean I'm building two pieces of software that are significant. I'm not planning to release them, they're basically just for me. One is a financial tracker which seems to be what a lot of people do. But my intention is to replace my Airtable subscription which is far too expensive now. And so I'm building and I'm sure there's people that would advise me against doing this, but I'm building a local Docker based and I'm actually using Supabase to pull in all my transactions and to normalize and dedupe and do all this stuff. And it's probably the most ambitious project that I'm working on. So that's one and then the other one which was sort of a fun project to play around with Swift UI is to build an app that allows me to look into my ring cameras from both like the Mac menu bar or from like a desktop app. And the reason why I need this is because like my shed where I work is in like the back of the house and the doorbell is in the front and it's a pain in the ass for me to like pull open my phone and like see when the doorbell rings. What I want is just to like go to the menu bar and be like okay, what's, what's happening in the front? And so I built my first Swift app to do this and I've been like iterating on that. So those are two other project, you know, one is saving money and hopefully building something that's very specific to my needs. And then the second one is more like a life quality of life enhancement.
- [62:28]
Brian
The financial one is interesting. I did remember how there was a time when I had somebody create an app for me to be like the 52 week highs just so I could know what companies. But also I have been paying for years a service $100 a year for most of my money is in a strategy called dividend appreciation. So like you invest in things like Procter and Gamble or whatever like dividend aristocrats that raise their dividend every year if you get in early enough. I see in your life you can like your, your yield on cost will be like 80% by the time you need to retire, whatever. So the key to that strategy is that works. But if any of those. So like I have a portfolio of 50 companies that are dividend raisers, right. But if any of them run into trouble and cut their dividend or like then I want to sell out of them. So I paid for years, a hundred dollars a year for someone to, like, do that for me. I created a thing that does that for me. So once a month, I get an email that says, hey, by the. And not only because not all the time is someone in the portfolio in trouble. It's. I had it rank it. Here are the ones you should be worried about. Here are the ones that you shouldn't be worried about or things like that. So again, that is something. That's finance. That's something that anybody could do, maybe if you're not investing in dividends. But, like, one of the things that I also have been motivated for this Yi AI stuff is, again, being as old as I am. I remember the adults with the Apple iis, not even the Macintoshes, that were the first people to show me the earliest personal computers. You basically had to know how to program to use them. Right. And would be like, look at what I can make this machine do. And I would be like. And they would literally. This is the joke. Like, I put. I organized all my resumes into. Not resume my recipes into this. And like, that's always been the joke. Like, what do you do with it? I store my recipes. Yeah, I've got a book over there. Okay. But there was a very quick period from where it went from that hobbyist thing to people doing actual tangible things that improve their lives. I would say that from again, putting on my Internet history hat thing. For most people, computers really didn't change their lives until email and the web came around. Because then you could do things with computers that were like shopping and talking to grandma.
- [64:56]
Chris Messina
Well, I mean, like, you know, laying out pages and printing things. Yeah, you know.
- [65:00]
Brian
Yeah, we did print stuff everywhere. Every time someone had a birthday party.
- [65:03]
Chris Messina
Yeah, exactly.
- [65:05]
Brian
But then. Then one day you wake up and it's like, oh, yeah. But also, all of this stuff is happening. And because I wasn't paying attention to it, I missed it. And I hope my dad does not listen to this. He won't. My mom will, and I hope she doesn't tell him this. But like, my dad, I feel like he's a great guy, but he missed out on things because he was at the stage when the personal computer entered his office and he couldn't handle. I mean, he had a career 30 years after the computer entered his office, but he really didn't get with that paradigm shift. Right. He worked with it, but he didn't really understand it. And I feel like that kind of killed what, number one, it didn't help his career. And number two, it killed his joy in what he did. And so one of the things.
- [65:56]
Chris Messina
What did he do? What was his job?
- [65:58]
Brian
Well, he worked for General Motors doing finance stuff. And then because he hated computers coming in so much, he became a teacher.
- [66:09]
Chris Messina
Wow, I see. Okay.
- [66:12]
Brian
And then towards the tail end of being a teacher, retired when he was like, everything's on these. I got to do all my lesson planning on computers.
- [66:21]
Chris Messina
They're everywhere.
- [66:22]
Brian
So anyway, my other motivation for this is personal, not wanting to have that happen. B. Also, if again, history repeats itself and it's like, this is a. It was a shift where computers come in and change work, change life, and things like that. And like, you can be at the be. You don't have to be at the bleeding edge like the swixes of the world and the software developers that are all in cerebral valley doing all this stuff and doing. You don't have to be doing openclaw on your whatever, but you can also be aware of the things that are starting to become possible for your everyday life. And if for nothing else, you can be the asshole that tells all your friends, hey, I'm doing this AI stuff. And they're like. And then. But they'll be doing it too, in six or seven years.
- [67:16]
Chris Messina
You know, I guess like this. Yeah. I'm probably like a little bit in the future in terms of, you know, where you are. And that's normally been like our dynamic. I think it's actually what your. Your call to action here is even more profound than that for a couple reasons. You know, I think you're one. Yes. Looking back and saying, you know, don't let what happened to my dad happen to you with AI, I think it's. Is very reasonable and predicated on, as would be expected, your historical lens. I think, projecting into the future. What I see happening is that there's a lot of resistance and a lot of ignorance and a lot of assumptions about how technologies work.
- [68:03]
Brian
That's interesting.
- [68:04]
Chris Messina
It's actually, I think, removing people's agency. And there's a set of ideas about regulation and about how these things are happening to people, that if you don't engage with it and if you don't figure out what's possible, then you're coming into these conversations, you know, quite ignorant. And it's almost like, you know, you're a lamplighter, as you know, the dawn of electricity is happening, and you see that houses are burning down, and you're like, yes, but your lanterns are also burning down houses. And the potential for this is so much greater and broader that we are basically making intelligence available. And I know I'm sounding super pilled about this stuff, but like, as you and I have been talking for the last hour, like, we are just at the beginning of having intelligence available on demand in order to make us actually smarter and savvier in how we make use of the dominant paradigm for coordination and collaboration amongst humans. And if you don't engage with that, it is going to happen by a bunch of other people who are seeing the amount of power they're going to have and driving to that end.
- [69:08]
Brian
And also because this is part of the manifesto for this project of how I AI is that it's sort of like a Pascal's Wager. You know, God may not exist, but you know, if he does, might as well. Yeah, okay, so I'm not sure that I want this AI revolution to happen. In a lot of ways, I'm both
- [69:33]
Chris Messina
like, I think skepticism and doubt are very valid and real. You know, it was the same thing. And I guess, you know, I'm pilled based on my experience with social media and, you know, the way in which, you know. Now obviously it did turn out to be not that great. And that's because, you know, humans are not that great. You know, so it's just an amplifier. But for those humans who are great, they're able to reach an audience and express what they love to do and share those things better than they ever could before. And people can enjoy what they produce in ways, you know, that would never have been possible.
- [70:03]
Brian
So by Pascal's Wager, part of this manifesto is, I'm not sure that I want the AI stuff to happen and you might not be either. And you tried it out once and you're like. And it gave you the wrong answer and you're like, well, this is all bullshit. Yeah, yeah, yeah. Well, okay, none of that is wrong. But if it ends up being right, would or right not binary, like 100% right.
- [70:35]
Chris Messina
Totally.
- [70:36]
Brian
But if it ends up being right,
- [70:37]
Chris Messina
there's a spectrum of effects.
- [70:38]
Brian
If it ends up being right to a certain degree, you lose nothing by being conversant in it. Right.
- [70:47]
Chris Messina
I mean, there's an opportunity cost. You could go learn to bake bread or prepare for the zombie apocalypse. I suppose what I will say though is, and I think think we've touched on it, the ways in which AI has affected and changed me has been in some ways providing more of a mirror for gaps or ways that I could actually excel or learn better or ask better questions. And so to the degree that this is a tool of personal empowerment, which then becomes social empowerment, I think that's another angle to take this as opposed to being like, oh, this is happening to me and my work is going to suck. And I just have to like, no, like actually this is a moldable, shapeable substrate. You know, it's like clay. And so, you know, clay doesn't tell you what it needs to be turned into. That's where your creativity, presuming you have some and you want to preserve it, can be expressed. But you know, to this conversation that we've been having, like getting the vernacular and learning, I mean, you know, get. Is very valuable and very important in terms of actually doing this stuff well. And so once you learn some of those things, I don't know, it just like it's been reshaping my thinking and helping me to understand and see my blind spots, you know, in a way that if I had rejected it out of hand, I don't think I would have, you know, been growing as much as I have been as a human in the last, you know, year.
- [72:05]
Brian
And I think again, and I might be wrong about this, but my positioning, I think because like, there's no way that I am. You're further more advanced than I am. I'm nowhere close to someone like Swix or like the every folks like Dan Shipper and stuff like that. So I can't do something that's like everybody. There's tons of people filling the niche of, well, here's the latest in the cutting edge. I'm not interested in that because I'm not there.
- [72:32]
Chris Messina
Yeah.
- [72:32]
Brian
So I.
- [72:33]
Chris Messina
You are where you are, which is probably where a lot of other people are also.
- [72:36]
Brian
But also I'm probably ahead of a
- [72:40]
Chris Messina
lot of huge number of people.
- [72:41]
Brian
Right. So I want to then be the fast follower to the or I want to do something that will be like that. Where it's like, hey, this is possible and here's a tangible thing. All right, two quick questions and then I'll let you go. Is there anything else that you have done in your personal life with AI that you're like, yeah, people need to know about this because it actually is pretty fucking.
- [73:09]
Chris Messina
Well, I guess on the flip side of everything we've just talked about in terms of personal productivity, I would say in terms of my own qualities, I don't know as a quasi humanist technologist, is that I have been using products like granola and other AI notetakers in a lot of my personal relational conversations. And so you know, Portia and I will actually capture some of our heated debates and conversations and conflicts.
- [73:46]
Brian
Do you have a Plaud, which, by the way, is a sponsor, except for the fact that I already had one. So this is not.
- [73:52]
Chris Messina
You can send one to me.
- [73:53]
Brian
Yeah, and it's useful because it. Yeah, you can record your conversation with your partner in real time. Because it's like, I want a source of truth that both of us can access.
- [74:05]
Chris Messina
So there's that. Right. Which is adjudicating conflict. But the other side is actually, again, remembering, analyzing. Bring a third party, a third party who is on my partner side to then argue with me and then to use my own psychology against me to help me see in the ways in which I'm not meeting the moment or in which I could actually improve, you know, my understanding or how I communicate. So that's another area. And just by having data over time that allows you to see how your arguments are actually improving and the quality of connection and repair after them, I think is super valuable and useful. So that's one.
- [74:52]
Brian
So I'm going to end with the second one. Archie, shut up. I don't know if you can hear that. You started at the beginning by saying that, Brian, and you invested money in a fund that was an AI fund, but you weren't using it in your personal life. And then part of this is me explaining why I didn't think it was applicable to my personal life. And that's kind of, again, the thesis and the manifesto of this project. But again, acknowledging that you're always more technical and further down the road and quicker to new things than me. Would you say that again for using AI in your personal life? Could you point to one thing that you were like your aha moment of like, if I can do this, then what else could I do with it?
- [75:43]
Chris Messina
You know, I honestly, I just think that. So what I try to do with my plus one and what I tried to do with Claude Cowork before and what I'm continuing to work on, I think it's more useful as a conceptual frame than success yet. And the project is what I call reverse Claude code. And so it's less about me telling Claude what to do and more about figuring out how to explain or express my goals, to create ways of keeping myself accountable. And then to have, you know, Claude or whatever AI tool I'm using to essentially reflect back, you know, whether I'm making progress against those things, you know, over what time horizons I am. I'm so like ADH head, you know, where Like, I just can't. I have so many different things that are pulling me in different directions that I need something that's there that says, you know, by the way, Chris, you know, at the beginning in January, you said these were your goals. You know, let's keep coming back to them and then let's decide if they're still your goals and let's keep moving in that direction. And so the idea is for, you know, Claude, or whatever, generative AI to essentially program me and to protect my attention because there are so many shiny things for me to pursue. So it's a work in progress. But I've connected Claude to my things, MCP to reminders, like Apple Reminders. I've connected it to my calendar, I've connected it to my email. So it has this broad perspective of things and every week it gives me a list of all the distractions that I had. It's like, these are the things that are keeping you from achieving the things that you said that you want to do. What do you want to do about it? So gradually I'm trying to make progress in that way.
- [77:24]
Brian
And by the way, again, and to end here, it doesn't always have to be right. That's not the point of this.
- [77:30]
Chris Messina
Correct. It's more about bringing awareness or.
- [77:34]
Brian
It said nine things and only one of them was a good idea, but it's an idea that you hadn't considered before. That's right. Well, listen, you just said mcp and like, that's another thing that I would have known conceptually, but now I actually know in reality. And we should probably do an episode about where we can explain to people what that means, because again, that's something very simple that normal people can fuck around with and find useful. Chris, thanks so much. Is there anything that you're working on right now? I don't know. I gotta out find fucking fix the cul de sac. I went down on the code. I don't know when I'm going to put this out, but I want resume writing to be live first.
- [78:18]
Chris Messina
But if you don't mind burning the tokens, like, honestly, I think you should try a reboot and just be like Claude, from everything that we've done and achieved, I want you to write a new spec.
- [78:27]
Brian
Wait, you're saying, like burn the whole thing down as opposed to reverting to the previous.
- [78:33]
Chris Messina
Yes, because you might be on a cul de sac within a cul de sac.
- [78:36]
Brian
That's interesting.
- [78:38]
Chris Messina
That's what I'm saying. I'm literally about to Have Claude rebuild this entire thing that I just built because my assumptions have changed and my knowledge of the project has changed and how I want it to work has changed.
- [78:47]
Brian
Right. By working on it, I've learned more than I thought.
- [78:50]
Chris Messina
Correct?
- [78:51]
Brian
Yeah. Yeah.
- [78:52]
Chris Messina
So like, imagine if you were starting the project now, knowing what you know.
- [78:54]
Brian
Yeah.
- [78:55]
Chris Messina
How would that be different? And how much faster could you get to where you are now? Like, you could honestly just like take a bunch of screenshots of how it works.
- [79:01]
Brian
Well, also, because the code base exists. So, like, if we know what's.
- [79:04]
Chris Messina
You can always go back to it.
- [79:05]
Brian
Well, but also. But it can build off it. I can be like, hey, just use this for parts.
- [79:12]
Chris Messina
Yeah, yeah, yeah. So that's like all the. Like, I go to. I have a deep research project or skill that'll go look for existing open source implementations of the thing that I want to do. And we'll grab those as just reference and we'll say, look at how they solve these problems or look at these edge cases. You know, let's create a plan that incorporates the learnings from those projects. Then let's build our own code because oftentimes it'll be like Ruby code, but I'm writing in Typescript or something.
- [79:42]
Brian
Yeah. Hey, I'm writing in Typescript too. Even though I have no idea what it is.
- [79:50]
Chris Messina
It's a good choice.
- [79:52]
Brian
I just tell Claude to do it. Okay. Chris, I used to say I love everybody, but I love you.
- [80:00]
Chris Messina
You too.
- [80:00]
Brian
Thanks for coming on again.
- [80:02]
Chris Messina
Good to see you later.