Pragmatic AI with Matt Stauffer

The Pragmatic AI Podcast is sponsored by Tighten. T-I-G-H-T-E-N. We will take your AI ideas, prototypes or even vibe-coded apps, and we'll take them to production. Scalable and secure. Check us out at tighten.com.

In this episode, Matt Stauffer talks with software engineer and self-described "Renegade Historian" Dave Stanton, who has built an entire AI-driven workflow that lets him program — and write a 400-page history book — almost entirely by voice from his phone.

They get into the nuts and bolts: an always-on Mac Mini running stacked Claude Code sessions in tmux, SSH access from anywhere, voice-to-text tools, and a self-designed "PhD" where Sonnet plays research assistant and Opus plays a hypercritical dissertation chair, with every critique tracked as a GitHub issue. Dave explains why the book itself was never really the goal.

The conversation widens into the bigger questions — what gets lost when AI handles the rote work, whether to retain human connection on purpose, and Dave's honest 80/20 split between optimism and unease about where all of this is heading.

-----

Editing and transcription sponsored by Tighten.

Big thanks to the companies that support the show:

Mailtrap

Creators and Guests

Host

Matt Stauffer

CEO of Tighten, where we write Laravel and more w/some of the best devs alive. "Worst twerker ever, best Dad ever" –My daughter

Guest

Dave Stanton

Renegade Historian

What is Pragmatic AI with Matt Stauffer?

Pragmatic AI is a podcast about practical AI for business leaders, developers, and thinkers—the stuff that actually works, not the hate or the hype. We cover using AI tools day-to-day across the spectrum of human experience.

The Pragmatic AI Podcast is sponsored by Tighten.

T-I-G-H-T-E-N.

We will take your AI ideas, prototypes, or even vibe-coded apps,

and we'll take them to production.

Scalable and secure. Check us out at tighten.com.

Hey, and welcome back to Pragmatic AI, where we talk about using AI in the real world.

What works, how to use it well, and when it causes more harm than good.

Practical tools and real trade-offs for builders and business leaders.

My guest today is my friend.

We were just talking beforehand, my friend of quite a few years.

I haven't even seen him in seven years.

And so that's how long I've known this guy, Dave Stanton.

And the I I freaking love this.

I asked the guests to tell me how to introduce them.

And it's always I'm a software engineer at blah blah blah.

Dave's title here is Renegade Historian.

Dave, would you please say hi to the people and tell us what does that mean?

Give us some context.

Sure.

Hey chat.

Uh glad to be here.

Glad to talk to you again, Matt.

Been a long time.

Um so my day job is I work on this travel middleware for American Express travel for a
partner.

Um highly regulated industry, so it's a little hard to do just wild AI experimentation on
the work laptop.

So I'm also a recent empty nester.

The girls are gone, uh off at college.

Thank you.

And because of that, it's like, well, I got some free mental cycles and I start to like
fiddle with some other sorts of projects.

And like many of your guests guests have talked about in different episodes that that um
November 2025 inflection point, Opus four six, you you could just start to do stuff.

Like um you didn't have to look at the code quite as much.

So I was tinkering on my own, building things.

I started getting into this heavy voice to text workflow.

I do almost all my programming from talking to my phone.

Which you probably might want to like go down some of them rabbit holes.

And I actually I I built I'd been programming with my voice for a few months using the
Blink for iOS app, which is fabulous.

And Super Whisper, lots of bouncing back and forth.

I saw this joke on and we'll get to Renegade Historian, but there's a little bit of an
origin story here.

So the reason I

I saw this joke.

Maybe it was like I don't remember exactly remember, but you probably saw it too, where it
was like, all the new Mac keyboard that is just one, two, three.

Right?

Of like, that's all you need now because we're just vibe coding everything.

I saw that on a Saturday morning as a joke.

I threw into an image gen in like a uh open AI Dolly image gen thing, and I'm like, let me
build a phone case that just has three buttons that would power Claude Code.

And then I was like, well, why don't I just do this as software?

And by four hours later, using Blink and other things that I had set up, I had built the
iOS app to do what I wanted with the soft keys.

And within a week, I had it deployed into the app store.

99% of it written with my voice.

I still had to set up some certificates and stuff on my computer.

And so I was building apps, trying stuff, just constraint breeds creativity.

And I hit a point where I was the getting a little burnt out on just building things for
the sake of building things.

I've always loved history, history minor, the last couple decades, my main if I'm going
jogging or working in the yard, I'm usu usually listening to a history book or a history

podcast.

I had the somewhat programmer, like, well, what would I do if I give it up all the way and
be a goat farmer?

Sort of conversation you have in your head occasionally.

And I was like, well, what'd I do right now?

My girls are in college, I've got a lot of empty brain capacity right now.

I'll go get another PhD.

Another.

Yeah.

As one does.

as one does, I enjoy rigor, I enjoy like frameworks a lot.

And I was like, but that's silly.

I don't obviously I'm not gonna do it and pay for it.

I was like, well, what would I do?

I would want to go write a history book.

What would I write it about?

I would want to mash up because I was listening to a history book and this idea, this
little had this little um nugget of an idea, this curiosity, and I was like, Well, I would

want to write a history book.

That was a mashup of usability research, things that I learned from my uh human factors
PhD, things that I've learned as like a software engineer, and mash it up as this like

thermodynamic explanation for movements of people.

And uh and I was like, well, most historians go through and they work on a PhD and they go
through all the coursework and they do all this stuff to publish the book.

I'm not trying to make any money off of it.

I don't really need another PhD.

Um, it was just more of a framework to give myself for the intellectual rigor to just not
slop out a bunch of words for the sake of of slopping out words.

And so that's how I started of, well, I've got these four or five books that were the
intellectual foundation for my view.

So some things that some readers might be familiar with or viewers might be familiar with.

Things like Claude Shannon's information theory was very important to me during my
doctoral research.

and then you've got the diffusion of innovation, Everett Rogers.

You've definitely seen that S curve, innovators, early adopters, right?

So those two were really foundational.

And then there's a couple of these history uh perspectives that I found very interesting
when I came across them.

Jared Diamond's

He's more of uh geographic determinism.

His famous book is the guns, germs, and steel.

I think I got those three in the right order.

And clo it's c it's close enough.

The idea is that like where the people were seated on the map and what resources were
there predicted the outcomes.

Got it.

Okay.

Yeah, that makes sense.

I'm not a determinism guy.

I'm a probability guy.

There's so much randomness.

There's all this sort of stuff.

So, but I like the idea of I've never enjoyed the time place thing of history.

Like this person did this thing.

Then these people went to that place.

Um, and so I had a couple of these ideas and I started to see some parallels, and we don't
have to talk and like go super deep into the

The historical content of the book is that's not really what this podcast is about.

If somebody else cares, just go to my website and you can read it.

It's all open access.

Um I from the beginning would be almost like flip what would academia do on its head.

Um I've already gone through all the seminars.

I know what the structure works like.

I know the sorts of classes you would take to learn epistemologies and research methods
and all this sort of stuff.

And I just applied that.

To conversations with Claude and just imagined, well, here's my setup for what I would
want a history book of probability with geography for history.

What would be my first semester?

Oh, well, you should read some stuff about the Silk Road and you should read some stuff
about this.

And then I would read the books as I was walking.

I would have conversations with Sonnet via Claude iOS.

I would have some back and forths.

And then

when something came to my mind that like synthesized an idea from the book that took took
like an idea that was in the book and it landed it inside of the framework, like a like an

outline I would have for my book.

Then I take that little nugget of a conversation and I go into the iOS app that I built to
make this faster and I file it off as a note card.

Which is what which what you you do as like a doctoral student as well is

You get a piece of you get a journal article.

You're like, all right, I've got this rough outline for what my dissertation is eventually
going to be in three years.

It's gonna be a couple hundred pages, it's gonna have these sections.

I need to talk about these theories.

And each class you take is supposed to fill in a section of your dissertation.

And so that's just what I did of like, well, let me do a thing that's gonna fill in these
chunks and I put them all down as note cards, and then uh I haven't.

I kept uh the Opus models, I kept that to do to be my faux uh dissertation reviewer.

And so my prompt

familiar, the sonnet models are the faster, cheaper, dumber, and the opus are the the
slower, smart.

Yeah, and which was better for the flow, 'cause if I'm talking with you and I want you to
be faster so that I can have the back and forth and the I

versus a oh now we're gonna wait for five minutes, yeah.

And the goal was I didn't want the models to overfit or like give me the answer.

I wanted to say, this is what I am hearing, this is what I am seeing, this is a connection
I think exists.

Fact check me.

Help me go dig up some research.

And so that's kind of what I did.

And so from the the long-winded way of circling back to the the rogue historian as a
little bit of a of a uh

inflammatory, mildly inflammatory title is I'm just taking how a historian would work,
somebody going through a doctoral program and be like, well, if I'm not having the

gatekeeping of three plus years of coursework and the working in a lab and doing all of
this stuff and having to publish in pretty expensive journals to to feed the tenure beast,

how would I approach the synthesis and the publishing of knowledge that is to the quality
that I

No a dissertation should meet.

Yeah.

So what I heard there was a empty nesting software developer with a PhD said, you know
what I'm gonna do with my free time?

Is I'm going to go through all the work that it took to get my first PhD, but I'm not
going to make any money off of publishing it.

I'm not going to spend any money on well, other than the the existing subscriptions on
going to school, getting all the validations they require and everything like that.

And I'm going to end up building

processes that allow me to do the things that I would do in a PhD, but using AI and tools
that I build rather than using this the the existing structures of academic or academia in

order to end up with the same ish, you know, and we have to talk about what same is is is
and isn't same uh output of what that experience would have been like.

The you'll have the education, you'll have the prompts to think, you'll have the rigor and
structure, and you'll end up with a published work, maybe not published in the same

places, but a published work.

Is that a okay retelling of what you just said?

Okay, there are so many pieces to pull out of that.

Uh I'm gonna start at the end instead of at the beginning.

Uh the end result is you have a published work and you've done a bunch of learning along
the way.

Is there any goal with that published work being something that is uh influential and ref
ref ref referenced

Or is it sort of like that's sort of just a artifact and really the the the journey w the
you the the the journey was the friends we made along the way kind of thing?

It's I I wrestle with that a little bit.

I think the end result is very interesting and very compelling.

So I admit I would love for other people to read it or skim it or throw it into a model
and get digests.

Um so I think it is very interesting.

I I almost for sure, because I don't think my brain can stop, is going going to follow
some of the threads and try to

um continue the discussion elsewhere, but I but I I've been trying to know like when when
is the definition of done enough?

Uh I got to the point where I I felt I met the standards of what a dissertation defense
would be, and then I kept going because I just found more stuff that I thought was

interesting to fit into it.

Yes.

Totally.

So do you do you publish a physical book at some point?

Yes?

Well question mark question I um so I've on the I've been just doing EPUB for now, so at
least I have something that I can read on my phone and is kind of like done and for my

in-laws and stuff to read.

I am gonna do the book and almost a printed book, almost more as a forcing function, so
I'll stop.

That's what I was wondering.

Because I'm like, if it's an EPUB, you can just change it every day, right?

Yeah, yeah.

Okay.

So if someone were to take the particular generated EPUB that you have today and were to
translate it into a physical book, are we talking a 700-page-long, highly dense academic

thing that nobody but people in your specific niche of historian world could understand?

It's a little more public-facing.

Like, what's the general vibe of it?

And that's interesting.

And also I um it's it's just fascinating to see what the models do as you get conversation
going and you give it the the the per the boundaries of like of where you want to of what

you want to stay within, but I try not to give it um too much direction of structure.

For example, like if you're building an app.

I find you get just such better results when you talk about the inputs and the the the
result that you want.

When you start saying of like, here's how I want you to structure classes, the output
tends to be not as good because you're like fighting against how the models how the auto

regression works.

How it's like if if you're doing things like Dave's way or Matt's way, um, that's going
counter to how the models are trained.

on whatever style guides and design principles.

so I tried to do this similar thing for this book of I almost didn't I actually didn't
even care about the book.

Okay.

That's why I asked, 'cause I was like, do you or do you not?

I was I was very curious.

But Okay, that's what I was wondering.

That's what I was like, it's an artifact.

But did are are in I I want you to keep saying what you're saying, but I need some
context.

Do are you writing the book or are you prompting Claude or whatever else to write the
book?

Like who's actually typing characters into the final document that generates an EPUB?

The typing of the characters is uh is ninety nine percent Claude

Okay.

So what where is your input?

Is it are you building the you mentioned you got outlines, you kind of got note cards and
stuff like that.

Where does it translate from what you have taken to the actual prompting?

Are you saying, Hey, we're gonna work on page seventeen?

Are you are you saying, Hey, I'm gonna build an outline and you're gonna build the whole
thing from this?

Are you going chat like what's the actual translation process from your thinking to words
on paper, as it were?

Imagine you're reading you're reading a book for your like history class or whatever.

And as you go, you highlight some stuff and you scribble in the margins of why you care.

Like why did I highlight why did I bother highlighting this?

So that's kind of the style is I'm I'm walking, I'm listening to the audiobook, because
that's just how I work better off of audiobooks than off of reading books.

And then I pop into Claude for iOS and I have

a little back and forth with a long-running sonnet session.

And I say, okay, I heard this.

This is what I think this relates to.

For example, of I'm hearing about the in Russian colonialism and the way that the
railroads were moving south across uh what is now Kazakhstan.

Like that feels a lot like the railroads moving west in the in North America.

Uh and at the very same time, clearly these aren't

Coordinated parallel events.

And then Sonnet replies back of yes, there's there's some academic that talks about some
similarities in this ways.

And then I go, oh, that point of it, let's talk similar about that.

And I might have a 15, 20 minute conversation and using Sonnet as like a research
assistant.

Go find me some books about this.

Go find me some journal articles that are primary or secondary sources that.

Validate this is a real thing and not a hallucination.

And I have some concept.

And then once I have that concept, I take the the whole back and forth chat, which might
be might be two messages, it might be 10 messages, and I feed it into a Claude code

instance from my phone.

I shell into a computer that has a Claude instance waiting for it, and I pipe all that
content in.

you just hitting copy and paste to get it into there through the shell?

Okay, got it.

copy and paste.

And you're saying of like, well, how many of the words that are typed are mine?

It's about fifty percent there of because it was voice to text, so I never typed any, but
half of the half the content are my thoughts, right?

so we're saying fifty percent of the of what's being pasted in a Claude code comes from
what you're saying and fifty percent comes from your conversational partner.

And then where does it go from there?

It goes into another Claude code instance that's running on a Mac Mini at my house.

And that Claude code instance's uh job is to file note cards.

So it takes the back and forth conversation and turn it into a note card that says uh
timestamped when it was generated, what was the book, what was the chapter in the book,

what's the synthesis of the this conversation.

And if there's any prior art or prior journal articles, bring 'em in, put as footnotes and
do some some basic footnoting as one card.

Yeah, and pause you there.

So I know that you got to hear my previous interview that I had with an academician,
right?

And so you you don't kinda and so w the thing that was interesting there is he said this
is Nick Peterson for those who have haven't heard yet.

And one of the things he said was, Thank you.

he I think is brilliant, right?

Um, and one of the things that he said was basically there's always been this way in
academia where the person doing the thinking, the research, is relying on the other people

to kind of do the the low level work and the research and all the individual pieces and
there's the th you know, and so he's like, So

People are just replacing that low-level work with AI.

And this so far, what you're saying sounds very familiar, right?

You're like, I got my conversational partner who's my research assistant who does, and I'm
like, cool, with you so far.

I'm sure there's some people who would be like, well, that's different in XYZ ways.

And I totally understand.

But so far, I'm like, yeah, that sounds relatively similar to what Nick was talking about.

Now that you have the cards, what happens from there?

Okay, so um I would I'm reading, I'm listening to a book like it's one of my semesters.

And I'm so my conversation in the app is like I would be talking the prompt that I have
for the Claude iOS app is like it's a seminar professor.

And I'm like, hey, tell me about this.

We did that.

And then I file to my research assistant, like you just said, and then when the book is
done, or I finish reading that book for that semester, then

I go over into my research search assistant and we say, All right, how many cards do we
have now?

Okay, we've generated 30 cards from this book.

Let's synthesize it into um a a book outline or um read through them and tell me if you
think we have an at the very beginning from the first book, what do you think we've got

here?

And it went through and says, Okay, well, I think this, this, this, we at we did some back
and forth of questions and and

My default was I would rather start with outlines.

We can stub some kind of sections within chapters and then tell me what are the gaps, the
intellectual gaps.

And then I would go over to my Opus model, that was my dissertation chair, and I would
say, Here's like the outline and some stub some of some ideas.

Be um hypercritical yet.

Friendly and collaborative.

You're not trying to end my career, but you're trying to be uphold rigor and tell me where
I have conceptual holes in this outline or this chapter or this paragraph that me and a

sonnet model generated from all these note cards with some back and forth discussion.

The Opus model would say, This looks good, this is vague, this needs supporting evidence.

I would get back a report from that Opus model.

And then um me and my research assistant, we would open up a GitHub issue for every
criticism from the professor, from the dissertation chair.

And so after like a semester, well, maybe we open up 30 GitHub issues because of weak
evidence, whatever.

And it's all there.

So you can go, it's a public repo.

If you're if weirdly curious, you can go and look at all the closed issues.

And then I started to approach it a little bit like test-driven development.

I've got these conceptual holes.

Okay, research assistant.

We've got these conceptual holes.

Look up and recommend what might be a good topic for next semester.

Okay, next.

Exactly, exactly.

We'd go through the same process, we would do more note cards, we would synthesize the
note cards.

We would compare against all of the open GitHub issues, critiques from the advisor.

And then once we had answered not all of them, but at least many of them, we would do
another round with the advisor and just kind of rinse and repeat for about six books.

Okay.

So we have 'cause you you're fully aware of the level of criticism a lot of people have of
of AI, right?

And and so you're both doing something that is exploratory and creative and innovative and
then also can be scary and for some people can be very impersonal and human.

So I want to kind of poke at the so so we've got research research research assistants uh
that are AI.

We've got the advisor that is AI, um, which I think

some people might be unhappy with and some people might not care.

I think the biggest and first criticism that I we haven't still saw our way through is is
how much are you losing your own interaction with and your own thinking and your own kind

of innovation.

And I think it's very helpful for you as being someone who has gotten a PhD before to be
able to compare.

And obviously you're I'm asking you to criticize yourself, but still up to this point, we
haven't gotten to the point where it's actually a book yet, but up to this point,

the thing I think that was closest to being being the potential of, you know, Dave's not
doing the thinking is where the synthesis aspect.

So can you tell me a little bit more?

Like you've got all these cards that are clearly representation of your conversational
thinking together with Sonnet.

You decided which cards, you decided which books to read, you were super v involved, and
then the synthesis happens.

And is this is the synthesis, hey, you know what?

Like if I were to write a book, take all those cards and give me something, and you're
like, Yeah, it's good.

I'll make, you know, a couple notes on one of them.

Or or are you st still very involved in like the thinking and organizational processes
that happen?

trying to be as hands off as possible because it's a wild experiment anyway.

Hey, this is the rogue historian part.

Yeah.

So it's not just rogue in terms of what you're doing history of, you're rogue in terms of
your relationship to the content, the thinking, the the entire academic concept.

The I am being very intentional around the note cards so that each

Exchange in isolation.

Um, I read every word, I go back and forth a couple times.

So each note card I have I'm very familiar with.

The synthesis, I'm kind of trying to not guide it.

I'm trying to just out of curiosity of being like, if I'm almost like fine-tuning this
model off of all of these thoughts, all of this like

Now I'm at like three hundred note cards of all of these like Dave's thoughts on history.

On this broad topic that I'm working on.

huh.

Right.

And you're trying to write a book off put Dave's thoughts into a uh s synthesize it into a
book format, then do it.

And then I read it when when it was when I would have the the first few um semesters, when
the first few books, I read every word.

When I had like a couple chapters, I've just got a enough for a couple chapters, I read
every word, even a couple times and I'd like, I don't know about that, that doesn't quite

sound like me.

As we've gone further and further, and I when I spot check the pros, some I read more than
others, it sounds like me.

There are Claudeisms, of course.

There are Claudeisms, but it sounds a lot like me.

Maybe I'm adapting to the models.

but I think a I I've got now a lot of

I mean 50% of the source material is generally Dave's voice to text that goes into these
note cards.

So it's like if you're writing blog posts and you generate, well here's twenty, here's
twenty of Matt's blog posts that show the tone and the structure.

Now, if you go ask Claude to generate me a new blog post on this topic, you got a pretty
good chance it's going to come out sounding like Matt.

So so again, we still haven't gotten to the end, so I keep poking and you you know, I'm I
we are we are long to friends.

If you are not bothered to read every word that has been written, how do you ask other
people to bother to read every word that's written?

It is a good question.

One I with my eyeballs, I read very little anymore.

Maybe it's from staring at computers from work and code and stuff all day.

so um part of it is I I kept going.

So when I read the the version that was 140 pages, I was like, okay, well I read all this,
and now that I'm adding and now that I'm up to 400 pages, I'm like

Am I gonna read it again?

Am I gonna add something again?

And I can, I could, but I was like, half of this is from a learning uh learning
experience.

So what I'm my current thing, so that I'll actually stop it at 400 pages, I'm gonna read
the first.

I'll read some chapters I haven't read recently, but I'm gonna read some number of
chapters, some sample of chapters, along with this phonetic corpus.

for voice to text model training, and I'm gonna train a Dave voice, then generate the
audiobook with Dave's voice so that the first time that I actually listen to every word of

the book, my AI generated voice will be narrating it to my human self.

Okay.

So you're not like, I'm gonna put this out in the world as a final thing without having
listened.

It's just like, look, the longer it gets and so I've written a fifth 500 page book.

When I have to go through and make a massive change, and I'm like, I've read this chapter
twenty seven times.

The words are just blurring together.

So yeah, there's I'm not saying I never read it because I wrote it in the first place, but
if you get to a point where you're right reading the same thing over and over and over and

over and over again, that I get that.

That that's exhausting.

Um okay.

I'm gonna pause because I remember that I'm uh have new exciting thing that you're aware
of and nobody else's, which is we have our first sponsor.

Uh and I'm very, very grateful for people who are willing to sponsor my podcast.

I it's a idea I just introduced starting like a month ago.

Because I've done podcasts for decades at this point and never taken sponsors because I
was like, I don't want people to think that the only reason I'm doing this is

sponsorships, or if I talk about something it's not gonna be, you know, honest because
someone's paying money.

But what I realized is that my podcasts have also been inconsistent.

Um, and uh they're inconsistent because I'm doing it because I like I like doing this, but
sometimes liking do this gets in the way of like, yeah, well, I also need to make money

and I need to make Tighten successful.

Sponsorships help solve that because now this is work just like anything else.

And it's work I love doing and I'm so glad to do.

But once it is something that is work, it helps me say I can validate this as being worth
doing.

So anyway, sorry, Dave, to make you sit and listen to me talk about this.

Um, I'm very grateful for sponsors to make these podcasts.

Slot into something that makes it a lot easier for me to stick with consistently.

So and Mailtrap was actually my first sponsor for the Laravel podcast.

They reached out and asked to sponsor, and I said, you know what?

Hold on.

This is a great idea.

I don't know why I've done sponsorships.

So they were my first sponsor there.

And so they're now also my first sponsor for Pragmatic AI.

So today's sponsor is Mailtrap, enterprise grade email delivery infrastructure for modern
software teams.

Mailtrap helps teams deliver email reliably at scale with the security, compliance, and
infrastructure control modern organizations require, built for enterprise environments and

ready for AI era communication workloads.

Mailtrap integrates easily into inf existing infrastructure without slowing teams down.

See how it works at mailtrap.io.

Mailtrap, you guys are amazing.

I am so grateful that you are making all of these things a more viable uh possibility for
me.

And Dave, thank you for sitting here while I just blabber at you for a bit.

Okay.

Well, I'm think I'm thankful for folks like you that have long running podcasts that are
entertaining and informative.

I love that.

Okay, so we have gotten to the point where uh those uh synthesis happened, but a synthesis
is not a a chapter in an actual final book.

So there's still some process that goes from synthesis of uh a semester to chapter in a
book, page a book, whatever.

So what goes from from one to the other?

Or is it that's is that it?

It's is a synthesis just a chapter in the final book?

It is not necessarily a chapter of itself.

I might be reading a book and making note cards that are natural that naturally collect
into a chapter.

As I got further along, I would find more examples of things that I had already written
about.

For example, diffusion of innovations, the S curve thing, about why certain things get
adopted uh fast.

and

I would go through and I'd be listening to some other book and I would find some example
that I felt was related to diffusion.

And I would pluck that out into a note card.

And then we might end up with three or four different diffusion note cards, a few
information theory note cards, and then when I do a synthesis, uh it

Claude goes and it looks through the table of contents, looks through the other card
references, follow all the footnotes, and be like, that paragraph that had this example,

this is a thing that makes that argument more powerful.

This is more supporting evidence for that paragraph in chapter three.

And so it adds, it might only add one line.

It might only add a word and a footnote.

And the footnote is like every one of these cards, if you read the EPUB or look at it
online, it's it's pretty dense.

Uh as like a dissertation, you wouldn't expect that every single term is completely
explained, or else you would never get to the point of the synthesis of the dissertation.

But as you're going, you see a thing, a word, a reference, a mention, footnotes there, hit
them, and then you might get.

three to four hundred words that's explaining the reference for that note card.

And so that's kind of like what happens is the note card finds a place where there a word,
a sentence, a paragraph is relevant to support it, and then does all the linkages.

Okay.

So one of the things that was hardest for me in writing a book, uh, and I think one of the
places that I was proudest of my contribution was 'cause I wrote a technical book, right?

So what's the overlap between a technical book and technical documentation?

Hopefully you're not just you know, reproducing the docs.

And and I I I was able to say, Well, I'm not just reproducing the docs because the docs
were the thing I happen to write a book about

weren't as rich as the book is.

And so every time I found something that I was writing in the book that wasn't in the
docs, I would write it in my book and then I'd go add it to the the publicly available

docs to make sure that was the docs were better.

So I was like, okay, so part of that is I'm I'm learning how to write about it so I can
make everybody's access to it better, including putting it in book.

But the biggest thing that I was trying to say is the narrative structure and storyline
that helps people understand that regardless of how academic the thing you might be

teaching is or is not, or how structure how technical it is,

At some point, we are humans who understand things with beginning, middle, and end, with
rising climax and all those kind things.

Like, even if it's tech, we need to understand what's the story behind why we're doing
this and what comes after it.

Well, how does this piece lead to this piece?

And so figuring out how to do that, first of all, is tough overall.

You know, my wife is a screenwriter and she talks a lot.

We watch a movie afterwards and I'm like, something didn't click.

And so she's Yeah, it's because the the writing wasn't good and it didn't guide you
through.

And I was like, Great, that's it.

So understanding how to structure narrative.

storytelling, even when it's not actually a narrative form, to walk people through the
learning process or the, you know, the the natural conclusions.

This, then that, then therefore this, then that, you know, is a lot of it.

And so my fear if I were kind of stepping into the writing process that you were in, is
that I would end up with a book that basically its structure is not necessarily ordered

and narrative.

It would be purely based on

Well, that's what I studied in semester one, and that's what I study in semester two, and
that's what I stu so that's chapter one, two, three.

That's purely that.

And it may or may not actually be something that walks somebody through that way.

Um, is there anything you're doing to try and address that?

Or are you sort of like, I'm just crossing fingers and hope it makes sense?

You know it but oh god so several things to jump off there of one of um yes, writing
narrative prose is a massive skill and takes a lot of time and heart.

And my wife uh wrote five or six books, uh self-published author on on Amazon, sold a
decent amount.

We still get a little uh audiobook royalty check, so that it's it's sweet.

And I am not good at that part.

Uh-huh.

good at writing um at at prose.

I can write academic stuff all day long.

So I could have.

I could have done it all, but it the the prose part wasn't personally interesting to me
off this project.

It was almost an excuse, giving myself a framework to help me learn and think about things
and the and the the output having the look of a book.

that's the seeming like a book is just a is a convenient byproduct.

But but but i I I I do not at all um discount how hard that is.

How hard that is.

And so one of the c I did a couple things from the beginning, like I put Claude as a co
author, secondary author, right on the front.

And not nothing for me.

Anthropic didn't give any sort of provals here, so Anthropic, if you want me take it off.

But I'm

You're trying to be honest about what's going on.

Yeah.

like, I'm not trying to pawn this off as Dave had every single idea here, whether he spoke
it all or synthesized it all.

Not everything here is Dave's.

That's not the point.

It's more of a an experiment and and some curiosities along the way.

Also made it of putting it like just put it on the web, open access.

I put it with like an open access license from the beginning so that there's I'm not gonna
try to sell it, there's no incentive.

And even if I did it and I was gonna like how would I would I have sold any?

Who knows?

But I didn't want the subconscious things that you do when the output is for sale, that
you you do things to make the output more sellable.

And so I did some of these guardrails up front.

So if the output, if the goal was not sellable, but your goal was to create because we we
have both kind of I I suggested and you said, yeah, that was the the idea that the the

book is almost coincidental, right?

It is a structure and framework that is helpful for Dave to have the learning experience
and also the experimenting with AI experience that Dave wants to have, not because the

goal is to prep create a book.

If you were to say, you know what, I think these are really interesting ideas I have here,
and I would love to get a book out in the world.

that every person I know should read because it's so captivating and interesting.

This is such a fascinating idea.

I want I do want it to get it in the journals or whatever.

I do want it to get published, whatever.

What would you do differently about your writing process than what you've done this way?

Or would you take what you've done and then do something afterwards?

Like what you know, does that make sense?

question.

No, yeah.

It it's um I thought a lot about this.

So let's say that if I there's certain some subtopics that I would want to submit to a
academic journal.

I could try and I probably would try just to see of submitting it with Claude as a
co-author, just to see it.

Because I got nothing to lose.

I got I I I've I've uh it's not gonna affect me.

I've got no tenure packet that I have to to get through.

So almost just to see, but there are some uh topics that I find very fascinating that I
would want to dig deeper on.

And if I wanted to submit it to like a like a peer peer-reviewed journal, I don't think
they're ready for it.

Um, and so I would just do a standalone piece all Dave doing all of the writing and a lot
of those journal articles are eight to fifteen pages on very, very, very narrowly scoped,

and I would just hand roll all that stuff to go to a peer-reviewed journal.

Yeah.

You'd take all that thinking that you've already done, all that planning, you'd use it for
structure and you'd use it for the but you didn't you'd just go type something.

Okay.

Yeah.

I'd take the four hundred page book and be like, well, I'm just gonna do a thing on
diffusion in uh Han China.

Okay.

We pull all the note cards that are related to that from my research search assistant and
be like, all right, well here's the the the thirty.

Let me write the prose that stitches all this together and then go get some additional
footnotes.

Okay.

So one of the things that's fun, and this has been true since I've known you, is that you
get excited about something and there's just a little bit of glee.

Like I hope y'all are being able to watch the new tube.

You can see glee in your face as you get to share about this thing you're doing.

And you're like, Yeah, I did that.

And you're like, you know that someone's gonna be like, What?

What are talking about?

And there have been multiple kind of Dave Glee moments.

one of the Dave Glee moments across here was when you talked about everything you're doing
with your phone.

so you remember that there's a lot of people here listening who are super mega nerds and
gonna be like, I want to figure out how to do that.

And there's also a lot of people listening here who are like, I use Chat GPT once a week
most, right?

So we've got a huge range of people's comfort level.

So let's talk a little bit about what um not primitives, but what tools do you have in
place that allow you to be as productive as you are.

And some of them you built yourself.

But if someone didn't have custom built tooling,

What would you be like here?

Like the couple things that people should be doing on their tool or on their phone to have
a much better kind of like set of interactions with AI than the average person has.

I didn't s and I didn't start by building any tooling.

I started with take stuff that's off the shelf, fiddle around with it and remix it, find
where there were friction points for what I wanted to do, and then I just kind of sawed

off the the the pokey bits or the rough edges so that I could do the thing that I wanted
to do faster and get more to the fun part.

So the way um the the basic setup that I have

I'll also I'll do of when I started.

The basic setup that I started with is I use uh and no sponsorships, um, but super whisper
for iOS and Mac OS is just an awesome voice to text app.

I pay for it because I use it so much.

You just hit command spacebar and then you can just talk and it kind of gets pre-polished
through a model of your choosing.

So using a voice to text for me is very important.

I really appreciate the initial polishing because it gets me from being where you're hung
up starting because you want to get the exact right word.

And I found liberated by using the voice to text, I care less about getting the right
word.

If I can just word vomit the concept, the model is gonna get pretty good at um cleaning it
up for me.

So super whisper, then I've got Mac Mini, but I don't use open claw.

I just have a Mac mini with a bunch of Claude code sessions running in different terminal
nope.

a lap okay.

So you have for those who are not familiar, it's a increasingly common thing for
programmers to have a second computer.

The Mac Mini is a very small uh Mac device that doesn't have a screen, it doesn't have a
keyboard, it's literally just like a little brick that you put somewhere.

And I'm assuming yours probably doesn't even isn't even connected to a screen and
keyboard, you're just c remote connecting it from other machines, right?

You do.

Okay.

Yeah.

need to do system updates, but I generally only remote connect because it's another of
force constraints on myself, uh create constraints breed creativity.

interaction with it.

Yeah.

Okay.

So he's got a little brick computer sitting that its pure goal is just to run a bunch of
AI sessions.

I assume there's a combination of the programming power, the fact that it keeps running
when you've closed your laptop, and also it's safer because it can't access the files in

your laptop.

But I'm actually curious to stop you there.

What does motivate and what are the benefits of having this separate machine that's doing
this all the time versus just doing everything on your laptop?

well I guess t the way that I have to reach it is I have to I have I use SSH to reach it.

So I use from iOS.

SSH is almost universally going to be if you've ever seen the movies where you've got a
terminal where it's like green text on on a black background, the majority of time someone

is SSHing in, it's just typing little terminal nerdy stuff, but instead of typing it into
your computer, you've run some things, so now what you're typing is sent to another

computer.

Continue.

familiar with remote desktops.

It's like remote desktops, but for the nerdy typey thing.

But just but the similar concept.

I have to be able to um shell into the into the Mac mini.

So you have to use some sort of cloud proxy to do that.

Common thing.

I use Ngroc.

I've used Ngrok for a long, long time and I like it.

Other common ones, a lot of people use Tail Scale for this.

And so to be able where you can shell into the Mac Mini.

Then at the Mac Mini, the reason I run it as a Mac Mini versus, say, like on my other
laptop, is the Mac Mini needs to be running all the time so that you can shell.

Um, if you close the laptop, then you can't shell to it.

And I want it available so that if I just have an idea of a thing that I may not spend
sometimes I'll spend three or four hours straight of of of talking and iterating on an

application or some note cards of the book, and I want to read everything synchronously,
sometimes

I just need to update a link and I can just shell in and say I update this link on this
website, go.

And I don't wanna have to like spool up the whole system.

So that's that's the big reason.

and that this this shelling concept for for y'all.

So a lot of us are like, well, if I'm not on my laptop, that means I'm not working.

And one of the things that often happens when you're building with these kind of workflows
is what if you're out and about and you have an idea and you want to tell the thing to to

go?

Or what if you're working and you have about to have to close the thing and go, right?

And so there's this constantly running access to the systems that are happening.

allows him not just to say, well, it's on my laptop.

So if I'm not he's Well, talking about walking around on your phone, your phone needs to
be able to connect.

And what if your phone's trying to connect to your laptop that's closed in your backpack?

So is this benefit of something that is persistently available, space for these kind of
these things happening.

Now I before you go any further, I do want to check, do you have security constraints
around it?

or is it sort of like, no, everything that, you know, could happen on my laptop could
happen on my Mac mini.

Well, it's different because I I don't install or authenticate everything on Mac mini that
I would have on my my laptop.

So it's kind of like having an admin user and a standard user kind of a thing.

Okay.

All right.

So I I interrupted you.

So you mentioned that you got Super Whisper.

You've got the constantly running Mac mini that your phone and laptop can SSH into.

So that's kind of the place where your all your AI agents are running.

What else is a part of your initial setup?

Or is it mainly that?

W the the next thing you kinda have to have is using tmux in the terminals.

So so tmux is a way that if you've got your little tippity tappity shell thing that you
can name that window, that Claude session, you can name it so that you can attach and

detach to it over time and from different devices.

For the folks who use if you use Claude code today, um

And you see like the re what is uh what is it called?

Remote control in Claude code, how that you can do like a little slash remote control and
it gives you a URL that you can then launch inside of the Claude for iOS app and it will

tunnel into Claude code.

Yeah, on your local machine.

Onto your local machine.

That's basically what I'm doing.

But instead of uh having to fire up, get that link from Claude code every time I want a
remote control, Tmux gives me a wrapper around each Claude session that is named that I

can reach 24/7 365.

Okay.

I so I am gonna admit I've been around the internet for a long time.

I've been on the terminal for a long time and I've never gone down the TMUX route.

I even I'm even a Vim user and I've never done Tmux.

But I was part of the reason why was I was always under the impression that the detaching
and attaching stuff like that was purely for the sake of your own local computer.

So I was just like, it's a really convenient terminal window manager.

Great, that's fine.

I've got iTerm.

I don't need that.

I didn't know that.

there was anything about connecting it from the outside.

Is TMUX providing the facility for that or is there some Okay.

imagine you what you're just shelling.

So if you had two computers and you shelled into your work laptop and then you attach to
the TMUX session, well now you have two machines that can attach to the session.

So it's basically programmatic programmatic access to be able to say, I want because like
right now, I've I at any given time often before I started using the tool solo, which we

may or may not talk about later, I would have 10 tabs open by iTerm.

And so SSHing in and being like, well, grab the contents of that one tab was not something
I could have done.

I would have had to actually have a visual window where I'm actually screen sharing and
then saying I would click on tab number nine.

Whereas with Tmux, you can programmatically say, I want to be on tab number nine and
you're na now able to interact with it, even while theoretically the iMac itself is also

attached to that same uh Tmux session.

Okay.

going two down that nerdy rabbit hole and for anybody who wants to like chat about this
other times, I've got s a decent amount of blog posts written on my site as well that kind

of go through this local setup.

The there's a lot of configuration that you can do.

I have optimized it all for one Dave that maybe is almost always connecting from one
machine.

And so if you need to like detach other machines and let this machine attach to it, it
just makes it more resilient than trying to like multi connection stuff.

Yeah.

Okay.

Um, I won't go far in this, but I have to mention that friend of the show, Aaron Francis,
has created a program called Solo.

And if anybody's interested in learning about this kind of stuff, but they're like, you
know what, I might want it at a more somebody's managing it for me, you know, graphic user

interface level.

The fun thing about Solo is it's like iTerm or whatever else, Ghosty, where it you've got
a whole bunch of different tabs that are each terminals.

But solo exposes an MCP so that every single instance can talk to each other like and the
agent control them.

So that's where I've been explor exploring.

Um but I definitely gotta look in your T mux setup.

Yeah, solo's super fun.

It's super cool.

And think if for most people, I acknowledge that I'm a weirdo in this way and I want to
try to do everything from my phone all the time.

If you're sitting at if you're at your laptop, most people want to see stuff most of the
time.

Something like solo is amazing.

Yeah.

Well I have s I have tail scale and screens or whatever it's called, so I I can control my
computer for my phone, but trying to do anything on a massive widescreen controlled from a

little tiny touch screen on your phone, you know, it's not quite the same.

But I mean that's uh what's interesting, but I've I I've I do just find it fascinating of
uh constraints the the constraints and I do most everything from voice.

I type very little.

I built a custom iOS app for my shell app so that it would have larger buttons, Claude
specific one, two, three escapes, hooking it into super whisper so that I have fewer

Switching between apps just make it much more seamless for me to do how I have evolved how
I work.

And then when I I don't really the only stuff that I would really want to see is like I
want to see the end result on a website sometimes.

Okay, sure.

Do I want to see I used to want to see more like tailing standard output or looking at
logs or traces, the observability stuff?

Uh what I've started to do is

I still have it available, but I don't look at it while I'm building.

I run a I set up a separate project for a local observability stack so that all of my
Claude instances and all my apps send OTEL to this open telemetry standard stuff to the

stack and so that when the Claudes are doing their thing, they know they can go look into
the whole observability stack and just trying to shorten the feedback loop.

Yeah, totally.

The a big takeaway for people who aren't maybe like professional software engineers or
infra people is that the models work so much better when you they have short feedback

loops, in my experience.

And so I just want to make very, very short feedback loops.

People are better with short feedback loops.

And so where I might ask to build this butt add this button that does this thing.

All right, well, it takes 10 minutes to build it, run some unit tests, ship to a website.

A lot of things went could have gone wrong if you've got access to all the open telemetry
so that when the tests fail or you're running into integration tests, you can just go

fetch the traces from there, you get a a much cleaner output the first time through.

Okay, so if somebody wants to just start down this world where you're like, hey, I'm doing
all this stuff for my phone, we've got if you're a coder, the potential of having an

always-on machine that you can connect, whether it's connecting through something more
rich like Tailscale and screens or something more narrow like either Tailscale or Ngrok

and SSH.

You've got uh voice to text.

So you mentioned Super Whisper, although I think you said there's something else you're
using today instead of Super Whisper.

You've got one of the more rich voice to text.

Now, for those who so my wife uses um Mac OS's built-in voice to text all the time.

I use Mac OS's built-in just I mean all the time.

People are like, my god, you do it more than anybody else.

I know you just say, hey, S-I-R-I, I'm not gonna do it right now.

Remind me to do XYZ.

They're teasing me all the time.

Uh I do use super whisper on my local machine because I've found I get better formatted
output and I spend less time saying comma, semicolon, space or whatever.

I've never tried it on the phone.

So regardless, for someone who's never even heard of this tool, what is the benefit you're
getting for these AI-backed uh voice to text versus just a more traditional one?

essenti what you just said that I don't have to deal with um punctuation, I don't have to
deal with the like I don't know, I guess structuring the prompt well.

I can just

you don't like saying exact no exact word you're gonna say.

So you could just kind of word vomit and it's f turns into something useful.

Yeah, okay.

Okay.

so that's the big thing, and I find it's just using um that S I R I ain't great at that.

Hopefully they'll do a good update in was it twenty seven coming up.

But there's just like a lot of misses and so uh it's so that I don't have to go back and
do too much editing.

I uh that's why I use one of these other apps.

Yeah.

And I'm I'm a big fan of that.

I mean, regardless of whether or not somebody likes AI, I have found that um things where
it's just a little bit of cleanup um is one of the best options.

It's just saving me time.

Like I I recently was um making a slide deck and this I I was using software deck set that
lets you re uh build your presentations in Markdown.

And so I had my whole all my stuff there and then I was gonna do the part that I don't
like.

I was gonna spend an hour doing the part I don't like, which is turning my markdown into a
perfectly formatted whatever.

They had an AI skill that Dexet publishes, and it's like, hey, turn this thing into this.

It was my deck, my content, my writing, my authoring, but I didn't have to spend 45
minutes doing rote crap.

And I was like, yep, this is what I want AI for, you know?

So, okay, cool.

All right.

What other things uh I mean, we went down a 45-minute rabbit hole that I loved about

The rogue historian and the book and all kind of stuff.

What other things are putting that glee in in Dave's eyes?

Like what other things, whether it's related to this or not, that are just like, Man, I
just love I love this.

I love telling people about this.

This is such a fun place.

I'm exploring.

What else are you doing that direction when it comes to your creation, your free time,
your AI usage or anything?

well.

the other project I'm working on right now is a a model evaluation service.

So so the idea is that if you're whether the it's a a local model or whether you're paying
for a cloud model, which model to use for which sort of tasks.

I have some day job reasons to do this.

I've got um a a thing on the side that I'm fiddling with as well.

So of building this service, and it's it's fun of of

Where you create a task and then the task gets fanned out to n number of models.

I might run three or four locally, small, medium, and large, and then also send a couple
off to uh anthropic, sonnet, and haiku, and then get back the response, same input, get

five or six outputs, and then you see them all, and then have an interface for doing um
evaluations and scoring.

So that you can start to build a corpus of for these sorts of tasks, what are the the
right the best models?

And it's funny 'cause you said module evaluation, I'm like, are you trying to be the
person that says these models are good for this globally?

And maybe you could, but you're saying more each person kind of understands for the type
of task I'm doing.

Yes, I and and this is an open source thing as as well.

So it's on on my GitHub of my my day job purpose for this is that I want to be able to do
like site reliability engineering stuff.

I wanna pull down a bunch of logs or metrics or traces and start to do things of I start
to see some this latency is spiking on this operation and this monitor is alerting.

Is there a correlation?

Yeah, yeah, yeah, yeah.

probably correlated.

A small model is fine for that.

And I can run automated loops that will do some of this correlation stuff.

I get into something more complicated, like if you're shopping for you're going for trying
to get um book travel for an airplane, and you get a burnt red banner that says temporary

system processing problem.

You as the shopper are pretty annoyed.

That thing pops in my console.

Yeah.

Uh-huh.

being able to using like larger models about being well automatically let me try to pull
that error code.

Let me go look through the code base of the service where that was where that was logged
and try to find actually the straight up like line of code where there's a missing like

catch for that error.

Um and so those sorts of things you need

bigger models and so I'm building this whole evaluation tool that hopefully we can use at
work, but I'm doing it for fun stuff as well and for learning purposes and it gives me

glee.

Um so and it's so it's there if somebody else wants to fiddle with it as well.

I love it.

Um, this reminds me every single time I've given a talk, because I I've started giving
talks to companies where they kind of bring me in and they say, We you know enough about

AI, and I'm like, I don't know as much as you think, but whatever.

They're like, You enough about AI that we want you to come in and kind of talk to our team
about uh what should we know, what should we not know, what should we care about, what

should you not care about, how are ways that our specific uh industry, our specific needs
are kind of best met or not met by AI help us really kind of come up with a very

pragmatic, but I'm ching um kind of approach to this.

And one of the examples I often give to give people the vision of what these practical
day-to-day interactions are is I have a tool called Built with Laravel where people can

suggest sites that should be included in this, you know, hand curated thing.

It's hand curated.

I'm not gonna have AI make a decision for me.

But man, when I first opened it up, I got 250 submissions and I could just barely work
through all of them because each one I'm doing these same steps over and over and over

again.

So I built this workflow and it says, hey, go out on and figure out their traffic and
figure out d

Do they meet these particular metrics?

Like one of my metrics was I'm very interested in things that a VC or PE person trying to
decide whether or not Laravel was a good thing to let their portfolio companies use, would

they be impressed by seeing this or not?

Well, it turns out that you can actually ask AI those questions and get pretty decent
answers.

So I'm still the human curating.

I'm still the human kind of making these decisions.

But when I make a decision, instead of getting website, I now get

Website and a little mini report from Opus that says, we think this one is a four out of
ten likely to do this or whatever.

And I've found that it's like things that allow humans to well, it's the same as what I
said before.

You're spending less time doing the rote crap.

We're spending less time doing repetitive research.

We're spending less time doing repetitive typing.

It's not taking our soul and our humanity and our our you know instincts and our vibes or
whatever away.

It's allowing us to.

Hone those and spend more time on those and spend less time just doing rote crap.

So hearing you say that I'm like, yeah, that's that is the dream for me.

That's what I want it to be.

Like an SRE, you still are the one who's going to evaluate what the problem is.

You are still the one who's gonna recommend the solution.

You're gonna recommend a solution that, given your decades of experience in your career,
is the best recommendation for this thing going forward that makes it least likely for it

to continue to break now and in the future.

But you now have more information to more quickly.

make those decisions than you did before, right?

uh I can go and see that for this error, here are some logs, here are some sessions,
here's the whole request response payloads, work with Opus, put it all together and create

like a little Google Doc one sheet with examples and all the stuffs, and then I can go
talk with the dev team that's responsible for that service.

Whereas in the past, you style a bug report, you give a couple session IDs and you and you
um

You kind of just file it into a backlog and it probably unless it's critical, it gets
looked at whenever there is time, which is never or or way later.

And if I can do I'm I'm not gonna find the optimal solution.

I'm not trying to.

But if I can get it eighty percent of the way there and narrow down where for the devs to
look, I'm saving them so much time and like help me, help you, help help the customer.

help you.

It's exactly it.

You're like, hey, if if if if solving this problem that Dave cares about now becomes 80%
easier than it was before, it's better for them.

And it's much more likely this is actually gonna get solved.

So it's like everybody wins.

I love it.

Okay.

we're at an hour.

So we're gonna get close to wrapping.

But before I ask you some of my kind of like end questions, I just want to check in with
you.

Is there anything that you were really hoping we get a chance to talk about today or any

Furtherances of the things we've started that you want us to get to before we wrap?

No.

uh well funny that I did as a I'll I'll make a mention that I did funny as a little bit of
preparation to come and talk with you and trying to get like getting a little bit into

podcast voice and try to slow down the speaking like when you're doing a conference
presentation, you have to be like talk a little bit slower.

I put I've got the whole book and I've got all these notes and I did another little Claude
session.

And I was like, hey, I'm going on a podcast in a in a week to talk about this.

Here are the things that I think are important.

why don't you some role play?

And I did a simulated podcast transcript and posted it on my website.

it in a blog.

I love that so much.

And it was kind of weird because I'm like, wait, that's that's me.

I'm the I'm the pragmatic AI and I can't read too much of this because I've gotta actually
be real But it was fun, man.

And I mean you you you said you liked the Nick Peterson episode.

He's the one who told me about saying, Hey, you know what?

Like if I'm waiting for a human to review something and it's gonna take three weeks, why
don't I first throw it to an AI and say, Hey, here's who you're pretending to be in this

moment.

Let me do a first round with you so that by the time it gets to the human, they're getting
the

the stuff that is again more unique to them versus the rote stuff that I could have got
out of an AI or something.

Oh, totally.

And we all know this in in in work and in life.

There's the little tiny thing that really doesn't matter to the to the to the big picture,
but people just latch onto it.

That word choice, that color, that whatever, that position, and it disrupts the whole
value of the human to human conversation.

If you can refine any of that sort of stuff ahead of time, it just makes I think it brings
more uh like brings more humanity to the like

to the exchange because you don't have to have the humans just doing the rote, grunt,
grinding that you used to have to.

Yeah.

so I don't ask this all the time, but I do kind of want to know from people generally uh
what they think about the future given kind of where we are.

And I sometimes we'll have a conversation like this where it's just so pro AI the whole
time.

I sort of assume, but I don't want to assume.

What do you think um the future of our world is given the current position and trajectory
of AI?

Um, and does that make you hopeful?

Does that make you nervous?

Does it make you a little bit of everything?

Like where does what's

10 years from now.

I mean, your kids, you know, you said you're nephew and Esther, so your kids are out in
the world.

How do you feel about all this?

Definitely a little bit of everything.

Yeah, okay.

I I I'll generally position myself as an 80% optimist and 20% pessimist.

Different domains I'm more optimistic about.

I I'm hopeful the the end res I'm optimistic that the end result for at at the macro level
for the human experience will be better in ten years.

There's a big difference between macro and micro stuff.

A lot of people are going to suffer in the transition period.

That makes me unsettled and uncomfortable.

However, I feel it's inevitable, like with other big technology transformations steam
engine, locomotive, electricity.

etc.

Big disruptions that a lot of people hurt a lot during the transition period and beyond,
but the overall safety and uh welfare of humanity at large at the big macro level improves

over has improved over time life expectancies and right the overall ability to

Do the things you want and take care of your family and uh live a meaningful life
improves, though there's gonna be a lot of pain.

And so I don't want to just discount or or um gloss over pain that will be suffered.

Yeah.

And it's very and I mean, I'm not a historian.

Uh I do know one.

He's on my podcast right now, but I I as a non historian it does seem to me that normally
when someone says, you know what, some people are gonna suffer, but it's for the greater

good, there's usually like that might be okay ish on the the top of the thing, but when
you really kind of like dig into what they're feeling, they're gonna be like a lot of

other people are gonna suffer, but I'm gonna be better off, so it's gonna be okay.

And I'm not hearing you say that.

Not at all.

And and I think that's a great would often this the the follow-on to that is the well this
is inevitable, so get out of the way.

I very much disagree with that second part.

Everything needs pushback.

Um and getting back to the history book.

That's the big thing in the history book is that people it's it's masses of people and
masses of energy.

that over millennia move for different reasons, geographical, technological reasons, ebbs
in ebbing and flowing of emp empire powers and stuff.

But just because something is coming in doesn't mean what was there just should go
completely away and get washed over like a tsunami.

It's healthy to have pushback against things.

So folks who are the maxis that just say, this is gonna win, stop standing in our way.

huh.

the we think this is a good idea, but please push back, criticize, debate.

That it leads to healthier ideas and healthier outcomes.

Yeah, I like that.

Yeah.

Um, okay, so uh one of the things that you know because you've listened before is that I
have been asking the community to say what are practical ways that you're using AI in your

day-to-day life, so that can we can read at the end.

Today I'm gonna flip it on its head a little bit, and uh I did not tell you this was
coming, so apologies.

But I read a poem, I think you would call it, called Please Use AI.

and I don't think I can screen share, so I'm just gonna

Try and put it in the notes.

Let's see.

I've never here.

There's a chat.

Let me hit just you and Dave.

I it says I'm in there, but there's nowhere for me to actually type.

There we go.

Okay.

So I'm going to throw this in here so you can have this open as I'm looking at it.

I'm not going to read the whole thing, but I will at least kind of give the premise.

And then for listeners, this is called Please Use AI by Sean Smucker, and I'm going to
throw it in the show notes.

Um, so he says, uh, be sure to, and this is written in a

poem form, so I'm not gonna give you the the perfect presentation.

But he says, be sure to use AI when making your next, I don't know, meal plan, for
example.

Definitely do not call your friend who loves to cook and ask her for her favorite recipes
or tips or ways to save time making meals.

Because you'll end up talking for longer than you'd hoped, hearing perhaps about her
father's cancer diagnosis or how lonely she's been, or even what she's planting in her

spring garden and then lost with the early frost.

And be sure to use AI when planning that next camping trip, the last one you will take
with this particular child.

Definitely do not text your friend who's fly fished every river in Pennsylvania and biked
every backwards trail because you might end up texting back and forth for the rest of the

day, or even meeting up later for a beer and hearing how he's he ended each recent night
blackout drunk, or perhaps you'll hear how his cousin is an idiot on Facebook, or maybe

just that he repaired his own washing machine and pretty damn proud of that.

And then this poem just goes on and on like this.

And so it was so heart-wrenching and beautiful um to think about the ways where we could.

replace humanity and replace connection and replace intimacy and replace those kind of
like fortuitous moments of human connection that come from these other things bit by bit

by and it's not even necessarily AI because it could be a Google search, right?

And I mean we could have do it you and I know you well enough to know that you and I could
do an entire podcast just talking about human connection and culture and sociology and and

what modern connections versus you know pre-modern connections look like.

So as I end every single one of these with what's a practical way you've thought about
using AI?

I also there's a part of me that's like, what's a thing that you choose not to use AI for
so that you can retain humanity?

So again, I'm putting this on you on the on the spot, but I'm curious for you, is there
anything that you have tried and stepped back and said, you know what, that's not for me?

Or is there anything that you're like, you know what, this is a this is a habit or a
tendency that I'm actually trying to unlearn?

Or anything that you watch out for other people as you're

t having your kids go out in the world using AI that you're like, I want to make sure they
don't do this.

And and you could tell me, no, that the prompt doesn't trick trip anything for you, but is
there anything that brings to mind for you?

I'll I'll redirect it a bit because of inspired by the poem and of your reading of that,
because it is beautiful of so maybe not a specific activity, but um that I do differently.

This this book thing, and as I've been going through it, I do a weekly call with uh aunts
and uncles that live in different places that I haven't seen, and we started doing this

during COVID, and we still do it.

Every single Tuesday, six PM, we get on and we talk.

Nice.

And just talk about anything or nothing, news, politics, sports, ping pong.

My aunt's like huge into ping pong, and she ends up telling me a story a bunch about of
her ping pong training, loves it.

And and like that I do all the time.

And part of uh in our conversations, they will sometimes read some of the the the book
that I've written or certain things that I've written and they'll ask me questions about

it.

And I find that's kind of been very interesting of these are not

techie folks, what at all?

And they're asking me questions and and trying, they're trying to make sense of their like
70s and 80s of the world that they are, they've they've kind of what is this world?

How is it changing?

Is this good and bad?

Is this thing that I'm seeing in the news scary?

Hey nephew, what are your thoughts on it?

And it's opened up some really interesting conversations that aren't just

like technical or logical, but that are very philosophical and and emotional and it's been
um very nice over the last couple of years to be talking about some AI stuff's for good

and bad.

Yeah.

I love that.

Um, obviously this isn't your point, but just telling me that, I'm just like, Man, I've I
have an aunt who I love who I have not talked to in years.

Actual years.

I'm like, Yep.

I'm gonna give her a call when we hang up here.

So whatever the output of that thing is, it certainly has has has done a good thing for
me.

So thanks for sharing that.

I accidentally used AI in front of one my kids for the first time last week.

He was trying to figure out what is the cheapest way.

He's really into soccer.

He's like, what's the cheapest way I could possibly start recording myself as a goalie?

Um so I can get videos together to start seeing like what did I do well and what I need to
grow in?

And he's like, I want a GoPro, but I don't have GoPro money.

And so he was we were doing research on like what what is the cheapest combination of like
uh some kind of GoPro knockoff and a good enough but only just barely good enough memory

card.

And what is the smallest size of memory card he can get that will hold, you know, a full
hour and a half game of footage at the quality level.

And so it's all these very technical specifications.

And at some point I'm like, my brain hurts.

And so I was just like, you know, without any thinking, I pop over in chat GPT and I say,
We're picking this camera.

Here's the specifications of its output.

we're picking this memory card.

Please make sure that this particular memory card has the speed that's needed for this
particular type of output and make sure that the size that we're doing will give us an

hour and a half of footage at this particular speed.

And it's like, bip, bip, boop, yep, it was good to go.

We might have done one other interaction and I close it.

And then later I see him Googling Chat GPT.

And I'm like, what have I done?

Um, so I don't know where this goes, right?

teach 'em, the kid on the streets they're gonna teach

Yes.

And I hear that.

And I you know, I I I don't know if you got to hear it, but I had a interview with an
elementary school principal where she was very intent on saying we want to teach these

kids how to think for themselves and have, you know, critical thinking stuff like that.

But also like he's thirteen.

And I'm like, I hope that the critical thinking skills are already there because he like
you said, like he's his friends are already doing it and so but like we also we don't

allow them to use social media.

So just because the kids around him are doing it doesn't mean he gets access to it.

So there's this whole world of um

Who do I want to be around my kids?

How much do I want to use my phone around my kids?

Right.

Like that.

And and this has now just added a whole stack of new, new things that I don't have answers
for.

So my answer to my own question is I don't know what the hell I'm doing.

Wish me luck, you know?

So all right, Dave.

Yeah.

If people are fascinated by everything you're doing, obviously we'll put it in the show
notes, but I'd like to have you say it out loud.

How do they follow you?

How do they keep up with what you're doing?

you can find Dave Stanton dot com.

I post most of the stuff there.

And that probably find all the the links to GitHub and on socials it's go to plan B.

So if GitHub or Twitter.

Okay.

Well we'll link all in the show notes.

and Dave, thank you so much for hanging out.

I really appreciate you being here with me.

This was a great time.

Yeah, super fun.

Thanks, Matt.

Awesome.

And the rest of you, we will see you next time.

More episodes

Chapters

Creators and Guests

What is Pragmatic AI with Matt Stauffer?