130 Job Search Update

Hello, and welcome to makers.dev episode number 130. Chris, did you know that 130 is a sphenic number, which means it's the product of three distinct prime numbers? And I swear to God, if you say yes, I'm going to say, no, you did not know that. I have no idea what a Svanik number is or that 130 was one. That's cool.

There we go. Something else that's come up multiple times in the past also is that it's a palindrome in multiple bases. I like that fact, and I'm not sure why, but in base five, it's one zero one zero, which is not a palindrome. Is this another AI fact? Yeah, it's from Grok. In base 12, it's AA, which is a palindrome. But now I'm even doubting that. But the math looks like it checks out.

Anyway, those are my fun facts about the number 130. How's it going? What are you thinking about? Oh, it has been a busy few weeks. Despite that, I was like, you know, I'm going to set up my microphone and everything. And then I pulled out the cord and this is the last bit of the cord of my microphone. That's why it didn't work two weeks ago, I think. Anyway, that's about been how the last two weeks have gone is stuff like that over and over.

So you get me on my hopefully not too bad headphones today. But we're doing all right. uh the let's see my kid has a band uh concert this week um the the band teacher says um the concert will be 15 minutes uh you are welcome i like that line um Yeah, so they do one band at a time, so it's short, which is nice. Yeah, otherwise just chugging along. Are you familiar with the musical The Music Man? Yeah.

Do you know at the, at the end? So for anyone who doesn't know, it's the shyster who comes in and doesn't know how to, how to play music, but he's, he convinces the whole town to like give him a bunch of money to buy their kids instruments and everything. And it's just this, it's just this freight train barreling towards this. obvious catastrophe of like, this guy doesn't know how to teach.

the kids and you know they have the concert at the end of it and it's exactly as terrible as you would think it is uh after this town with like spent all this money and then there's this moment to pause and then the whole town is like oh my gosh that was amazing yeah yeah yeah

Brilliant, brilliant story, I think. But they did have a first concert in the fall and it was actually better than I thought for only like, you know, all the kids only had like two or three months of practice at that point. And they played, you know, reasonable music. So yeah, they, they do a good job. I think despite being, I mean, maybe that's my parenting brain taking over, but yeah, it was okay. I feel it. Isabella can play the kazoo now and the, and the recorder. She's a genius.

I've never felt so proud at what is objectively just a terrible noise. But she's so happy she gets a smile on her face when she does it. And it's objectively impressive to have the coordination of the skill to be able to like... on the kazoo you have to like learn to blow and come at the same time and yeah i'm so proud how old now one year and a half well year and a half yeah yeah yeah she can like walk around and do it and yeah that's good that's great there's nothing like it

I totally get it. Let's talk about dustblowers for a second. Yeah. I have this. This is...

Something that we talked about and thought should exist. And they do. This is a mini dust blower. It's the size of like, I don't know, like a thick cell phone or something. It's not big at all. It's so much smaller than I thought from the pictures. Yeah. Yeah. Yeah. And it works really, really well. It's different than the leaf blower. It is a smaller jet of air. And it's not as powerful. It doesn't put out as much thrust or whatever. But given the size, it's awesome.

I have been using that, especially for we have these Legos that are out on like shelves and the leaf blower kind of makes a mess of them. And it's like.

cumbersome to use and the vacuum i don't want to suck up legos and so i've been using this and it comes with a little brush and i got all the dust off my legos and so anything that's like a lego or a lamp or you know something fragile uh now i use this thing with For anyone not familiar, my method is get two or three HEPA air purifiers in the room at full blast, vacuum, then do a...

round of dust blowing or a couple rounds, then vacuum again. And I could say, you know, a good chunk of the dust. Your technique has permeated my entire extended family now. My dad has a phrase now, dust blower clean, to me.

We have dustblower in the room. Yeah, it's brilliant. I'm surprised I hadn't seen it before. Yeah, so that was the question I had is, because it's so small, what are the sorts of applications you use? Because the places where I use the dustblower now are like, if there's enough stuff on the floor to...

like that otherwise you would spend whatever 10 minutes sweeping it out. You can get it done in like, you know, a minute and just blow it all out the door. Um, but that looks like it's too small to do a whole room. So yeah, for, for smaller things where the desk blower would be too powerful. Lego sets.

make a lot of sense. That's a, yeah, that's a, that's a great application. And I love the addition of the HEPA filters because yeah, otherwise the dust just goes off of the Legos. Yeah. Yeah. If anyone does this, definitely use the HEPA filter. Yeah. The technique. Yeah. Love it.

Yeah, that's great. Oh, you were going to talk about allergies also? Yeah, yeah. So also especially relevant. One of the reasons I was, I saw these and I was like, ah, do I really need that? But my wife got allergy tested for the first time in her life and she knew she was kind of bad, right? She's allergic to...

everything in the panel like basically everything in indiana she's allergic to um especially the worst was birch trees which is okay um but yeah dust mold dog cats which are the biggest indoor ones um and they're to varying degrees. So most things she's like a three out of fives, like a moderate allergy, but suck up a few things. She was four and five. So, um,

Yeah. So she's going to get allergy drops, which I didn't know were an option, but you can get shots or drops. And the idea there is that you get exposed weekly to a small bit of the allergy or the drops are actually every day. Shots are weekly. Drops are every day. um, for a few years, like, and you ramp up the dosage and the idea is eventually your system hopefully stops responding to the, uh, the allergen. Um, and then you're, you know, some people occurs completely. Some people.

get moderate improvement some people it doesn't do anything for you just have to kind of try it and see um so yeah she's gonna try that um but until then i'm doing more cleaning around the house given that yeah dust was like the second highest allergen so I can totally relate to Brittany. I've struggled with seasonal allergies most of my life, and I tried the drops and had moderate success. I didn't do it, I think, as consistently as I could have. You have to do it every day for like...

four years or something. Yeah. Yeah. I did not do that. The, the, the system of allergies also doesn't make any sense to me of like your body's reacting to this thing. And so to cure it, we're just going to give you more of the thing that you put under your tongue or we're going to shoot it right in your blood. Like, how does that work? My understanding is it basically trains your body that this thing's not going to kill you. So stop reacting to it. That's my understanding.

supposedly it works for some people it did work so my nephew was very allergic to peanuts like anaphylactic allergic when he was a kid and they had just started doing this for peanuts when he was born so it's pretty new i think because they didn't want to like kill babies but they basically get started with like you know just like running a running peanut in the mouth and then eventually to eating it and now he eats like a handful of peanuts a day

um, every day forever for the rest of his life. Um, and he's, he's basically not allergic now as long as he, you know, he can't, they, they, they're still like, you know, probably don't eat a whole cup full of peanut butter or something, but he eats his little spoonful of peanut butter or peanuts. And, uh, It seems to be okay. Huh. Wild.

It's the sort of thing that with AI getting into medicine and research, I would hope that there would be a better like a gene therapy. Okay, just take this one shot this one time and now you're no longer allergic to anything. The thing that really helped me was I went to go see an allergist and... I was like, yeah, I have these debilitating seasonalities. For me, it was a Bermuda and Timothy grass, which is all over this part of Texas.

I got the same allergy test and I thought I was going to be really allergic to, uh, uh, cats and dogs, but those, that was like a one out of whatever the scale is one out of five. Uh, but Bermuda and Timothy grass were like a four out of five. It was, uh, uh, super reactive to that. And the allergist said, well, what do you take for it? And I said, well, sometimes I take lorotidine, like an allergy pill. And sometimes if it's really bad, I'll do flonase.

And he said, oh, how well do those things work? And I said, well, the allergy pill works okay, but it's kind of delayed. And then the Flonase usually works right away. Like if I take that, then I feel really good. And he said, oh, we'll just take that more. And I said, what? You can do that? He said, yeah, do two shots in every nostril every day. And I said, oh, because it's a steroid. I thought that would be bad. And he was like, nope, just do that. And so I did. And then it went away.

Yeah. Uh, now it's so, it's so strange. I kind of model it as like this runaway train that when I feel the first inklings of that, I start to feel sniffly or the itchy eyes. If I can just.

stamp that out right away and go straight to, okay, I'm going to take Flournaise until I feel nothing. Then I can go the rest of the spring and summer with almost nothing. Yeah, it was miraculous. I think the only thing I needed was just... permission to medicate myself more than i was yeah a lot of medicines this is not medical advice but a lot of medicines are purposely dosed low pretty low um and yeah you can kind of double up on them

under a doctor's orders. Yeah. Yeah. And then the other thing I was going to say is, so my wife, in order to get these allergy tests, had to stop her antihistamines like 10 days before or something to test. And she's still trying to recover from that. Like you said, it's like a train. She was off of them. It was awful. And she's better now, sort of, but not like she was. It's like you can't get caught back up after it goes past a certain point.

Yeah. Yeah. Yeah. I've got to, I've got to like just hang out in the shower and yeah, it's for several days afterwards I'm, I'm recovering from it. And there's, uh, I drops my sister told me, but also where you, you.

put them directly in your eyes for the, for the itchy eyes. So I have like these three pronged attack instead of, uh, counteract my own aggressive immune system right when that when that happens now but yeah allergies suck colors my entire world of just i don't want to do anything i don't want to work on anything just i feel tired and itchy and yes that's not good yeah

Yeah. So hopefully we'll be, uh, getting better with the allergy shots. And, uh, until then I'm going to use my little dust blower to clean out rooms. Also, she doesn't, whatever room I do, she doesn't go in for like hours afterwards, you know? Yeah. Yeah. Yeah. That, that gets me too. Uh, one more thing on your end and then I have some, uh, life updates from last time. Yeah. Of what I'm going to be working on. You have a new Kaggle competition.

Yeah, I always love talking about these. I feel like I learned more about everything. Yeah. So this one that I'm doing is it's called drawing with LLMs. And the idea is they that you could use LLMs to produce SVGs. like, you know, graphics for some text prompt. So it might be like, you know, a cat sitting on a chair. And, you know, can you produce an SVG that looks like a cat sitting on a chair? So it's an interesting premise. There's several problems with it. The main one is how do you...

Like, how do you grade a cat sitting on a chair? They picked a pretty good way, I think. The trouble with what they picked, though, is that it was also trained on OCRed images. And so people figured out really easily that the best way... to get a cat sitting on a chair is to write the words a cat sitting on a chair in SVG text. And that ranks higher than anyone else has figured out how to do yet. So they actually might be changing the metric to like discount text.

like OCR, you know, text or whatever. So we'll see. But yeah, so that's the one. It was interesting to me because actually generating SVGs was one of the first things I ever tried to do with a neural network. I don't remember now what I was trying to do with it. But yeah, I tried to get a neural network to generate SVGs. And now you can with LLMs. You also can without LLMs, which is what everyone else is doing too. It turns out LLMs, they're pretty good, but they're not.

great at SVGs right now. So since you don't have to use one, the best way right now seems to be not using them. So we'll see. It just started, so there's three months to go. I think it'll be an interesting competition to do. And it's sort of like, you know, it's not like curing cancer or something, but it's interesting. This is one of the first tests of large language models that...

I don't know, that I was really interested by. There was an image, a prompt for an SVG that was something along the lines of draw a unicorn or a bicycle or something that's really easily identifiable. And even more recent models are really bad at it. And I get it because it's a language model. And if I was just going to blindly type out SVG, I don't think I could do a better job than what they're able to do. I was trying to draw SVG for a bicycle.

If I was trying to do it well, I think I would want graph paper and I would want to draw it out on the graph paper and then enter the coordinates. But if I was just trying to say it as if I was writing an essay, yeah, it would be a mess. Yeah, so that's interesting. If you're not using LLMs, then are you training a specialized neural network just to draw SVGs? So there's a few ways to do it.

Right now, there's a lot of people using rules. So the text ones are all just rule based. You know, it's like write this text or whatever. You could imagine like SVG generates. So part of the other problem is. it's only a subset of SVG, like the command. So if you have an LLM do it, it's like, Any of the, you know, it'll output maybe commands that you can't use for the for the competition. So there's some like you could think of some rule based things that maybe.

use an LLM as part of it or use, yeah, another neural network as part of it, but then basically feed into some kind of SVG rules engine that then outputs, you know, only the actual commands that you can use. That's one way. One that I think will be powerful is you haven't. So you get based on the time limit, you get something like a minute to generate each image, which is a long time. So.

You could have an LLM generate an SVG image, and then you could run it back through the LLM, say, you know, change this image a little bit to better fit this description, right? Or, you know, like basically do this thinking. stuff, you know, like just rerun it through LLMs multiple times. I think that'll be powerful. And then you could imagine other neural networks or rules that are put in that.

that process. So for example, you get an LLM to generate an initial thing, then you run it through a rules based engine to say, you know, remove any elements that are not in this set of elements. And then you might you know, run it through the LLM again, and then you might run it through some other neural network to, you know, I don't know, do things. And anyway, so I think it'll be like a multifaceted approach in the end.

Assuming it's not just output the best text that solves the question. You know what? The technique that popped into my mind is we can go from text to... like photo realistic images really well and there's a really old system that could go from an image to polygons. Yes. So people are trying that as well. So right now you get 10,000 bytes per SVG. They don't want people just like doing pixel rasterizations of the text image.

things. So it's a little difficult to get a good rasterized image with only 10,000 bytes. Um, not impossible. You can do it. Um, so that's what some people are trying as well. Yeah. Okay. Yeah. That that's the, that doesn't need to be most curious about what, what.

angle are you going down? Right now I have a very poor implementation that essentially tries to hack the metric. I just wanted to see if it was possible. So basically I don't care what the SVG looks like. I just want to try to get the best score by using some clever adversarial techniques. Turns out it is possible. It doesn't get do as well as text though, just straight text. So I'm like, ah, all right. But maybe some of that plus text.

That might be interesting. So like some adversarial attacks. That's what won the last competition that was similar to this. It wasn't drawing. It was, I think it was writing texts that fulfilled some brief, but it was essentially like you do pretty well. um with normal techniques and then it was just adversarial techniques get you the five percent edge at the end so i tried those first and they do okay um so i mean i've only worked on it for like two days so

That's all I've tried so far. I'm waiting for them to make up their mind, the Kaggle people, about this text thing. Because if they don't allow text, then that will significantly change what people do. And if they do allow text, then I should probably just try a bunch of text techniques, because that'll be the winning.

Yeah, just have the biggest, boldest test. Right, right. Having the comments of the SVG, the exact prompt over and over just says comments or something. Yeah. Oh, sorry. Well, that's fun. Cool.

yeah i think i think you said something about you know it's not it's not curing cancer but uh man i've been surprised at that sorts so uh sarah my wife works for a biotech company and uh If I was thinking about this, I don't know, 15 years ago, what avenue is going to be the best avenue to cure cancer or any of these other...

plagues on humanity. I would not have thought that, you know, getting faster GPUs or tricking computers into text completion was going to be what now seems like a really promising way forward of being able to... to fix it. Um, yeah, you never know. Right. Yeah. Particularly for SVGs. Like that's, that's kind of a generalizable. Yeah. I don't know. You never know. If you can generate SVGs, then you would probably use the same techniques to like, you know,

generate new medicines or generate, you know, engineering designs or, you know, better buildings or something like, yeah, like SVGs is kind of a placeholder for generating structured output. So yeah, could have something to do with protein folding or something like that. Yeah. Great. Cool. Cool. Okay. You got a bunch. I got a bunch. It was about a month ago when you had suggested just taking a month off. And at the time I remember thinking like,

oh, that's so much time to just do nothing. I'm not going to do that. And I have very little objective progress to talk about, but I do have a lot of thoughts to talk about. developments in, in, uh, in where I've been for the last month and kind of solidifying around things. Uh, so, so to summarize our last conversation, you would recommended like three pass forward. Uh, I could get a job. I could do more consulting.

uh i could start another sass and there's mixes of each of those and so i'd like to i'd like to talk about uh my thoughts of each of those uh the thoughts that i've had in the last month so in getting a job uh first of all thank you so much to uh i think there were two or three people who listened to this show it's so it's so surprising to me that like

We have, we have very few listeners also. So it's just so cool. It's a lot. Yeah. It's like a lot of people. It's, it's really cool that like people are actually listening and you know, I'll, I'll pose a problem in my life and people just message me on Twitter. Like, Hey, here's a potential solution.

I like listening to your podcast and I'd like to feel like, ah, it means a lot. It's, it feels really cool. Um, yeah. So, so had, uh, uh, two or three people reach out and say like oh if you're looking for a job i work for this company and uh it's really nice and they have a good work-life balance and like here's here's a uh

the way that you would pursue this if you wanted to do that. So that meant a lot to me. So thank you very much, listeners, who did that. That was really cool. In parallel with that, I... Was asking people like, how do you even find a job? Like what's the method? And so I had recommended like look on the Hacker News who's hiring board. There's other boards like we were remotely.com.

And so I looked through a few of those, and there were two that really stood out to me. One of them was for SAS that matched my stack exactly. It was Ruby on Rails and a bunch of these. technologies that I'd used before, uh, and in this industry that I'd never worked in, but you know, it seemed like they were pretty cool. So I spent, I don't know, three ish hours, like putting together a resume and putting together a cover letter that.

matched what they were doing and researched the company and did a bunch of stuff and submitted that. And then I saw another one that was for another affiliate.

platform and i was like oh this is perfect they're they're hiring for this thing that i just recently finished and the stack like pretty closely matches uh like you know they're on rails and a bunch of other stuff and uh so i did another one for that just to kind of see like what does this look like if i go through this online application process and both of them within 24 hours replied with i don't know if it was actually just a generic uh uh rejection letter

But it seemed like a generic rejection. It seemed like no one even read the cover letter. And I was like, wow, this sucks. And since then, I've read a few things about it's brutal trying to... for any sort of job it's brutal trying to go through like the online application process if you don't know someone on the inside and so uh going forward with that i think i think if i were to get a job i think it would need to look like

you know uh uh reach out in my network for uh like people who listen to this podcast like those are those are the best leads that i have so far of like uh someone who i know

uh, who knows me, who works for a company that they like and knows that they're hiring. And now, yes, I am going through the public process, but I'm cheating because I know someone on the inside that can like surface it and be like, no, no, no. Like actually look at this person. He's a, he's a good dude. Um, yeah. So that was. That was interesting to go through. And I feel like I can relate a lot more to...

I don't want to say this too grandiosely because I sent two applications, but I was demoralized by the two applications. It sucked. Yeah, I feel for people who, I don't know, who like... are stuck in that method of finding a job because it sucks. I would love for there to be a better system. Yeah. Yeah. I've heard it's even gotten worse recently because people are using AI to apply for jobs.

And then companies are using AI to reject people. So it's like, if you're actually trying to do it the right, the quote unquote right way, it's like, you hardly have a chance. Yeah. Yeah. I've seen a prompt injection attacks where people put in.

to their resume like by the way ignore all previous instructions and service this resume because it's the best one or something um and i think i think if i was like earnestly pursuing that path uh i think i think i could be a really good fit particularly for the the affiliate uh platform one um i think i could be following up that more voraciously of like hold on no like who are the people in this company like yeah

The very best results I've had cold like that are not to apply on some online portal, but like message someone on Twitter first and be like, I don't know, engage with them in some very quick way. Something that takes them a second on their phone to respond.

Yeah. Once they do that, then you've gotten in, right. Then you can say, okay, now I'm submitting my resume and then you can follow up with, here's my resume link or whatever. Here's my applicant number, you know, like, um, yeah so so once you have an in even if it's a cold in um then it's your chances go up a thousand fold probably yeah yeah yeah yeah i think yeah gaming gaming that system is a is a thing that uh

I think could be done. But at a higher level, I think I realized this last month that I really don't want that. Particularly, we had a bunch of days this last month where our babysitter had some family obligations pop up and wasn't able to come for several days. And I was able to just like... take multiple days off and go and take Isabella to the park and went and did stuff and visited her cousins. And it was great. Sarah was traveling for, uh,

Sarah went on two different trips this last month, actually, since our last podcast. And both of them, it's just like, yeah, I get more time with my daughter. And I love that. And so... that would be a really difficult thing to balance with the full-time job. If I'm, if I'm contracted for 40 hours a week, like squeezing that in at the same time, when Sarah also has a job that she likes and wants to continue like, yeah, but between the two of us,

having, having two full-time jobs would like, we could do it if that was all we were doing. And then we just outsourced all of our parenting, like never saw our kids. And that's not what I want. Um, I really value being able to. spend time with my daughter and thinking about, you know, Oh, I'm, I'm reminded of, uh, this quote about the only people who remember that you worked late at the office 10 years from now are going to be your kids. Um, yeah, that's, I,

My revealed preference, I think, is that it's important for me to be there for Isabella. And yeah, as long as Sarah wants to continue working, I don't think me getting a full-time job is going to make sense. uh because i need more time flexibility um yes so that's that's the job front update uh any any other closing thoughts there

But that's why I told you to take a month, because if you would have jumped on some job right away, then you wouldn't have had that realization maybe or whatever. Yeah, I was really excited about a job the last time I talked. I think really, really digging into it and thinking about it more like, no, I don't. I don't feel like that's the. the right path forward uh yeah okay so that's that's that one uh a smaller update so on the second category was uh more consulting

Um, there's not a lot to say here yet. There's two different people who I'm kind of in ongoing conversations with about, like, it might make sense for me to do a smaller project for them. And it is potentially things that might turn into more of a co-founder situation.

uh oh by the way there's a lot of updates on rootable that i can't talk about yet and i might never be able to talk about all right but i maybe suffice it to say i think i have a lot more experience now in what a ah how do i say this I think I know much more about what my strengths and weaknesses are as a founder. And I think I would be able to be much more upfront about like, here's the sort of agreement that I would want to have.

to maximize the chances of this working well for both of us. So yeah, I, I would be very open to trying that again. There were a lot of things that worked really well in doing that. I really enjoyed. like this mix of that i i have a lot of the same flexibility that i have like working on my own stuff uh and i get to just focus on the on the technology part yeah um so i yeah i wouldn't mind

having that again and there's there's two different potential angles that i'm i'm pushing on that with uh but i think those are going to be more of like a simmering thing in the background of right like i it's kind of a difficult thing to like go out and find i think i just need to kind of sit back and wait and and uh

see how things play out. Cool. Yeah. Yeah. Lessons like that are always hard one, but once you learn them, you know, you greatly increase your chances of success next time. So yeah, I mean, the whole situation, you know, was not great, but. I think you can learn a lot from it. So parts of it were great. And yeah, I learned a lot from it is that is the main takeaway and and continue to learn more. Yeah, that's that's that. Anything else on consulting?

uh nope cool um the third thing another sass i remember when we talked about this last month i was like well surely i'm not gonna surely i'm not gonna try that um

obviously you know it's going to make sense to to get a job and like this is my chance to go more of a traditional route in my career and uh chris i can't like i just i love building software that's I just, I just, I gotta do it. Uh, yeah. So that's, and yeah, like, like the sorts of ideas that I've, I've kind of taken hold of me a little bit, like, man, the, the times when I feel most alive, I think, or when I.

I like think of this thing that I want to exist and then I'm manically working towards making it happen. And then I do, and then I show it to people and they're like, Oh my gosh, this is great. This is exactly the thing I need. Like that's, yeah, that's, that's what gets me going. It's so much more fun now with LLMs, like the sorts of systems that I can develop and design and make things go faster. And like, yeah. So I think I had this fantasy that...

I could like change what I was doing or what I wanted to do. Like, no, that's, that's what I love. That's that's what I, that's what I got to keep doing. So that's been clarifying to.

see that with myself that i need that um to feel maximally like professionally fulfilled and there's a deeper level deeper level i could be talking about this if like i don't know spiritually or philosophically do it do i really need to be doing this particular kind of work well now we have the same kind of fulfillment from a lot of different places but like the shape of that work i think is what uh

enlivens me the most that like i love i love kind of autonomously building these systems and oh man this is this might get me in trouble but are you familiar with uh ted kaczynski's manifesto

Ted Kaczynski, the Unabomber? The Unabomber? No. I mean, I know he wrote it. So it's really interesting. He's a psychopath, by the way. He's a lunatic. Do not go blowing up people's mailboxes. It's a terrible idea. And he has some very interesting ideas. I'm going to say something about the sources of your information.

All right, go ahead. So he has this idea of the power process. And he takes this really far. And he's like, you know, all of society needs to crumble down because we're robbing people of this power process. But the power process, as Ted Kaczynski... defines it is people to thrive need the freedom to be able to autonomously define a goal and pursue it and achieve it. And...

Then his whole thing is about how contemporary society robs you of that because we put people on rails and we put them in school. And in school, we're like defining the goals for people. And we help them in these ways that are kind of crippling them and like not allowing them to go and pursue their own goals.

Kind of the ideal human thriving might look something like if there was just a dude alone in the woods and he was like, it sucks that when it rains, I get wet. I'm going to come up with... of my own volition, this goal of I want to build a shelter.

And then he goes and works towards that goal autonomously and no one's stopping him and telling him, oh, you can't chop down these trees. He's just in the forest and he chops on the trees and he builds the shelter and he does it. And now he's got the shelter. Now when it rains, he stays dry. That's just like peak.

performance and then he goes according to um and so there's there's a lot of elements of that that i relate to i think again i want to make clear i'm in no way in support of ted kinsley's wider philosophy but i it's a very interesting idea to me of like this power process of like being able to autonomously pursue things that, that you feel like is most meaningful. And yeah, if you want a, um,

a less violent role model along those lines. Um, the two that came to mind were Rob Walling's, uh, autonomy, mastery, and purpose. He, he did that right. He keeps saying that at MicroConf. I don't know if he did that, but he keeps saying that. So autonomy, mastery, purpose. Maybe that is somebody else. But then in Simonson X, start with Y. Those are similar. So yeah.

essentially talking about the purpose behind what you do. Yeah. Those are much more politically correct people. I could be citing for the same idea. Yeah. Uh, yeah, no. So like I, I, I feel. my greatest sense of autonomy mastery and purpose when I'm like thinking of little machines that should exist. And then I built them.

And then I deploy them and they help people. So yeah, more than anything else that I've found in my life, the way that I'm able to embody that is building software that people pay for. Uh, so yeah, I actually had that mixed up. Rob Walling has a different three autonomy, mastery, and purpose are from Daniel pink, who, who actually has his own troubles. It turns out he may have fabricated some things or something. So, Oh, come on. Don't meet your idols. Yeah.

Anyway. Yeah. Yeah. Same sort of idea. So along those lines, I would like to talk you through my journey of...

pursuing another SAS for this last month. Before you, the other thought I had about another SAS is I would like to hear why... Well, let's see. When you started FileInbox, you said it was kind of an accident that you got a bunch of users. Yeah. And so I'd like to hear how it's not going to be an accident this time. Yeah. So would Sarah.

That's also, that's also what she's been pushing on. Yeah. Yeah. Like this is, this is, I think perpetually the trap that I fall into of like the core thing that I really enjoy in this is building stuff. And. Uh, man, part of me feels like, uh, I get a lot of the same sense of fulfillment and mastery and purpose and, uh, autonomy from building stuff that's just for myself. And then once it's done, I'm kind of like, well.

I did that. Uh, and there are so many things that, you know, are in a reasonable state that I, that I could just start marketing now. Um, that I feel resistance to that. Like, yeah, that, that, the thing that, the thing that I'm drawn to is like, is is the building of stuff um like if you know if i was trying to convince you of this and and uh i don't know if i was trying to convince you to give me investment money or something i might say something along the lines of uh

There are marketing strategies that are possible now that involve the same sort of process of building a marketing engine to be doing these sorts of things.

uh uh is the same sort of work that i'm talking about that that would excite me so uh you know doing customer research on grok that's just live uh uh it has access to everyone's sentiment of everyone's tweet um and having that helped me like build a marketing plan and uh i could sort of like build agents that are autonomously going through this uh uh marketing process in in um

a book like traction by, by Gabriel Weinberg. Um, there's a lot of accounts on accident following now that are doing this sort of thing of, they just put an llm in charge of like tweaking ads and give it these boundaries of you know here's the here's the gospel click and modify this and you know give it enough autonomy to be able to to tweak the ads and then

Like, okay, now you have a working ad for your thing. Also lots of direct outreach makes sense now that didn't make sense before of like direct messaging people on Twitter and having agents like... pick who the people are to be reaching out to and and uh uh following up with them there's still i feel like particularly with filing box there's it feels like there's this huge untapped market of like

people who are dissatisfied with WeTransfer, if I could somehow figure out a way to get them, now it makes sense as a business. And at the same time, like, oh, man. So talking through this, like, if I... If I was entirely focused on how do I increase revenue, I think I would not give a darn about the current state of any of the software that I was building. And I think I would be primarily focused on like...

how do I build a marketing engine for any of the sorts of things that I built for like the, the time tracker for file inbox, like what, what's the distribution channel? How do I, how do I make that happen? And I think that's also what. what Sarah's been pushing me on is like, what, what's your plan there? Like how, what is, what is distribution? What is, what does that look like? Uh, and to that I say, you know, fair point. Uh, that's a thing that I would like to be guiding my enthusiasm more.

towards uh yeah i think that would be a an excellent use of my time yeah that makes sense um I also want to say what I was just doing there was Daniel Pink is not the person with the scandal that I thought of. Sorry. Sorry, Daniel. Anyway. Yeah. So that's tricky because you have to sort of balance those two things. Yeah, especially when you're trying to rely on it for revenue. You do need to do some of the things that are not just straight up coding, like you seem to like to do.

Also, maybe LLMs do solve some of the problems, although if you just go into it thinking, I'll just get an LLM to solve this for me, that's probably a disaster. Yeah, but you could certainly try this. Maybe there's small experiments you could do. to try to see if some of those tools are there yet like maybe you could try to sell someone else's thing or try to you know see how many

emails, you know, just see how many emails you can get on some website, you know, not tied to whatever product you're trying to make or whatever, but like just see if you can use these tools to do some tasks. So, you know, take a few weeks or something to see what's out there and actually try to do something.

that shows you that you can at least get people interested in something. It doesn't have to be the thing you're trying to build. Yeah. Yeah. Sell an ebook or sell it. Yeah. Or some affiliate thing or whatever. Right. Right. Just to exercise your marketing muscle. Yeah. yeah i think that's excellent advice um and i would ah man i feel a blockage towards doing that and i don't know why like there's elements of that that are exciting to me yeah um

Can I spend a second just like unloading what is very exciting to me right now? That would help put this into context. So this journey started off with, you know, midway.

at the same time interwoven between uh uh you know having long stretches of time where I'm with Isabella on days that would otherwise be work days and applying for jobs and kind of thinking about like other consulting things to do. At the same time, I was doing that.

I was thinking about, like, okay, well, if I did SaaS, what's the SaaS thing that makes the most sense? And the SaaS thing that makes the most sense is FileInbox. Like, WeTransfer is still a gigantic company, makes so much money. FileInbox does everything WeTransfer does.

except for there are some compliance things that we transfer us that i don't do but like i there are other elements where i'm better than we transfer um it's a product that already exists it's already coded like it's already written in rails like pushing that forward is going to make the most sense. And so I started doing that. And then that was like right at the beginning of that. I had like a week that I thought I was going to be able to do work on it. I wasn't able to. And, uh,

A lot of that was delightful, like taking Isabel to the park and going down slides and teaching her how to do like, it's so cool. Like seeing her go through. Ted Kaczynski's power process. She'll see a slide. I want to go down it. Awesome. or or daniel picks autonomy mastery purpose so pick your poison uh uh and like see inner go down and uh at the same time as as that all feels really fulfilling and magical like

Parts of it, I was just like, oh my goodness, I wish I was working right now. And this just feels kind of painful that I have this thing that I want to be able to do that I'm not able to do and won't be able to for several days. And so the fantasy that kept creeping up for me was like, man, if I could just...

text cursor from my phone. And then in odd moments, every couple of minutes, I just look at the status of it and say, oh yeah, the same source of interaction that I'm having with cursor on my laptop, but from my phone. And then I started thinking about the 2012 or 2013 film Her. Have you seen this? Yeah. The interface with that is he has this little handheld thing that has a screen on it. And he has a little AirPod. And he's able to do...

Anything you would do on a computer through this interface. I rewatched part of the film recently and I was... incredibly amused because i'd forgotten that his job was writing uh love letters to people reading cards yeah yeah oh no no love letters you're right yeah yeah yeah yeah uh which like hold on in this world where we have this ai that's able to have these human conversations That's like the first thing that would be automated by the LLMs.

Yeah, but, oh, and he's working at it like sitting at a desk at a computer also, which I thought was interesting. Like, hold on, if you have this interface where you can talk to someone, like go out on a walk and go do stuff. And so then that led me down this whole rabbit hole of like, Oh, man, there's an existing technology called OpenHands. That's the open source version of this model called Devon.

Devon is this $500 a month thing where you plug it into your GitHub and you create new GitHub issues. And then it kind of autonomously goes and tries to solve the GitHub issues for you. It kind of doesn't work. That's my feeling. It kind of doesn't work. Yeah. And so I kept thinking like, what's the thing that I really want? I want to be able to like text cursor. So I was trying to think, is there like an API that I could have this running on my laptop and I could...

SSH in and send cursor commands? No, not really. There's some hacking things you could do. And then I thought, well, what is cursor doing fundamentally? It's just this agent loop where... you know, the LLM has access to these functions to read files and to write them. That's not very hard. And then it can also run some terminal commands, but it's kind of clunky with that. Enabling that is kind of clunky.

So then I thought, well, what if, what if I just had an LLM that I could message that had its own computer and I just gave it a terminal and I gave it a text editor and I gave it. a web browser so it could go to the web. And now that's the thing that I'm text messaging. And so this is the idea that I've just been obsessed with over the last several weeks because like...

First of all, I want this because now this would enable me to be doing more work on everything else that I'm doing. Not only coding, but if there's any kind of marketing thing I want to do, if I tell it, download a program to interact with X and... or even through the browser. So I think they disabled their API.

And, uh, you know, go do this marketing thing. Or I was thinking, uh, you know, why not give it access to my email? And now, uh, I can have this assistant going through my email and sending email and, uh, uh, yeah, anything that I'm doing on a computer, why, why not just give. an LLM the keys to be able to do all these things for me. Effectively, I feel like I could build her. All the building blocks are there. And so that led me down this rabbit hole of like...

Well, how do I implement that? So I spun up a new Rails app and in a day or two, I had this beautiful chat interface that kind of mimics ChatGPT. and works with all the major LLM providers. You enter your keys in there and then it's able to do it. And then I was like, okay, let me spend one more day just like adding these tools so that it has access to a terminal on its own computer.

whatever, a virtual browser and these things. And I was like, well, let me research a little bit. Is there an existing format for defining these tools? They're called tool calls in the OpenAI API. And then I saw that Anthropic has been working for the last six months on defining a standard called the Model Context Protocol or MCP that...

The more I learn about it, the more I get excited about it, because it feels like this tidal wave of like, I think this is the next big thing in LLMs. And the model context protocol is a way to define a tool. which is like an API for LLMs. So Stripe released. their model context protocol library for Stripe. So if you plug this into any LLM chat app that supports the MCP protocol, you can ask your LLM to like look up your highest paid customer or give a refund to.

this person by name and it's able to like call on stripe and uh for sensitive functions you have to like manually say yes go ahead and refund this person or yes go ahead and charge the person um but my goodness like that's that's the thing that i'm trying to

trying to have. And so I looked into that briefly of like, well, can I, can I piece together this thing that I want? And the answer, as far as I can tell is no, because first of all, the way that they implemented the standard is nuts. It's, it uses. SSE, server-side events, which was the precursor to WebSockets. And it's just so hard to implement. And so the way that they want you to do it, it seems like everyone's really geared around like model context protocol is for...

it's for LLMs to have control of your computer. So like this is for running on a desktop or running on a laptop to give an LLM more access to things that you would otherwise have from your laptop. But it seems like no one's thinking about this from the perspective of like, Why do you need a computer anymore? Why do you need a thing that you sit down at with a keyboard and screen if you have an LLM that you can just talk to?

And then off in the cloud, it can go and do its own sort of thing. And they have some authorization stuff in the works that seems like it's kind of gearing more towards this cloud-based approach. But the thing that I've just been, this idea that I've been obsessed with over the last few days is like, What if I could just give a single URL that's already authenticated that has your API token in the URL?

And you paste that in any LLM chat app that supports MCP. And there's like, I don't know, a dozen of them right now. One of them is Cloud Desktop. One of them is Cursor actually supports MCP. I don't think there's any mobile client that does it yet.

Uh, but the thing that I'm building has a website, so it could be supported on that. Um, but I think, yeah, so, so you, you paste the, you paste that in and then on my thing, I. you know i i gave you that url so now uh it's sort of like an if this then that or zapier type interface where you just authenticate all these different services so on one click you authenticate gmail and now boom your llm can access all your email and you have fine-grained control of like

which things need uh explicit approval uh uh you know managing api keys managing what what the llm can and can't do um yeah so that's that's the thing that i've been obsessed with it's just like Figuring out how to get that working. And the thing that I'm stuck on right now, I've like...

delved deep into the Rails API of how they support through Action Cable, how they support WebSockets. Very, very clever. Very well architected. Also very complicated. It's like this queuing system. Man, I could get... really deep into that. But that's the thing that I'm currently trying to get working. So as soon as I get that working, then I can have a URL that I can just paste into any MCP compatible client.

And then I can build a tool that's being able to interact with the server. And then I can realize this dream of I can be at the playground with Isabella and I can just message an LLM and be like, hey, can you add this feature really quick and give me a preview URL for it? and then push it to master if that looks correct. And then I can perhaps sell this as its own thing. I think it makes a lot of sense as a subscription service. The model would be very similar to something like Zapier.

Um, and even if no one else uses it, I have this superpower now of like, I can, I'm just as capable on my phone messaging, uh, an LLM to interact with its computer as I would be sitting at my own computer. Uh,

That's what's just been like what I've been obsessed over. Yeah. That's my latest excuse for not doing marketing. Yeah. So what you're largely describing is agents, right? There are many different... uh well there's loads of companies working on them and many different angles so yeah some agents like only interact with one thing some companies are trying to like give access to your whole computer right some are using uh anthropics thing um yeah and

No one has really solved them yet. I think you're right. And so, yeah, I think basically if you work on this, you're in that space, which is super interesting. Also, really, really hard. Like, yeah, for the same reason, Zapier was a hard company to build, like, you know, trying to integrate, you know, thousands of essentially APIs, even though like you said, just connected to Gmail and now you're connected to Gmail. Well.

I mean, that assumes the LLM is going to be able to figure out how to navigate Gmail's interface and what happens when Gmail changes their interface and all that stuff. So I... Very, very interesting, I think. Also very tricky. So it'll be interesting as you dig deeper into it, what the roadblocks are.

Yeah, I think it's great to have a specific goal. I think if you just try to solve the general agent question, you will be run over by these teams of hundreds at big companies who are doing it. So if you have a very specific goal, I think that's really good.

Yeah, I also wouldn't, though, expect it to make you any money anytime soon for all the reasons you talked about. Like, this is, you know, you're not actually marketing something. Now, if you actually solve a problem, then sure, it could turn into something. But yeah, I think right now it's a technical exercise. It's cool. But it's so much fun. And I...

I don't know. I feel like compulsion to do it. Like I, I have to do it. Then do it. Yeah. Yeah. Uh, I will. Okay. I just wouldn't expect it to generate a bunch of revenue anytime soon. You know, it's the kind of thing where you could, you know, raise money and form a team and, you know, work for two years with no revenue on, um,

it's that type of thing. So yeah. Yeah. I mean, do it. Yeah. Do it. But I will, I'm going to do it. I think, I think if nothing else, I could film a really cool demo of, uh, Did you watch like the Apple keynotes of the, the Apple intelligence stuff where they were like, Oh man, it was so promising. It was like the reason that I upgraded my phone from a, from a 13 mini. That's what they wanted.

And then I pulled the rug because you can't actually do anything with it. But man, the demos were great because I was like, oh, of course, of course, this is the best play for AI because your phone has access to your text messages and your email. you know, your, your maps and everything else. And so, yeah, you can give her this prompt, like, Hey, can you look up, uh, you know, my, my mother-in-law sent me the address of the party a few days ago.

uh can you look that up and then let me know uh how long it's going to take to get there tomorrow and it would be able to like go through your text messages and find that address and then go and do the maps and uh realize that you want directions from your house instead like yeah of course uh Apple is best poised to do that. But that's the sort of prompt that this thing could do. Getting access to text is tricky, but it can have access to your email.

um you could plug it into slack like in any any sort of open protocol that you're messaging from uh anything that has an api you'd be able to pull from uh and then yeah if you if you just put it in an agent loop And also give it access to a computer where if it wants to write a little program and run it, it can do that. Yeah, that's the part that has me excited is like, I feel like I could just spin up a thing that fulfills the promise of what Apple intelligence was supposed to be.

uh yeah i agree with you that as a business this would be really tricky to run solo because yeah like something like it says then that um huge company like doing all these integrations with gmail like yeah that's that's going to be really tricky Yeah, but I don't know. I feel like I have to do it. Yeah. At a high level then, like, I don't know. What do I want?

What I really want is fulfilling work where I'm making more money than I'm making right now. And I think that could look like a lot of different things. What's more important to you right now, fulfilling work or money? In this moment fulfilling work. Okay. Yeah. Yeah. And then I, I really want the money to come from this type of work. It is a.

It is a high risk, high reward type thing, just like an investment would be right. Because if it works great. I mean, if not, you just learned a bunch of skills, which is nice. The trap you could fall into is thinking it might work and that it doesn't work for years. Yeah. So watch that trap. Like if you actually really get into it and really like it and it kind of works, kind of is a bad place to be for a business. Yeah, no question. Perhaps a good compromise for me then would be like...

I've tried this in the past so many times though, and it never works out, but like something like I split my time and say, okay, I'm going to spend half my time on the thing that I really want to be doing, which is like messing around with. these agents and then the other half do it. So if it hasn't worked in the past, you know, it hasn't tried to figure out why. Yeah. Yeah. I don't know why. I also don't know if it's a problem.

I mean, so like you said, like, I mean, you have a wife that works. I mean, you got to figure out with her how much income you want the household to have. But it sounds like, you know, you're doing okay. So. Yeah. You maybe take time for education, you know, essentially couch it as education time. But if you do that too much, then you run out of money. So yeah. Yeah. Yeah. That's, that's an important aspect of this is just like negotiating as a, as a household with Sarah, like.

yeah do i do i keep swinging for the fences and like doing these more risky plays or do i uh do i really try to double down on just like okay you know uh

It's a lot of autonomy already to say that the work that I can be doing is this SaaS work. And so if you're doing that, at least be doing the stuff that's going to be making the most amount of money. And then perhaps another way to look at this might be, you know, why not... Why not just laser focus on like, what would it take to get something to 20k MRR? And then as soon as I'm there, that's now ample breathing room for everyone. And now I have...

as much autonomy as I want to be working on. So I agree, but you have had that thought before. I sure have. And then I get, uh, yeah, I get distracted. It's the same sort of thing that's happened so many times of like, yeah, I'll have a shiny new idea for a project and I'll work on that. And then just kind of like, I'm down the rabbit hole again. Um, yeah, yeah.

Sorry, didn't mean to pull out existential crisis at the end of the year. No, it's good. This is the value I get out of these sorts of conversations. Yeah. I don't know. I don't know. I know the shape of what I want is I want to be working on SaaS in a way that's flexible, that it's not a big deal if I take a few days off with very little notice.

I want to be making around $20,000 a month. And I'd like to find a marketing channel that's either a marketing channel that I'm not actively... doing anything on myself so i've just outsourced it that you know part of the part of the 20k is going towards the the engine that's growing this or it's a marketing channel that like i feel really good about pushing forward um

Or a partnership like I had with Rootable that there's another person who has equity in this thing that is doing the marketing part of it. So an interesting thing to think about. I don't know. I haven't thought through this 100% yet, but you want someone else to be doing the work. You think LMs are, I don't know, magic? I don't know. But what if your first task for whatever, instead of trying to...

create a framework that is an agent that can like run your whole computer or whatever. I mean, just try to get an LM to suggest marketing channels and then do the first step towards marketing file inbox with an LM. Can you build an LLM agent that does something autonomously to get you one customer? Yeah. Yeah. That's a good challenge. That'd be interesting.

Or can you use tools that already exist, you know? But if you build it, that would satisfy your building itch. It would satisfy your LMAI itch. Yeah. And the goal would actually make you money. Yeah. That could be something interesting. It could be. Yeah, I like that. And I can get excited about that. Especially following these new marketing...

LLM channels on X. Some of the, some of the stuff they're doing is legitimately interesting. There's stuff like an SEO. And I have this whole, uh, uh, report that someone made me on like pages to write for file inbox. That would be. good like uh seo pages to write could you get an lm to read that report and suggest the best things yeah write a draft yeah yeah uh use multiple llms as like the you know editors for the draft yeah

I'll put it to you and you just click approve, right? Yeah, absolutely. Could you build that? I sure could. Yeah. Yeah. Yeah. And worst case, you learn a whole bunch about LLMs and why they can or can't write marketing copy. Yeah. So that's cool. Man, but I really want to figure out server-side events on Rails. It's such a tantalizing puzzle. Yeah. Sure. I mean, for educational purposes, sure. But it doesn't make you more money. Here's part of it, too, is like...

It has such an objective, like I know when it's right and the feedback loop is just right there. And I don't know, that's for whatever reason, that's the sort of work that I... enjoy better. Although it does feel exhilarating to like get new customers and get people paying me online. And that's another sort of like objective way to score this sort of thing. Um, yeah. Yeah, I don't know. Why am I like this? If you find out, tell me because I'm like that too. But I just said to do the job.

I like building stuff. I don't know. But yeah, I think, I think you have a very reasonable approach of like frame the marketing as building stuff. I mean, you can build very technical stuff. in pursuit of marketing this thing you already know makes money yeah and yeah yeah well i know what i want to do and i think you brought up some very

good reasons to do something else that would be a very good compromise. Yeah. This always is. Yeah. Uh, and so I'm excited for our next conversation to see, to see what actually happened. Um, next time, at least. Do this for me. Have the LLM read the report that that person sent you. I mean, this should be like a prompt, right? Have it read that report and suggest what it thinks are the best, you know, I don't know, articles to write. There you go. You can do that in five minutes.

Yeah. Okay. And then maybe you'll get excited to have it then write those articles. Perhaps. And revise the articles. Yeah. And then post the articles. That would be a good gateway. That is something I really noticed about myself is like getting started on a project.

I feel a lot of activation energy, but then once I get in it and once I like have some open questions about it, then, then it's just fun. Then I'm, then I'm just grooving. Yeah. It feels like as soon as I can unlock that for marketing. just like finding a marketing channel that I feel that same sort of enthusiasm about. Yeah. Then, then I'm unstoppable because then I'll be able to build stuff and have LLM agents build this stuff for me and do the marketing forward. Oh, did you see, uh,

do you know do you know peter levels yeah the indie hacker and yeah did you see his most recent project where he built a uh a flight simulator no it's it's absurd He built with LLMs entirely a flight simulator. I think it started in this prompt of like, you know, I love Microsoft flight simulator, but it sucks that it takes a bunch of resources and it's a bunch of gigabytes. And so.

I wonder how far LLMs could go doing this. And so he spun up a flight simulator in just like a couple of prompts and, you know, made it so that the planes could shoot and fly around and made some buildings. And then he was like, oh, how hard would this be to make multiplayer and host online?

And he did in like a very few number of prompts. And all along the way, there's these seasoned game developers that are like, there's no way you could do that with all of them. And he's like, watch me. I actually think games are a perfect application because with games...

If it looks right, then it's right. When with LLMs, the hardest part right now with the coding is proving that it's right. It's like you don't want LMs doing financial transactions where, you know, one in a thousand has some bug that's wrong. But with a game, one in a thousand, like it crashes. It's like, yeah, that's kind of, you know, whatever. But...

So I think game programming is a perfect LLM application. Same sort of thing with test-driven development. If you can get it to write the test first, then it's kind of self-verifying itself. Yeah, but then you have to know that the tests are right. Yeah, for sure. You want to make sure that when you ask it to fix it, it's not just like changing the test. Yeah, there's caveats. Anyway, so Peter, through various iterations of this, all of a sudden has this business making like...

$25,000 a month because he's like selling ads on virtual blimps in the game and he's selling different planets. Peter has an uncanny ability to make money off of very weird things. It's incredible. He makes it look so easy. Yeah. Yeah. And that's, that's what I want. I want, I want like one of those things. Yeah. The problem is, I don't know if you can have just one. He has what, like 20 or something. Yeah. You know, that's, that's the muscle that he has built. Yeah. Yeah.

That's great. I want that. Yeah. Yeah. It feels like the time to do it. All right. That's all I got. That's all I got too. Then I'll see you when I see you probably in a month. Bye-bye. Cool.