Michael Fafinski owns Gutter Galaxy. Last time he was on the show, back in December, we talked gutters and the homeowner side of his business. Six months later he's cut the homeowner division, gone B2B, and rebuilt the way he runs the company around an AI second brain. He's on the $200 Claude plan and maxing it out every week. Not writing code. Running a gutter business.
We get into the real mechanics: how he built an AI "executive team" trained on actual business books instead of generic advice, how he turns a 300-page book into frameworks his AI can pull up on demand, and why the whole thing lives in plain markdown folders he owns, so he can walk to a different AI company any time he wants. I share my side too, including the 500 pages of estimating tables I scanned into my own vault and the "engineer trap" rule my system uses to stop me from building things nobody asked for.
If you're a contractor who thinks you're behind on AI, this episode is for you. You're not behind. Most people are still using this stuff to check the weather. The ones who put in a real 40 hours are lapping everyone.
Watch: YouTube Listen: Apple Podcasts · Spotify
Chapters
- 00:00 - Gutter Galaxy update: cutting the homeowner division, going B2B
- 02:52 - The AI executive team: a CEO and CRO trained on real business books
- 05:49 - Building the second brain: whole books into frameworks
- 09:01 - Folders, lenses, and talking to it like another employee
- 11:53 - Skills: making the AI do real work
- 14:54 - Building in markdown before the second-brain wave hit
- 17:52 - Model talk: what the next model actually buys you
- 26:04 - Scraping and research: filling the brain
- 28:47 - Startups vs the model companies
- 30:11 - Portability: your brain moves when the tech changes
- 32:13 - Open source keeps the big guys honest
- 36:28 - Privacy and what stays out of the vault
- 38:34 - API keys, trust, and control
- 42:01 - The AI that knows your tendencies
- 44:00 - The engineer trap: validate before you build
- 49:21 - You're not behind: start the second brain
- 50:19 - 500 scanned pages: getting paper into the vault
- 54:18 - When does the non-technical user get in?
- 01:00:28 - A patient tutor for every kid
- 01:05:38 - AI, education, and who gets hired
- 01:10:55 - Just start: 40 hours puts you ahead
The big takeaways
- A second brain is just organized folders of plain text the AI can read. You own it, and you can move it to any AI company whenever you want.
- Michael trained AI "executives" (a CEO, a CRO) on real books so strategy questions get answered from real frameworks instead of generic model knowledge.
- Digesting a book into frameworks keeps it cheap: the AI loads one concept, not 300 pages.
- The organization matters less than starting. Michael's line: "just having it is 90% of the battle."
- Paper is dead weight. My scanned estimating class is now one of the most useful things in my vault.
- Nobody is behind. Real proficiency is about 40 focused hours away.
Full Transcript
Reasonable Tech Dad (00:00) Welcome, ladies and gentlemen, to another episode of the Reasonable Contractor Podcast. I am your host, Ryan Templin, the Reasonable Tech Dad. And we're back after a long hiatus. We took a couple months off, did some transitioning, do some project management now. And one of our all time favorites is back, Michael Fafinski with Gutter Galaxy. A new background, nice logo back there, sir. That stuff's classy. I like it a lot. So give the people an update, man. What's going on with Gutter Galaxy?
Michael Fafinski (00:19) stop.
Thank you.
Some big updates. I think what it was December last time we talked. So that's been about six months. that podcast was a lot about the homeowner division. but we've since really cut that, really minimized it. And now we're focusing on our B2B relationships. And I'm in the middle of a quarterly str strategy reassessment. so a lot of fun stuff with that and changing the business around. And then most importantly, the
Reasonable Tech Dad (00:55) Yes.
Michael Fafinski (00:58) AI stuff I've been getting into, the obsidian, everything like that. I have been on Claude, the two hundred dollar a month plan nonstop, every day. yeah. yeah. yeah.
Reasonable Tech Dad (01:06) Maxing it out? You've been hitting the limits on two hundred? That a boy. That's no small feat.
That's no small feat maxing out the two hundred plan. Cause when I first started doing two hundred, I actually I I was maxing it out for a little bit when I was writing a lot of code and I was letting them agents do in layers, but I ended up scaling back to the one hundred after you're not doing heavy stuff. How are you maxing out the two hundred? What kind of stuff are you doing?
Michael Fafinski (01:14) It's not.
Mm-hmm.
Well, it changes week to week, but I've been setting up my vault quite a bit and restructure I restructuring it. the biggest thing like I would think in like frameworks and concepts. so I utilize a lot of like books for that, whether that's Alex Hermozzi, Geno Wickman, Jim Collins, stuff like that. So I didn't want my AI to have like very basic knowledge when I came to it with like a strategy issue. So I built like my CEO.
or my CRO chief revenue officer, and all down the line. And then I trained
Reasonable Tech Dad (02:01) Do you do different vaults for each
of those or do you keep those in one vault?
Michael Fafinski (02:05) one vault different projects, different projects. And then I train each one of those models on like the specific concepts that I want it to. So it does like the research of what it should have for content based on like what I do. And then I can confirm or add books or whatever, and and then I digest all of those books. So it takes the PDF, digests it, then pulls out the frameworks.
And so that whenever I go like, what does the CRO think of this? It can pull up like the one page marketing plan and then I go, Okay, let's do this and figure this out. so it's using real concepts, not just some generic stuff.
Reasonable Tech Dad (02:45) Can I double-click on that? So when you say projects, I think of like individual projects in the clot interface, right? And those are isolated context collections. And when you say vault and obsidian, so for like second brain, for if people anybody listening to this AI thing, maybe, maybe not, but the second brain concept came from Carpa Carpathi, right? Andre Car Carpathy. And
Michael Fafinski (02:52) Yeah.
Yep, Andre.
Reasonable Tech Dad (03:12) The Obsidian craze was all about the fact that you could make every piece of context markdown, which is what the LLM's like. That make it's just and if you don't know what markdown is, it is it's a notepad file that used to be TXT and Markdown, it's it's straight text. And that's what the LLMs are built on. That's what the transformer architecture is. The reason I ask about when you say you broke your positions into projects, can you explain to that to me? So in my second brain, I use obsidian.
Michael Fafinski (03:21) Mm-hmm.
Yeah.
Reasonable Tech Dad (03:41) And I have top level directories. Is that what you're talking about in projects? And then when you said you train a model, walk me through that.
Michael Fafinski (03:50) Yeah, so I mean the real thing about this that makes it so amazing is that every person's vault is drastically different. And you get to customize it to exactly what you want. And so but that's also why it's so confusing, because everybody's talking about their personal experience with it, and then you start doing it and you're like, What the hell do I do? You don't know where to start. Yeah, yeah. And so then you really just gotta like dive in and then and then figure out along the way. When I say projects, I have like
Reasonable Tech Dad (04:00) I agree.
That's not what he was talking about. Yeah.
Michael Fafinski (04:19) I organize it. 01 is updates. So that's like it'll look at my emails, anything that I need to look at for that day, put a markdown file on that folder. That's like relevant information. And then zero two projects is my different folders that I have, the CRO in, the CEO in, and also other stuff like my squirrel cage. So whenever I have business ideas or any little ideas on my vault, I can just say
Take this to the squirrel cage, it'll organize it, put it in the right spot. I can find it later, but then I don't have to like start building it out because that's gonna take me an hour and then I've not gotten any work done. and then you go down, I have like review zero four reviews and maintenance, zero five people, zero six concepts, and and so on. And you know, it doesn't really make sense to other people what's in there, but to me, that all, you know, I know what's what's in there.
Reasonable Tech Dad (04:50) I love the scroll cage.
The rabbit holes. Yes.
So those are s
your projects are still just file folders. With subfolders and markdown files. Got it.
Michael Fafinski (05:19) Yes, they're they're just folders. Yep. And then
each folder has a lens, I call it, which is the the CEO's lens on this. So it loads the lens and then it can go, okay, what would the perspective of the CEO that's trained on these different books have on this? And then it knows, okay, what concepts to pull in, what books to pull in from that. and then you said training the model.
that for me, I created a skill called concept ingest. So it does the research for me of what what concepts, what books, what articles, what what videos should be brought in for this. And and that part you really need to coach it along. It's not gonna do it all for you. You need to give it the right context and and pull what you know. Yeah, it does. Yes. And then once once I say, okay.
Reasonable Tech Dad (06:10) Takes tons of time. But it's so worth it.
Michael Fafinski (06:18) Here's the research done. And then I go on to the next step of ingesting it. It'll take the books as a PDF, ingest it using Mark It Down, which is a plugin or a skill, you could call it, from GitHub, that changes the PDF into markdown files so that it it's not super token heavy. Because reading a PDF is super token heavy. then it has the raw book, and then from there it'll digest it into.
Like just the core concepts. And then from there, it'll take that one markdown file of the entire book that's called the digest, and then move it into separate frameworks. So I can say, hey, I'm setting my my EOS rocks, my my traction rocks for the quarter, pull in that concept and let's start working on it. And then it doesn't load in the entire book, it just loads in like that specific concept within the book. So it's super token efficient, and I'm not
Spending out the Hazusa on on tokens.
Reasonable Tech Dad (07:20) So at the end of the day, what you're left with is a distilled paragraph of text that is in your CEO folder that when it goes to reference that project, as you call it, it's still just a file folder, but it is text that outlines, it's basically the first chunk of text that's digested when that task is executed, correct? Am I thinking about it correctly? And everything that's the nice part about the vault in Obsidian is the wiki links and the graphs.
Michael Fafinski (07:39) Mm-hmm. Mm-hmm. And then it knows where to go.
Mm-hmm.
Reasonable Tech Dad (07:48) People look at and they're like, what in the hell? And they should think say that because yeah, it's super overwhelming and you'll never know until you like you said, you gotta dig in. And I'll tell you right now, mine me mine with the last couple months being in the way it has, busy and chaotic, mine's a mess. Mine's a mess. Like I've I've literally been bolting new pieces on because I know that I can just talk to it and go and build a new piece for it. So I've got these chunks and I mine mine are laid out the same way. I number too.
Michael Fafinski (08:16) Yeah.
Reasonable Tech Dad (08:17) My numbering is actually only because I like them to be in a certain sequence so that I can use a review when I'm talking about certain pieces. So I like in a certain sequence. That was the only reason I numbered them myself. Why did you number yours? Is that part of your link? Got it. Fair.
Michael Fafinski (08:31) can keep track of it. I
just was like, where is everything in my vault? And then I I there's actually a plug-in inside of Obsidian called Iconize. And it and it you can put an icon in front of each folder. And that was like, my God. All right, now I can actually like visually see it and find it.
Reasonable Tech Dad (08:35) Yes.
I spent
two, maybe three hours one night shopping for the right God, what is it? Do they call it layouts in the in the in the in the library there? So that I got the right color and the cause how you look at obsidian the original way sucks. There's no there's no lines, there's no hierarchy, there's no dots, there's no no hyphens. I can't handle that. I cannot handle that. And I go
Michael Fafinski (09:01) Mm-hmm. Theme. it's terrible.
I try to show
somebody it and they're just like, This is a foreign like this is code. Like what is happening?
Reasonable Tech Dad (09:18) Exactly right. Yeah.
Yep. Just not on the geek level, man. So and so but so you use a skill to build your concepts out and then your markdown is ready. And what's really nice about that, if anybody wants a simplified version, take away all the geek talk we're we've you know discussed. What's nice about spending the time to build the structure that we've talked about is that once you're there
I use Cloud Code still for the interface, but you can as long as you have a voice dictation tool on your computer, you can now go into Cloud Code and you can talk to it like a person. And you can ask it about those pieces and all you have to do is talk to it like another employee and it knows where to go. And it gets you the information and it can actually do work, which is the amazing piece.
Michael Fafinski (10:10) Insane.
Like it just the the abilities of it, especially if you own a business, is just 10x because it has all of the content and context that you would need, not just like the general context that a normal that that you normally get inside of Claude in you know, just it's it's memory.
Reasonable Tech Dad (10:32) Couldn't agree more. And you couldn't, you could never you could never have a person to cover the range that your your programmed AI does with a vault, the second print. Because when you build those skills you're talking about, you can like so for me, one of my key folders on number two, because if you you've heard of the concept of like an AI OS, obviously, like an operating system building your whole business or your whole whatever into an AI operating system. So mine
Michael Fafinski (10:40) Mm-hmm.
Mm-hmm.
Reasonable Tech Dad (10:58) Mine's called Codsworth, because I'm a fallout geek. I I like the the Butler robot. It's got the British vibes and it it gives it puts a little British fallout swag on there for me. and I've been thinking about avatars, you know, the AI avatars. If I could get it graphics so that it's the floating guy with the hands and the saws and all that. And now it's construction, so it'd have tools on. But the one of my top directories in my vault is OS. And what I've done, what's one of my favorites, is
Michael Fafinski (11:09) Yeah.
that's funny.
Reasonable Tech Dad (11:25) A lot of media tools you can build skills. So once you build those CLOD skills, you can use those by voice. So it was interesting when you were talking about PDF to Markdown. What I realized is that a lot of the command line tools for changing video formats, changing audio formats, PDF to text, generating a PDF from something that you an out like if when an LLM puts out a piece of information that you want to turn into a PDF.
There's skills for that. And once your skills are built, and I tell it, hey, look in the OS, look in the Codsworth, look in your OS, so that basically I don't remember all the media tools. There's like 15 of them. But this is the task I need. Go look in the OS for it and it'll find the skill. I don't even have to remember the name of the skill. Once I build the skill, I can tell it what I'm trying to do and it'll run the skill just on me telling it what I need. Which is unreal.
Michael Fafinski (12:10) Mm-hmm.
Mm-hmm.
Mm-hmm. And the beauty is that the Claude markdown file that it takes in on every chat can direct it to those. So you you know, I and that's and that's the thing is it the vault just becomes a a directory of go to here, go to here for more information. And you don't have to actually feed your claude.markdown file with all of that information you're trying to to jam pack in there so it understands you.
Reasonable Tech Dad (12:30) Correct.
Michael Fafinski (12:52) It you can just drill down and down and down farther and and yeah. What are some of the the tools that you've been using mostly?
Reasonable Tech Dad (12:57) The way that
I'm strictly in cloud code. I just use cloud code. When you say tools, you're talking about the skills or which interface am I interacting with it?
Michael Fafinski (13:06) Like skills,
MCP servers, connectors, plugins, anything like that.
Reasonable Tech Dad (13:14) So I don't I haven't really gotten into a lot of the plugins and connectors because I have been in the I've been so hardcore in the code side of things is that I was building so much of my infrastructure two years ago in Markdown that I had until the second brain concept came around, I was actually ahead of the Carpathy wave because somebody told me about obsidian.
Michael Fafinski (13:36) Really?
Reasonable Tech Dad (13:39) I don't remember exactly. you know where it was? I don't know if we talked about this, but I'm in a group. He's actually from Minnesota, but he's gotten kind of big. His name's Cole Maidine. And he has a big YouTube channel that's grown quickly. And it's two years ago, he's a former software engineer. And he had something he called the Dynamis community. And Dynamis community is hardcore agent coding. Like add it from the language. Like we're bringing building Lang graphs.
Michael Fafinski (13:49) Okay.
Okay.
Reasonable Tech Dad (14:09) agents from scratch kind of stuff. Node structure, writing the code. So I've been doing that since way back. And he was the first one to bring up obsidian. He had us building second brains. And then like six weeks later, Andre Carpathy came out with the second brain ex post. And I told him, like, Cole, did he just steal your swag? Like, dude, this did he just follow your YouTube video and make himself more famous off of your idea? Because we had already started building obsidian vaults in that coding community. And
Michael Fafinski (14:30) You're you're ahead.
Wow.
Reasonable Tech Dad (14:39) Cole's a rock star, dude. Like, shout out Cole. I'll probably send him a link of this. because he talks, I've told him we've we message from time to time on there and he told him about the podcast and that kind of stuff. But yeah, he's a rock star, dude. So I've never got into plugins because I've had so much markdown structure from the beginning. So way back when I was trying to build apps, I was trying to do second brain in my code repos. I was trying to, I had my architecture.
Michael Fafinski (14:54) Okay.
Reasonable Tech Dad (15:08) architecture folder that would lay out you know the PRD for the coding project. the it would tell it the front end stack, it would tell it the back end tech stack. And every time I was prompting cursor or one of the coding agents, I would tell it, hey, we're working on a front-end design piece, go look at my front-end design markdown. And then make sure you look at my tech stack. I would have to tell it specifically which six files to look at. But
Michael Fafinski (15:13) Yeah.
Reasonable Tech Dad (15:33) I spent so much time building that back end structure that ever since then all these ideas have just made it easier. I've just been able to delete, delete pieces of my system because now they've been packaging it. You know what I mean? To where we're at now.
Michael Fafinski (15:40) Mm-hmm.
And the most important
part is how you organize it. Like you can't do this without truly understanding how to be an architect in these in Obsidian or in these different files. Because yeah, the the AI can recommend things, but like it's not always right. And you gotta be really smart so that you know where it is, but then also that the AI c can find it and it can index and and just like the architecture stuff you're talking about. You had a head start. I'm jealous.
Reasonable Tech Dad (16:12) Yeah, dude.
Michael Fafinski (16:15) I got into this like
Reasonable Tech Dad (16:16) It
it was pure like we us us prosec architecture geeks, like it it used to drive me insane how much the coding agents would get off track. So I'm like, all right, cool, like let me just the you would just build more and more documents and more and more documents and more and more like I had the suit every every coding project I started, I would like stu I started copying chunks of it in because it would start replic it would start duplicating, right? Like if I was using NextJS and if I was using a back end technology like Python for
Michael Fafinski (16:29) Mm-hmm.
Wow.
Reasonable Tech Dad (16:45) Hmm.
I'm blanking on the the agent structure right now, but it's a Python back-end language. But you would basically bring in those markdown documents where it was, it would just know. And it's been cool. And when he in anybody listening, when he says architect, he's not talking about software architecture because that architect now you apply it to whatever you are building. There's marketing architects, there's construction architects, there are
Michael Fafinski (17:10) Mm-hmm.
Reasonable Tech Dad (17:15) Whatever field or industry, wherever you're trying to get work done, the architect, correct me if I'm wrong, but what you're talking about is building the ideas that allow you to think the way you think. Because the AIs can't do that. People are so scared of the AI as itself, but it it's amazingly dumb still. Still, it's amazingly dumb with if you don't arch build that architecture for what you're doing.
Michael Fafinski (17:41) Not fable, I guess, though. That's it's too smart.
Reasonable Tech Dad (17:46) Do you think
there's do you think there's more politics at play or do you think that was really dangerous with the cybersecurity stuff?
Michael Fafinski (17:52) Well, I just read something on X about how I think it was NASA said that it hacked into like all of their sensitive information within hours, not a day. like and obviously they're using it to try to break it, right? It it's like they're in safe hands doing that. But if that's true, then then terrifying. But also like they have mythos without the safeguards. Like with the safeguards, theoretically.
Reasonable Tech Dad (18:11) Terrifying.
Michael Fafinski (18:21) It can't happen. I don't know. Some people say it's just mismanagement of the the government agency, like how Anthropic was just not communicating well with them. Other people say it's a whole marketing ploy that Enthropic wanted this to happen right before the IPO so that they have the best model ever. I don't know. I don't I don't care. I just want fable, man.
Reasonable Tech Dad (18:43) Do you really? To me,
the models like I just actually the other night went back to four point six because I was getting frustrated with how long four point eight was taking. I was like getting I was like, Opus is taking too damn long. I know this is simple shit. I know you can do this. You used to be able to do this. I I switched back to four point six on the million contacts. I'm like, just go faster.
Michael Fafinski (19:00) When you're running six
six different chats at once, then it's fine, all right? Just going back and forth. Like, what is what did the even what did I even do the last chat?
Reasonable Tech Dad (19:11) I
you know what? I feel like what do you think the next model's gonna do for you that four point eight or four point six can't do right now? Are there are there use cases where you think that you're limited by the model? I feel like I haven't had that in a long time. I used to run into that a lot. It used to really matter to me on code. Now code, I trust any of the across the across the board.
Michael Fafinski (19:34) Well, you know, there's just one prompt I've been using that's been a lifesaver for me. It's check your work, look for any mistakes or or issues, and look for any potential improvements. And it always finds something. It always finds something. So that tells me that the model could be better, right? but also I think I actually never got to use Fable because I was like, I'm gonna be a good boy, I'm gonna like get my normal work done.
Reasonable Tech Dad (19:52) Fair.
Michael Fafinski (20:02) I think it came out on like a Wednesday. I was gonna play with it on on Friday. And then I never got to use it.
Reasonable Tech Dad (20:05) left it alone. I had it. I I downloaded it and left it on four point eight. I like I my
my c cloud code updated and I had it. And I'm like I don't need that for nothing. Like I run it and then later that day they sh I'm like wait what? No
Michael Fafinski (20:20) But I mean, honestly, I I think Sonnet to Opus is night and day. Like I truly it is completely a different model, a different person, a different platform I'm talking with. And so like I can't imagine what what what fable would be. And I think really the best part about it is like I would want it to review my vault, see where I could improve, and you know, some higher level strategic stuff.
Reasonable Tech Dad (20:36) Fabulous.
Michael Fafinski (20:48) You know, even just to start off the planning mode and then for acting run run to to opus. but yeah, I I want it. I want it bad.
Reasonable Tech Dad (20:58) As I
listen to you talk, there are there are times. I mean, a lot of my applications, I just have simple stuff, so I but you're right. When 4.8 came out, it's 4.8 is clean. It's cleaner than 4.6 was.
Michael Fafinski (21:06) It still agrees with you, you know, it's yeah.
Mm-hmm.
Reasonable Tech Dad (21:24) really like you're talking about, it was able to you know, so for so long it was about making sure with all this context that you gave it narrow focus. 4.8, you're able to give it very broad focus. And it understands with a when you as long as you have a vault, it's able to pull out the pieces it needs and it narrows on its own, I feel like.
Michael Fafinski (21:36) Yeah. Yeah.
Mm-hmm. I agree. And then also, you were saying you were ahead of the game with Obsidian. I had not used clawed code before, before jumping into Obsidian. So two, maybe two and a half months ago, I had never used clawed code. And then I just was like, my God, like this is insanely better. And then I just spent like 20 hours a week, like every week, just in it.
Reasonable Tech Dad (22:05) Changes everything. Changes everything.
Yep. When if you have used AI
and then you have not used Cloud Code, I only went through that 'cause I was on a a Windows machine that was aging. It was like a five year old Windows machine until December of last year. And then I got a new MacBook Pro. I got an four. And then I was able to run Cloud Code. And that dude, I was a kid. I was so happy. Everything was it was I had 'cause I had an old MacBook too.
Michael Fafinski (22:19) Mm.
Nice.
And you as a coder,
like you you you knew about it, you knew the strength, you wanted it.
Reasonable Tech Dad (22:37) I did
want it and I just didn't want to do the Mac update. I I tried really hard to run cloud code and then 'cause you knew for most of last year they they weren't building anything for Windows first. Like even Anthropic only had a Mac app for the longest. And so I was sitting with my Windows machine f Fritz Fritz. I was still using Windsurf and fucking Cursor was like it was Cursor was cool, but it was still on its way up.
But once I got cloud code it changed everything. Is that what you're using now? Do you work strictly in cloud code or have you switched to cowork? You started dabbling with cowork and browsers yet?
Michael Fafinski (23:10) I skipped right past cowork and went right into i into code. And like I feel like I have all the capabilities of cowork just in code. I just gotta get like a plug in for it. And it it I mean it I don't really use cowork, no. it's simpler but
Reasonable Tech Dad (23:23) Yes. Yep.
Cowork is their
I think cowork is Anthropic's attempt to make it mainstream. I think that they they're really trying to find your more average user. It simplifies things. especially the way you're able to everything is above I actually hate it. I hate it, it's clunky. But have you I what I haven't tried inside Cloud Code and I probably should is finding a skill or a plugin to do browser. Because that's what I've really been tempted to do.
Michael Fafinski (23:33) Yeah. Yeah.
Mm-hmm.
Reasonable Tech Dad (23:57) I gave Claude Cowork last week, I tried to let it put my Target grocery shopping list together. cause I had Markdown all built around it and my eating schedule and shopping for my kids and all that. I tried to let it let co work do the browser piece because it connects to Brave. And it ran up I gave it twenty-five minutes and it used thirty-five dollars in credits and
Michael Fafinski (24:03) really?
Tschüss.
Reasonable Tech Dad (24:25) And it only put 12 of 19 items in the cart. Because it in co-work, it takes pictures of each screen before it clicks. It like scans, it takes five pictures and then puts the pictures together and looks at them. And I'm just watching watching the token meter fill on my usage screen. I'm like, I wanted to give it a shake, and I'm like, nope, no way. So
Michael Fafinski (24:35) Yeah.
That's why I hate the clotting chrome. Yeah. Yeah.
I mean I so do you use Firecrawl at all?
Reasonable Tech Dad (24:50) I've dabbled with it. I've dabbled with it for scraping to do I've I've tried to build people's websites as a mock up to compare them for a new site I made. Like, look at this is your old site, and then I use that old site to build the new one because it's just simpler when you have all the data and it's more like modifying their structure versus creating something new. What are you using it for?
Michael Fafinski (25:00) Okay.
Okay.
a lot of scraping. that's definitely the primary, but I'm curious even if like you could kinda have it use clotting chrome and then also fire crawl. I've heard about playwright. I haven't really dove into that much. And then Yeah, and then I know there's one more that I heard that just like is better than playwright, because like playwright is more specific for coding usage, I think.
Reasonable Tech Dad (25:29) Me either. I've heard playwright's really good for plot browser stuff.
Michael Fafinski (25:42) And then there's one that's released I have not used yet, but I did download. but supposedly there's better one for just search, which I No, because that'd be like a fire crawl similar, I think. so it's different than that. Yeah. No, I can't. But if you looked it up, what's better than playwright? It'll probably come up.
Reasonable Tech Dad (25:49) Apify? Is it Apify?
Yeah, exactly right.
Just can't remember the name right now. That's all good.
What are you what are you doing all your scraping for? Is it work stuff or more of you building your framework stuff?
Michael Fafinski (26:10) some a lot of it's the framework, like scraping and doing the research for me. there's this one skill called last thirty days where it looks at like Instagram and and X and all that stuff. So like when I built out my Alex Harmozzi persona, I really like his philosophy. Yeah, I'm a big fan of him. So I I had it search his X, which he posts a lot on, and then it could get some more of that gradual stuff, not just like his books, right?
Reasonable Tech Dad (26:27) Same. Same.
Yeah.
Michael Fafinski (26:39) But a lot of scraping too, getting getting leads. I also couple it with Apollo's MCP server so I can get that. But I it I feel like you kind of need both. Like people are saying it's like replacing Apollo, but like I I no, I I think just using the scraping with Apollo then makes it like really good. Just using one, you're still having gaps.
Reasonable Tech Dad (27:02) Anytime so I don't like to do that like when you're on X you can say certain words you want to eliminate like you can give it phrases where if it if a post has this phrase in it, don't show me it. And and one that drives me insane is it's over. You know what I mean? So heavily used or it's over for Apollo. It's over for this. And after three years, like for a while for the first year I used to like, Yeah shit, they're right. That shit's fucked. You know what I mean? Like that shit's dead in the water. Nah.
Michael Fafinski (27:09) Mm.
I didn't know that.
Yeah, yeah.
Reasonable Tech Dad (27:32) Nope, nope, nope.
Michael Fafinski (27:33) It's just the hype. And like I I mean, obviously like I feel like you start on a lot of those influencers, you learn from like the people hyping this stuff up. And then I just start like cutting those people out and going to people who are actually giving high level information, not like this next plug in is wiping out this entire company. It's like no.
Reasonable Tech Dad (27:52) There it's it's so wide open, it's so new, and it's gonna be like not I don't wanna use like any other market because it is a transformational technology, but the people that make prognostications in my mind are just full of shit. Like no you just can't know because the c the think about the amount of work that we can get done as your average Joe, like obviously we're ahead of the game in AI, but like we're nothing to these companies comparatively, right? The coding loops that they have, the agents that they have built.
windsurf supposed to be dead, you know, and then they got acquired, you know, bought by OpenAI. You know what I mean? Like the the the acquisitions and it's it's just it's it's you can't you you cannot say anything's dead. So like when people are like this replaces that this is gonna die and this is where we're gonna be it's like okay, whatever. Because this is where we're at right now and in the next month it's gonna change and the next tomorrow it's gonna change and it's it's fun.
Michael Fafinski (28:26) Yeah.
I will
say for the the startups though, it's like it's so competitive because your entire startup could be just toppled over by a feature update in in one software, right? It's like Apollo adds one thing and now your entire software is dead. And yeah.
Reasonable Tech Dad (29:05) Your business. They they said that two
years ago that that was gonna happen. You will get consumed by the model companies.
Michael Fafinski (29:11) Yeah. And it's that's that's a little scary. It's like, how do you come up with a new idea in this market and launch something and and when these mammoths are here and can just be like, we'll just add that. Yeah. Yeah.
Reasonable Tech Dad (29:23) And they do. It's like a push of a button, dude.
Balls of steel. Balls of steel is the only way because you gotta go out there and you gotta grab the cash. And and like some of these people, the stories, it's crazy. The amount of money you can grab, like open claw. Open claw's a great example of what we're talking about. Like, open claw's gonna do this and Claude's gonna kill it. Open claw's still standing. Hermes still standing, you know.
Michael Fafinski (29:30) True.
Mm-hmm. Well, it's open
source. So like it's always gonna have its place in the market for people who want to be able to go, like with the obsidian stuff we're talking about, we can switch to codecs or wherever we want to go. We don't have to stay with Claude. So if Claude IPOs and and they they decide, hey, we're gonna we're gonna screw all these people and charge them two times as much, you can take your AI OS that you've created and just uproot it and bring it to another another model.
Reasonable Tech Dad (30:11) I think that's a super important part because anybody that was listening that glazed over the obsidian piece. Obsidian is very important. It's very well, it doesn't have to be obsidian. Once you learn what it is, you actually can just go create folders on your Mac or on your window. Just start creating folders. And that's all it is. It's just a folder directory. But it what it does is it does protect you. It does protect you from it gives you the portability.
Michael Fafinski (30:26) Mm-hmm.
Reasonable Tech Dad (30:37) to change when the tech changes. And I think Obsidian's brought me a lot of peace. And my second brain has brought me a lot of peace that it's portable. cause everybody was saying that Anthropic was going to take over the enterprise game and they were so far ahead in coding, but Codex is made. Have you gone to Codex? I haven't even used Codex.
Michael Fafinski (30:37) Mm-hmm.
I haven't once. I've just stayed in Claude because I've been like I don't wanna mess around with a ton of different stuff.
Reasonable Tech Dad (30:58) It took me so long to learn clo
clot, I'm like, I'm not I'm sitting here talking about how portability is important, but I'm stuck right now. I'm fucking stuck. Nope, nope, nope, nope, nope, nope. So I'm just over here contradicting myself, don't mind me.
Michael Fafinski (31:02) Yeah.
Yeah, we we haven't touched anything else.
They I think that Claude almost made it so we have it tied into our usage in Obsidian, I'm I'm guessing where it actually when you're using it in Claude, sorry, in Obsidian, it's taking from your token session hours or or your your your windows. and they almost made it so that it those had to come from API keys. And it was super exp
Reasonable Tech Dad (31:34) Yeah, they they did it for like a
yeah, they did it for like a week and then they backtracked.
Michael Fafinski (31:39) Yeah.
Yeah. And then they went right back like, no wait, people were leaving, people didn't want us. And I was I was happy about that.
Reasonable Tech Dad (31:43) People are pissed. Agreed.
Same. anybody doesn't know what we're talking about? When op OpenClaw, so if you're not familiar, OpenClaw was an open source kind of cloud code type of thing, but more advanced. It gave you you could basically run a you let agents go free a little bit. You could tell it what to do and it would autonomously keep trying things without you having to prompt it individually each time. And everybody loved it because it was open source. It wasn't controlled by anybody.
And Anthropic didn't like that. They didn't like that. So they 'cause what people were doing is they were taking their open claw and they were running it with their anthropic membership, their subscription. So if you paid for like this bulk plan like we talk about, the hundred dollars, two hundred dollars
You could run open claw with that and Anthropic's like, no, no, no, no, no, no. And they tried to put the kibosh on it and everybody riot rioted and revolted and they like, Okay, okay, we were just kidding. You guys go ahead, use it on your membership. It's cool. and I think they just sent out another email like a week ago saying that, yeah, never worry about that. You don't have to do API. You can use your subscription on whatever. It's basically what I took it as. I don't know what you saw, but
Michael Fafinski (32:49) It
it's truly funny because I think there's genuine poll that the community has. Like an anthropic listens, like there are the the X and the Reddit like loop rabbit holes that they start bitching about something that Claude does or that OpenAI does. And yes, and then they're like, crap, and then re they reverse. And it's like you actually have power. Yeah.
Reasonable Tech Dad (33:05) And the fire burns and it grows
The pitchforks come out.
It's cool. I think it's important. I think it's very important because early on I was I loved Sam Altman three years ago. Loved him. Now I hate him. I don't well, that's a strong word. Hate. I don't know if it's hate.
Michael Fafinski (33:19) Yeah.
Reasonable Tech Dad (33:31) But he was, you know, we're for the people and we're making sure that this is proliferating across the board. And over time, the only guy that's told the fucking truth is the guy that's always told the truth. And it's fucking Elon and Grok. And I it I love that Grok has, you know, made them bow and kiss the ring. And, you know, Anthropic's coming to him for usage now and and compute. but I wish Grok it have you seen, have you tried I've been dabbling? Grok has a
They have a code. It's not it's not clod code, but they're they're starting down that road. It's in beta.
Michael Fafinski (34:00) Well they bought
it wasn't cursor, what did they buy? what's the other major code platform?
God, I use it. I should know it. Maybe it's cursor. But the Okay, okay, then it was Cursor. Okay. Yeah, they bought Cursor. And I think that is like really like they were really behind on a lot of their models. But then once they bought Cursor, then it got really good for coding, I think.
Reasonable Tech Dad (34:16) I thought they I was gonna say I thought he did buy cursor. It was cursor, I thought. They did buy cursor, yeah.
See, I think that
they were their models, Grok, because I'm I'm an Elon nut and I've been on his team from the beginning. Their models were fine. His models were always excellent because he's taken a complete, he's taken the radical truth, the same thing he applied to X, and he's put that into the models. If you talk to Grok, I'll tell well, I've tested Grok up and down, like, because I'm a big radical truth seeker, and a lot of my prompts are built that way. Like I I'm a and that comes from what's that?
Michael Fafinski (34:37) Okay.
Okay.
He'll let you do whatever you want.
Do you use the love mode? The love mode?
Reasonable Tech Dad (35:02) I have not dabbled with his modes. I don't use his UI at all. No, it's I try to stay just in my projects. I say I don't use his UI, but that's a old Ray Dalio from the principles, Ray Dalio's principle book, Radical Truth Seeking. And I built that into my prompts for all these years and Elon has always gone that route. And his I think his models were always good, but he he never built the interface because Claude Code is why Anthropic got ahead, right?
Michael Fafinski (35:04) Don't Don't
Yeah.
Mm.
Mm.
Hundred percent.
Reasonable Tech Dad (35:31) And OpenAI got the early jump on everybody because they had Chat GPT. An interface that people could use. Grok and X never made an interface for you to use their models. That's where they were way behind in. Is they didn't have any way to interact with their models. Because once in a while they would they would do a Grok update and cursor would let you put in the API keys to use your Grok, and that thing would code like a rock star, even though it was it was behind.
Michael Fafinski (35:39) Mm-hmm.
Okay.
Reasonable Tech Dad (36:00) But you could tell that it was good. So anyway, yeah, I just got off on a tangent there. But to finish the thought on why I think the open source and the revolution, why that's important. I guess I was talking about Sam Altman and how I liked him in the beginning. And I really liked Anthropic and Dario initially because I really believed him when he said that they were about safety, right? And over time it's it's
Michael Fafinski (36:26) Mm-hmm.
Reasonable Tech Dad (36:28) He's like every other CEO, they get consumed by the the board members and the money. And I do believe with the stuff that I've listened to that a lot of their p their fear-mongering and their announcements, even with the DOD stuff, have been about market capture and regulation capture because they are taking the old corporate strategy of if you can get the politicians on your side, you can f push out the little guys. And I think that what's nice about
the open source and the nature of our world right now is that those fires you were talking about and the revolutions that push those guys back is the only way we keep them from doing that. You know what I mean?
Michael Fafinski (37:05) Mm-hmm. Mm-hmm.
And you know, I I know Anthropic released something after they controlled the the the use cases of mable fable and and mythos. basically they got some bash lock on that backlash on that. And their response was, hey
Reasonable Tech Dad (37:24) Backlash. There go.
Michael Fafinski (37:30) We want to have control because we think that other people aren't to be trusted with this this model, this information more than us. Like i and and honestly, like it's partly what they said was we are God, we are AI God, and we want to be the only player because we are the only people that are gonna use this safely. And you can argue that a that's
Reasonable Tech Dad (37:39) Well said.
Michael Fafinski (37:57) moral in some sense because I think they kind of believe that, but also the other side it's like, okay, so you wanna be AI God. Like that's a little scary. Then if you take that power, yeah.
Reasonable Tech Dad (38:07) You you're the one? Where's the humility?
You're the one, you think you're the only one? The rest of us are too fucking stupid? Yeah. No I agree with you, it's scary, dude.
Michael Fafinski (38:14) Yeah. But it's true.
It's like you do need to control it. Like if we if they were just releasing it to anybody, like that's scary too. So th there's this middle ground. I don't think there's a good there's there's not a good answer, but I don't know. I mean I kinda trust anthropic, I kinda don't. I I'm just I'm just using the one that has the best model right now.
Reasonable Tech Dad (38:31) I'm in the same boat.
I'm in the same boat.
on that line, I'd really like to get your insight. What do you think about the privacy piece? Do you just let it all fly? I'm kind of Jesus take the wheel at this point, man. I
Michael Fafinski (38:45) I'm not good. I mean, it has my API keys. It's it's it's got everything. But I I am in the process of building my CTO. and then inside of my CTO, there's security regulations so that for the future I'm gonna be able to say, like, scan my vault, help me with security, or like, hey, I'm building the skill, let's make sure to use the best practices for security on this, and then I I'll get better. but right now I've not been good.
Reasonable Tech Dad (39:12) You let that bitch
fly, don't you? Fucking take check it out, Claude. Tell me how tell me what you need. I'll give it to you. you do you let it read all your API keys and do you ever get into like ENV files? Have you gotten that deep in your coding stuff? So your ENV is where you keep your API keys in a code base, like in inside of a repository and inside of a I mean a vault really at this point.
Michael Fafinski (39:14) Yeah.
ENV files, what's that?
Okay.
Reasonable Tech Dad (39:37) So when you make like a code base, if it needs to call Appify, if it needs to anytime you wanted to call a model inside of a code repo, you would have to put your AP. It's just a fac it makes your code more efficient so that you don't have to keep repeating your API. You put your if you need five different APIs for an app to work, you put the APIs in a in a.env and that ENV stays local. You don't push that to your GitHub repo when you commit. And then
Michael Fafinski (39:49) Mm-hmm.
Yeah.
Mm.
Reasonable Tech Dad (40:03) Basically you're taking a code variable and you're plugging that in every time. And the theory behind the code for a while was that if you keep ENV files away from the model, you're not sending your API keys to the model.
Michael Fafinski (40:15) Yeah. I I primarily keep it away from it. there was like there was one I I use Plawd. it's like an audio recorder, trans transcriber. And they didn't have like a native API key to use. So I had to like hop into the back end of like their website, grab it, and then put it inside. So for that I was like, all right, there's no other way. Like I'm not gonna figure this out like manually. so then it I let it have it. But like my QuickBooks, like
Reasonable Tech Dad (40:40) Let me ask.
Michael Fafinski (40:45) I'm not giving it like access to that one, so
Reasonable Tech Dad (40:47) Sure,
that's fair. I was in a repo. I well it was my brain. I don't need to sit here and play like I played it safe anymore. I used to try to do a lot of that stuff and it was telling me the other day because I gave it a rule to not look at my ENV file. And it told me Well, even if I did look at your ENV file, I wouldn't ingest the key. I'm like
Michael Fafinski (41:11) Even
if I did.
Reasonable Tech Dad (41:14) I'm like, I'm way too far down this road at this point. I I'm just whatever I've put in here. The only thing that really gets me weary and the only thing I really protect at this point is my kids' names and kind of personal stuff around that. I when I I try to I I like to build family schedule stuff and family meal plan stuff, and I like it I like that it can reference that when I try to lay out schedules for work task. So I talk about
Michael Fafinski (41:16) I ain't trusting you on that.
Okay.
Yeah.
Reasonable Tech Dad (41:42) Ages, unfortunately, they can probably still tie it to me one way or another, but I try to protect them as much as I possibly can. But what are you gonna do? We're down this road at this point, I feel like, with a lot of it. And a lot of people are look at that idea and they're like, you know, how could you? But my answer is how could you not at this point, you know, for a lot of it.
Michael Fafinski (41:53) yeah.
Mine
mine has my deepest insecurities. So like I it literally has a file on like me and my like my personal stuff. And I'm like, help me work through this.
Reasonable Tech Dad (42:05) Couldn't agree more.
But how important is that to get in your reels
w how impactful is that to when you get stuff done, right? Like it it tau it it it it it it gets so dialed in on how it helps you. I don't see any other way than to give it that information.
Michael Fafinski (42:17) It really is. It is.
Mm-hmm.
It it it will tell me like you are you are doing this because of your tendency to do this and like just start doing this framework and I'm like, holy crap, like it knows it knows too much, but like and that then I use it and then I can get the work done. Yeah.
Reasonable Tech Dad (42:40) If
I'm glad
you're in the same boat. I'm w I was wondering about that because I know how process driven you are. one thing my vault knows about me, and I'm so grateful that it knows this about me, is it's something I've labeled as the engineer trap. And it's something that when I get a new idea or I'm trying to get something done, I want to build something around it so that it's streamlined. And it it caught me and it saved me from hours of building, days of building.
Where I could
Michael Fafinski (43:09) Mm-hmm.
Reasonable Tech Dad (43:09) just simplify and validate the idea first. Validate that this is gonna be useful before I build the thing around it. Cause I used to spend crazy tokens building things to get ideas that were not validated, right? Versus just bringing it to a customer.
Michael Fafinski (43:26) Just wrote it down, man. That's a great idea.
Reasonable Tech Dad (43:30) And yeah, tell it it I think it understands what that term means, engineer trap. Tell it that you have a a tendency to to build things before you validate. And it's God, it's just saved me so much time. It's it's caught me so many times and say, Ryan, you're going down the engineer trap right now. I need you to just and it I think it's helped me be a better human in the loop. Because when I talk about business stuff, a lot of validation, you know, if I'm gonna try to sell something or if I'm gonna build an AI product for a lot of this construction stuff I do, I'll have this idea.
Michael Fafinski (43:39) Mm-hmm.
Mm-hmm.
Reasonable Tech Dad (44:00) And what I've learned is that I don't need to build the process. What I need to do is me, I need to get in I need to clarify my ideas in a simple way with the AI. And then I need to go talk to some people that don't do AI. And then they need to be in the field to tell me whether the idea is worth a shit. And if they tell me it's worth a shit, then I can come back and get the validation. Now we can build because somebody else has told me that knows more than me that it's a good idea.
Michael Fafinski (44:24) Mm-hmm. And when you
Reasonable Tech Dad (44:25) Because I'll get into that brain
and build it out and be like, this is gonna be great and then bring it to somebody and they're like, That's ship
Michael Fafinski (44:31) And when you talk about human the loop too, like I catch myself, it's like sometimes I'm terrible. I'm just relying on it entirely to make decisions and then like I'm in a different mood and I'm like, I'm firing on all cylinders because I'm making the decisions and I'm orchestrating. But like there's times where I just start relying on it too much, and then I'm reading every little thing, and I'm like, no, I can just I can go through this because I know what I need to do.
And then I n I know I'm like, I've been on here too long. If I start like not telling it what to do and start asking it too much for for its advice, then I'm like, All right, all right, I need to take break. I need to walk.
Reasonable Tech Dad (45:08) And then I think
you're on to something really important. What's your outp like when you get done with those modes where you're not making the decisions, how good is that at the end? Like what you've been trying to build or what you've been trying to put together in my experience is not good.
Michael Fafinski (45:24) I feel like it's not as good, but mostly it just takes so much more time. Because I do ask things in the way that it will give me the genuine opinion. And then eventually I come around to like what I think is best. But then I have to go through more prompting and more reading. And I'm reading this entire thing, multiple prompts, when I could have just said, hey, do this, because that's what I know is best.
Reasonable Tech Dad (45:47) Yeah.
Super well said. Correct.
Michael Fafinski (45:50) And then I just confused myself. but
I think at the end of the day, the quality is probably pretty relatively the same.
Reasonable Tech Dad (45:58) That's a good point. That is a good point. There's a lot of times where I'm in that mode where I'm letting it do a lot and I'm I'm I'm overprompting and I'm kind of letting it when you let it build the context, it's like, shit. Now I got three more pages. But one way I've looked at it is that I wouldn't have gotten shit done in that mode anyway. Right? I would I would have been sitting here fried and that would have I would have gotten nothing done anyway. So it it it carried me a little bit there and I'm all right with that.
Michael Fafinski (46:08) Mm.
Yeah, yeah, yeah.
Mm-hmm.
And you said the br like context in itself. And I was I was right now, like I said, I'm in the middle of my my strategy review for for quarter three, and I almost told it, load in whatever context you feel like is needed for this. And then I was like, wait, dude, this is my strategy. This is my entire business like for the next three months. Yeah. Like you're just gonna let it do whatever. Like it's it's not there yet. Like
Reasonable Tech Dad (46:39) I've done that.
That's the way to do it, this motion right here.
Michael Fafinski (46:53) So then I backed it up and I said, do not add anything without my approval. I want you to suggest things, but not add and specifically look at these four files or folders or whatever. And I do want you to use these. And then now I'm like, all right, this is gonna be a way better strategy creation because it's not bringing in context that's not needed. Because I know my vault and I know what's needed, I know what's not.
Reasonable Tech Dad (47:15) Yes. I kill a large number of sessions where this term I heard I I don't remember exactly who I heard it from. I think Andrew Ng maybe. context pollution. If you start getting down that path where you keep where you start asking it that exact thing, go get whatever you think is best or grab whatever you think you need from this, you know, piece.
And I really watch context a lot. I really watch the context window. Like when they started doing the one million context, I still didn't let it get past quarter million. Because if you let it if you if you let it think about too many things at once, it really starts to sway and waver and bring too much bullshit into what you're working on now. It's still very important, even with big context windows, to keep your sessions focused, I think.
Michael Fafinski (47:44) Mm-hmm.
Mm-hmm.
And the beauty about being an obsidian with it is you can just build that out in that folder that you have. Like I have project templates or I call like AI project templates is a different one where it's like building out something for my vault, right? And so I can there's a s there's a template of like a status, a plan, specs, design. And so then it just I can just say, All right, I'm about to delete this chat, put whatever context you feel like is needed for this next, you know.
task that we are doing in the in the project and then it updates everything. I load in a new window and it's back to to you know 20% context and I can and it's it's performing well. And that's the beauty of obsidian.
Reasonable Tech Dad (48:47) That
That's probably the key point I think anybody listening. I we have gone on some really super geeky tangents and I know there's people that are like, What the fuck are these guys talking about? if there's anything they're like, dude, well, I think I don't even know that. I think anybody I think the only way to get where we've gotten is to spend a lot of time with the models. 'Cause the i and if anybody that I had a friend text me today, he's like, Dude, I've been getting into the AI stuff and I'm so far behind. I'm like, You have no idea how far ahead you are. People that
Michael Fafinski (49:00) Only computer science people are gonna be able to understand this point.
Yeah.
Uh-huh.
Reasonable Tech Dad (49:21) Think they're behind when you get into AI, it's the first thing you're gonna start feeling is that you're behind and you're not. There's still people, I would say ninety-five percent of people still use these things to check the weather and they use it like Google search. And if there's one big takeaway from this podcast, anybody listening, please understand that learning about obsidian and learning how to build your vault and your second brain, that is a crucial, crucial skill right now.
I would very much and what's nice about it is that you don't have to know any of what we know. You can ask it. Ask it how to use the vault, ask it how to build it, and ask it to keep it simple. Don't let your vault turn into something that you don't understand. Because if you don't if you keep your vault inside like you have a feel for it, then it is a super, super powerful tool that keeps you above water. And I think it keeps you caught up. It's I don't know how you I could.
Michael Fafinski (50:05) Mm-hmm.
Reasonable Tech Dad (50:19) Be f I went to take out the garbage and I was walking down the hallway here in the apartment getting ready to do this and I'm like, one thing I gotta tell him is that you can tell me how you feel, but I don't know if I could go back. I don't know if if all this shit blows up, I don't I don't know. I don't know how I'd get work done again. Writing in a notebook and keeping my ideas like I used to and having notebooks and like, you know, files and books, like books not in PDF.
Michael Fafinski (50:34) I can't go back to normal work. Like Yeah, no.
[conversation trimmed for print]
Reasonable Tech Dad (50:59) Those PDFs are critical though. Cause I have mountains and I mean in my storage unit I've got four totes of books and I'm like, I wanna if there's not a PDF available, when I did this store when I started doing the project management thing, I had an old estimating class from the lumberyard that I took. And it was five chapters of table span or like span calculations and I ripped two weeks ago I sat down on a Sunday night for three hours and I pulled out my scanner and I scanned no shit, five hundred pages.
Michael Fafinski (51:00) Я
Reasonable Tech Dad (51:28) and made it a PDF to put it in my vault. And it is now one of the most powerful parts of my vault. I would not have been able to do this new project manager stuff had I not put that PDF into what I'm doing.
Michael Fafinski (51:42) I one time had to go, so I I bought a book called Top Gradients for hiring. And I had to go two pages at a time and because there's like a certain amount can copy and copy and paste and copy and paste and copy and paste. And then you hit your limit for like the day. And then the next day you come back and you copy and you paste. You copy and paste. And I did that for like two hours. But I was like, I know that like.
Reasonable Tech Dad (52:01) What page was I on?
Michael Fafinski (52:09) I need this in vertical or in virtual, or else it's not gonna happen. I don't I can't train my system on it. And I'm also gonna push back on your organization of Obsidian. because so I think it's important to keep it organized, but I think it's more important just to jump in and do it. Because it was about two weeks ago where I before I had my trained like CEO and all this stuff.
Reasonable Tech Dad (52:15) No. I I
Please do. Please do.
Michael Fafinski (52:37) I did not understand my vault on a high enough level, but it was still insanely useful. And then I revamped it. And it's just the constant revamp. So it's like, it's okay if it's not fully structured because you're not going to be able to make it perfect off the first week. And it's just every week just figuring out how to improve it and what folders to add and to delete and and to make simple and and all that stuff. So like, yes, keep it organized, but also it's like, it's just just having it is like 90% of the battle.
Reasonable Tech Dad (53:07) I think that's a much better way to say it. And I stand corrected for sure. I was just trying to help I felt like we got so far down the road I was trying to help people like don't get overwhelmed, but you're right. Just do it, like anything else. Just tear in and let it and ask keep talking to it. Tell it what you need. Tell it where it's not working. Tell it what you don't like. And it'll help you fix it. And you're right. It's right. It's never gonna be perfect. My vault's not perfect. My vault is not the way I would like it right now. But so that's a good point.
Michael Fafinski (53:12) Yeah. We did.
Mm-hmm.
Hey, another
another night of me building stuff in there and it might be one percent closer to perfect, all right? That's what I keep telling myself.
Reasonable Tech Dad (53:41) And I would happily spend so many nights like that if I could. my happy place. Fire away.
Michael Fafinski (53:46) I do have a question for you though.
When do you think that the non technical user is going to be able to utilize AI the same way that we are now?
Reasonable Tech Dad (53:58) I think that's a long way out. I think it's a really long way out. I think that there's I'm I'm I'm probably the wrong person to ask, right? Because I go back like we were just talking about with cowork, right? That that's a built for your average Joe. And me and you, I I wanna throw up when I use a low like I for me writing all the code I've wrote and going back and using like a no code tool or the stuff that is built for mainstream, I can't. I'm like this is it it just I feel like I'm my hands are tied.
Michael Fafinski (54:18) Mm-hmm.
Reasonable Tech Dad (54:26) So I'm probably the wrong person to ask. I don't know. I don't know when they're gonna do that. I know it's hard. I know it's a super hard problem to solve. And that doesn't mean it's unsolvable, but I think more than anything, the only way that we're gonna get there, and it's mandatory, is that people gotta stop being scared of it, man. Have you ever looked at the numbers of polls on how people in this country feel about AI versus like China? How the sus
Michael Fafinski (54:52) I got a feeling,
but I've not.
Reasonable Tech Dad (54:55) People in China because I mean, now anybody can argue that communist countries are brainwashed to a certain point, right? They're limited on what they can see and what they say, but they feel optimistic about what it's gonna do for their culture as a whole. And I think one of the surveys, 'cause the all in podcast, you l you listen to those guys once in a too, right, haven't you? Paul all in podcast, or have we talked about that?
Michael Fafinski (55:06) Mm.
I don't think I've listened to them at all.
Reasonable Tech Dad (55:19) No,
they're they're a a venture group that four venture guys that I like because they have very different opinions and they bring it out on the table and you they really they talk about geopolitics and economy. It's it's cool because I think that they bring different views and they argue cleanly about it. And they brought one of these poll things in that said people think better of ice as a country than they do of AI. And that's saying something, right? About in our country in particular, that's like whew.
Michael Fafinski (55:25) Mm.
Yeah.
That's yeah.
Reasonable Tech Dad (55:48) And until
until and I and I and we can go back to the CEOs things and and when we were talking about market capture and and how I feel about Sam Altman and and Dario and how our trust levels have changed for him. And Elon and Elon and how people think about Elon. It makes me sick because if you've read his book and you've ever listened to anything that's came out of his mouth, he has never changed in two decades. He's never changed. He has always been about open truth and open transparency. You know, he's still a CEO.
Michael Fafinski (56:09) Mm-hmm.
Reasonable Tech Dad (56:16) He still has IPOs to sell, he still has a business to run. So there's times where he's touting his business card, but by and large, he keeps humanity in the front of his mind and what's good for the whole. And I think
Michael Fafinski (56:31) I truly
believe that. Like I and and you could argue that, hey, I disagree with some of those ways that he's doing it, but I truly believe he's acting selflessly in the sense of humanity, not for his own monetary pleasure. I mean, I think sometimes that might happen, but like I mean he doesn't really
Reasonable Tech Dad (56:50) Comparatively,
comparatively to the rest, it's not close.
Michael Fafinski (56:53) Yeah,
not close. And he doesn't buy stuff like that. Like
Reasonable Tech Dad (56:56) No. Dude, the yeah, we could go down the long list of sleeping on people's couches and selling his stuff.
[conversation trimmed for print]
Reasonable Tech Dad (57:44) But to to answer you to just wrap up on your question, and I'd love for you to add more on what you think, but when are mainstream people going to start using AI like we are? I think it has to happen because it's important and I think it's useful. I think at the end of the day it's a good technology because any technology that improves our productivity helps us increase the size of the pie.
Michael Fafinski (57:47) Yeah.
Mm.
Reasonable Tech Dad (58:10) Which is what capitalism is about. That is how we've killed poverty across the globe. That's how we have made the country that we have where things are so abundant. It's because technology and cre increases productivity. And I don't think that we're gonna get there until the messaging and the media stop destroying it and scaring people. Of the thing we were talking about with it's so over and this pe you know, there's just and it's unfortunate because the media is incentivized to get clicks, which is always through fear.
Michael Fafinski (58:39) Mm-hmm.
Reasonable Tech Dad (58:40) And bad
things. There's nobody not nobody. There's a lot of people out there that are trying to preach the optimism, but it just gets drowned out. And I think humans by their nature are not optimists, right? Most people are we're we're biologically programmed over how many millennia to find the bad so that we stay alive. until people start thinking that AI is a good thing, we're not gonna get there.
Michael Fafinski (58:58) Mm-hmm.
I think it's hard because
You gotta watch what's in the in the vault for context. And I don't know how a non-technical user could truly trim and and and keep only what's important in there. and and I I talk to even business owners at times where they're looking for more of a a white glove glove solution to insert it into their business. But
Truly what it is is just building documentation for AI over time. And I have no idea how you would do that with a non-technical user. Like I I mean, maybe the model could get really good at well, I don't because everybody's life is so different and what information they want in there is so personalized. So I think we're at least five years out to there's like a there's a software or some.
wrapper or something that you can just talk to it in your car or on your phone or whatever that is gonna get a good enough idea and be able to really be productive from it. But I think unless you're a technical user, you're never gonna truly be able to utilize the the full capabilities.
Reasonable Tech Dad (1:00:28) I think you bring up a really good point that triggered a thought while I was listening to you, but I think the voice having something that people can talk to and that's really good that taking the conversation and and talking to people in a way that they understand or that they can help clear up their vault until we have a a voice system. and this idea just came to me while you were talking. people learn in different ways. I think that's pretty well documented. the only way when you said trim your vault, that's for somebody who likes to read, like us, right?
Michael Fafinski (1:00:43) Mm-hmm.
Mm-hmm.
Reasonable Tech Dad (1:00:58) People that
like to read book you know, and they I I think it's do you really? That's surprising to hear being the process in the book a interesting. Cause I think about learning types a lot with my kids 'cause I I'm really deep in trying to figure out how to prepare my kids for what's happening. And that's a whole nother fucking episode we could do.
Michael Fafinski (1:01:00) I hate reading though. That's the only part that I don't like about this, about the the Obsidian Vault.
Mm.
Yeah.
Reasonable Tech Dad (1:01:21) I think that it's important to understand that people learn in different ways. And I think that's well documented. Like musical types, artistic types don't do books and logical thought. There's different brain styles out there and we're wired different. And LLMs are 100% text. Lacoon is his last name. The old guy, he he used to be the head of AI at Meta.
Michael Fafinski (1:01:33) Mm-hmm.
Reasonable Tech Dad (1:01:45) And I liked I he kind of fell off the map for a while, but for a very long time I really liked him because he kept telling people for like two years after ChatGPT came out that LLMs are not the answer. And I think that they've came a long way, but I think that he's still right in that the it it's not going to it's not gonna proliferate across our population because not everybody is
Michael Fafinski (1:01:46) Okay.
Reasonable Tech Dad (1:02:10) wired to take in text and talk in words and and think in text and write, you know, and it's just written language is what LLMs do.
Michael Fafinski (1:02:20) Do you think there could be a wrapper that makes it usable for those people in visuals and vo and voice and stuff like that?
Reasonable Tech Dad (1:02:25) I hope so.
My optimistic side that th thinks that there's super smart people out there that think like we do and I hope they figure it out because it's it's important.
Michael Fafinski (1:02:36) I don't think we're far from that. Like I I think I was talking to a buddy about this, but I think all that's needed is a wrapper that is so like whisperflow. You know, like I mean, I know you use that. Nobody that is in code doesn't use it. and then speechify, right? And those are broken systems in some sense where I need to copy the text and then hotkey and then it talks to me. and then whisper flow is I can just tell it stuff. But if you put a wrapper on top of it.
Reasonable Tech Dad (1:02:47) Yeah.
Michael Fafinski (1:03:05) Where it can take my voice and and text, understand when I'm like, mm, scratch that, or understand when I'm saying I'm I'm pausing. and then there's a, you know, a a command to end the the voice. And it can control multiple context windows. I think that's all that's needed. Like open up context window one. Where are we at with that? And then it tells me, and then I can go, okay, context window two, what do we need to free you up on so you can keep going?
And so I don't think we're too far. I don't I mean, I don't know that it'd be that hard. I there's a software called Hey Clicky that I've experimenting with. It takes control of your entire computer. and it's not the best currently, but I'm maybe I gotta get the pro version and figure it out. but it's I think it's the beginning. so I can do a hotkey on my keyboard and then it puts an agent off in the corner.
Reasonable Tech Dad (1:03:55) What do you mean takes control of your whole computer?
Michael Fafinski (1:04:05) It could be like go to the Google Docs, find this, and then make it into a PDF for me and put it in my downloads folder. And then it'll go put it in the corner of my computer screen. And then the next one will go right under it. I can have multiple of these agents running at once. Honestly, it fails a lot still. So it's not like perfected by any means. but you know, if I could have it just look at and be in my obsidian vault and click between different
Different chat windows and then give me an update of what's happening in there. Like that's all that I need. and so I I don't I think that'd still be clunky, but I I don't think we're too far away. Like I'd say like six months. And at that point, that's Jarvis. Like if you can say, open up this context window, start this, and then it notifies you that another window is completed, like that's that's Tony Stark at that point.
Reasonable Tech Dad (1:04:45) I hope you're right. I hope you're right. 'Cause
And I think
you you asked the right question in when you asked me how long until the non technical user is in, 'cause we need in. We need we need the collective. We need those people that don't think like you and I do to make it work for everybody. And that's until it works for everybody, it's it's cool for me and you and but that's not what we want. Going back to the open source idea. We want it to be f for everybody. You know what I mean? That's it has to be. Otherwise you're gonna have isolation and, you know
groups of power that, you know, pin other people down and that kind of stuff. That's not America. That's not freedom if everybody doesn't have the same tools. If we don't all have freedom of speech, what do we got?
Michael Fafinski (1:05:30) Yeah.
And that's what I'm truly scared of is really people not taking the opportunity, not understanding it. And then 20 years down the line, up there's to be a poverty gap purely based on AI. And at that point, I mean, we saw a little bit of controlling who can use which models, like the the government saying no foreign employees can use it, and then obviously they have to shut down the entire thing. And so
Reasonable Tech Dad (1:06:00) When they shut that down. That's yeah.
Michael Fafinski (1:06:04) Controlling who can use the best models at a certain point, like if that gap gets bigger and bigger and bigger, like that's that's that's a poverty gap right there. Like, and so that's what I'm scared of. I have no idea what's gonna happen with it.
Reasonable Tech Dad (1:06:09) That's bad. That's bad.
It's there's scary ideas out there. I'm I when people ask me about this, my new project gig, I've been there's a couple of younger puppies in there and they right away when I started doing the AI stuff, the office manager's like, I'd like generated this sheet to take out on a bid and she goes, my god. She's like, You just she's like, You don't even know what this stuff is and you just made this happen? I'm like, Yeah. And she's like
Michael Fafinski (1:06:42) Yeah, you don't even
understand fully what this bid is supposed to be.
Reasonable Tech Dad (1:06:44) I went on a gutter bit, gutter boy, how but it was a gutter bit,
and I printed up the measurements and stuff that I needed and the questions I needed to ask just by asking Claude Coat. and it made the PDF in such a nice format. It was like right in line. It had it in sequence of how I needed to ask it. And she goes, my god, AI's gonna take our job. So I I tried to, in my most soft, open-minded way, try to like tell them where I'm at on it. And the term I always bring up to those people that really resonates is
The Jurassic Park line, like life finds a way. You wanna, you know, no matter what these these the those poverty gaps happen and scary stuff. I I think by and large, the optimistic side of me there goes my headphone here. the optimistic side through all of these. Have you ever looked at that? Like when the cotton gin was implemented and some of these technology like the way people have revolted when new technology has disrupted things. people find a way.
Michael Fafinski (1:07:37) Mm.
Reasonable Tech Dad (1:07:41) At the end of the day we end up coming together and making it work one way or another. And there's there's there's disruption and there's there's messy spots, but we figure it out.
Michael Fafinski (1:07:50) I really try to be the link for a lot of my friends to like bring them into it. Because I'm at that age where like my age group should be really utilizing it, but a lot of people aren't. And they're like, I don't like AI. Like, I I I I don't I don't like it. I think it's gonna take our jobs, so I don't use it. And I'm like, okay, th honestly, after you dive in for 40 hours, you then you then you understand it's not gonna take the job, it's gonna leverage you. And so with anybody who I like says, like, I think AI is gonna take our jobs.
Reasonable Tech Dad (1:08:15) Correct.
Michael Fafinski (1:08:19) I really try to softly, like you said, kind of reframe it and be like, Hey, like once you dive in, you understand it's not going to take your job. It's just going to be who's going to get hired and who's not, because who's leveraging it?
Reasonable Tech Dad (1:08:32) The the kids I was talking to, I say kids, but the younger generation, they are your age. They're they're mid mid mid and early twenties. I brought up the cat 'cause she was she was taking a sti I let her like tell me why she was scared and she comes from a family from the education system. And I softly brought up the point about, you know, if you take a s she was bringing up specific details and I let her talk about it, and I said, if you take a step back and if you look at how the rest of the world does, how is the US education system comparatively, right? Is it is it pretty good? She's like, Well, no.
Michael Fafinski (1:08:36) Yeah.
Yeah.
Reasonable Tech Dad (1:09:01) Yeah, no, you're right. So then like I and then I went on the example of because I asked them how much long division they did as a kid, and I tried to bring up the calculator thing, because that's what they said in the sixties, was like calculator's gonna make us all stupid, right? People aren't gonna know how to do math anymore because we have calculators. It's like, no, there's still people that do fuck pretty good with math, like So the conversation ended up in our right spot. They've they've let me use my tools without talking that way. So whether I got across or they just don't want to talk about it, who knows? But
Michael Fafinski (1:09:03) Yeah.
Mm-hmm.
Like, can we just get this guy to shut up?
Reasonable Tech Dad (1:09:33) Right, exactly. And I
did tell him I'm like, 'cause I was talking about my kids and I think that softened her up a little bit. I'm like, you guys are the generation that's had to take the lump of it. Like when you talk to your friends and the people that hate it, you guys how you guys and the the people that are, you know, eighteen to thirty, you guys are gonna determine how my kids get there because you guys are gonna teach
Right? Who are they gonna learn from? They're gonna watch you guys. And then and however you g however things land by the way you guys handle it is gonna be how my kids that's what they're gonna take in. It's intense, man. It's intense.
Michael Fafinski (1:09:57) Mm-hmm. Mm-hmm.
I
I've already known like my kids, I I first gotta get a wife or a girlfriend first and find love. I'm still still working on that. But it's not going well, but it's going, all right.
Reasonable Tech Dad (1:10:12) Sure. Baby steps.
My vault girlfriend.
Michael Fafinski (1:10:26) There's a sp specific file that after after 5 p.m. I go into it. No, but my kids are they're they're gonna get a full like they're not learning from their school about AI. Like they by the time they're like 12 years old, they are gonna be more proficient than like people in the field. Like I want they're gonna be technical users because why why the hell not? Like
Reasonable Tech Dad (1:10:28) That one's got a lock on it.
That's good. What do you think about it? What do you think about the kids, your kids?
Michael Fafinski (1:10:55) If if you train them the right way, then they're using it as leverage, not as to not think. So like I'm gonna help them build something that to do their homework for them. Like I that's not gonna be a popular opinion, but like if they can build something to do their homework for them, but then I also make sure they're not being lazy with it, but they're actually getting better homework at the end of the or a better product at the end of the day, I'm fine with that. Exactly.
Reasonable Tech Dad (1:11:19) And they're the one thinking through the build, yeah. And w and so,
gosh, we like I said, we go down this. We'll do I mean we're already over an hour, so I do wanna so how would you do that? How would you think you do that? Because I have let Claude do a little research for me. and they say at my I got my kids are seven and four, and they say at that young of an age, you're not you should really just let them kind of see
Michael Fafinski (1:11:26) Yeah.
Reasonable Tech Dad (1:11:44) How it works, but don't let get too far into the weeds with it, is the simplest way I can describe it. So I don't know when I start letting them build. I think at seven I should start. I've seen some videos of kids coding, you know, and making making simple apps. And I've seen some really cool stuff on X where you can make apps that learn allow them to learn instruments faster. To sit on a keyboard and and have it teach piano. That's the piece that I'm so intrigued with is
Michael Fafinski (1:11:48) Mm.
Reasonable Tech Dad (1:12:13) is they say it can revolutionize where every kid has a tutor. That's what J Jensen's been talking about for the longest is that think about a world where every kid has a teacher that is infinitely patient and completely understands the kid and teaches them how they want to learn. Going back to what we were talking about and getting it mainstream, right? Not everybody thinks the same way. So I think it's very important that kids learn their way. And I think that I'm very optimistic and I'm trying to figure out how to build something like that for my kids. That they have that personal tutor.
Michael Fafinski (1:12:26) Mm-hmm.
Reasonable Tech Dad (1:12:44) 'Cause I'm being honest, I'm not qualified. I'm not qualified to teach my kids the way probably that they should learn the best. So I'm trying to find a better way to do it.
Michael Fafinski (1:12:51) Mm-hmm.
And I I think that starts with building stuff for them, telling them how to use it, and then they can use the tutor. Like I mean they don't need to build it themselves the first shot, but if they have a specific issue, it's like let's pull in this book and and and then train them on this. And he's like, Hey, you can start just asking questions and see what it has to say and and learn from it. And then I think it's Yeah.
Reasonable Tech Dad (1:13:13) Yeah. Grok has a kid mode I've let play with a little bit that's pretty fun where they just
do animals and stuff. That's pretty cool.
Michael Fafinski (1:13:20) Yeah. You gotta make it fun for You know, I I think, you know, building apps, building little stuff for like maybe their Minecraft world or whatever, like I think it's really important to like, hey, you know, encourage them on projects like that, but obviously they don't care about Alex Hermozzi's hundred million dollar leads. Like they should. My kids will. All right. At eight years old, my kids get trained on Alan Hermosy's hundred million dollar leads.
Reasonable Tech Dad (1:13:22) Exactly right.
Yeah.
They should. They sh they should
dude. Have you gotten into money models?
Michael Fafinski (1:13:51) I have gotten to it a little bit. Nothing crazy yet. Honestly, it's a little discouraging. My business model, I feel like, doesn't allow for some of the concepts that he's saying in there because it is much larger. it's truly like a hundred million dollar money models, in my opinion. So there's some aspects I'm gonna take from it, but some of it it's like
Reasonable Tech Dad (1:14:13) I pre-ordered a copy because I'm a fanboy and I haven't opened it.
I don't want to go down that far yet. I pre-ordered just I just watched the him set the Guinness World Record and all that and Ye Thanks buddy! That's what I was looking for! That's what I was looking for. I'm like, damn, I got this hard copy. Sit that till the day I sit down and scan those five hundred pages. Like
Michael Fafinski (1:14:19) [conversation trimmed for print]
Yeah. I still will buy
a book though. I I put it on the shelf. Like if I'm using a framework or I'm like actually utilizing it, I'm like, I I can put this on my bookshelf and brag about it. Yes, exactly. Exactly.
Reasonable Tech Dad (1:14:42) Support the author. Support the author. That's how I've been with Elon. Like
I don't use Grok, but I'm gonna pay the subscription because I support Elon. Alright, man, we should wrap it up. We'll go all night, dude. this was amazing, dude. Thank you so much for coming on, as always. I don't know what if we had any process to this or if we have any good lessons besides go learn Obsidian. Go ask your AI. Try start using it, people. Start using the AI. Just just tear in, like he said.
Michael Fafinski (1:14:48) Mm.
We should. Thanks for having me.
Yes.
Reasonable Tech Dad (1:15:10) Forty hours, you could do that in a month by just sitting down with the weekend. I've seen some good ex posts like, rather than watching a Netflix tonight, spend two hours watching this tutorial and just talk to it. Just ask it questions. Just be curious. Just be curious. And if you don't like typing, find one of these voice apps. Aqua voice is a great dictation tool I've used. Super accurate. It builds kind of how you talk. So it picks up more of your lingo and it's easy to talk into after times. So what else you got for the people, man?
Michael Fafinski (1:15:41) I'd say if you genuinely enjoy learning, if you're a person that enjoys learning, setting this up is going to allow for rapid learning in other topics, but also in AI, and you'll have a lot of fun with it. So if you genuinely enjoy learning, you f 50% of people I think are gonna genuinely enjoy doing this and setting it up. The other 50% probably won't. But yeah.
Reasonable Tech Dad (1:15:57) Yes.
And I'll add on that
and say, even if you think you don't learn like learning, you might like learning if you use AI. It's amazingly easy to learn with AI. Like what as someone who enjoys learning, I feel like I'm just learning faster. I can learn more than I used to be able to learn. So if you don't like learning, see if you can learn with AI just by talking to it. Just be curious to see how it goes. So Michael, always a pleasure, dude. We'll talk again soon. Appreciate everybody listening out there. Talk to you guys later.
Michael Fafinski (1:16:24) Mm-hmm.
Mm-hmm.
Peace. I'm gonna go talk
to my AI girlfriend. See ya.
Reasonable Tech Dad (1:16:39) That a boy, yeah, no more lonely. Fix that for the world next. Haha Later everybody.