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.

And welcome back to Pragmatic AI where we talk about using AI in the real world.

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

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

And my guest today is my, I'm sorry guys, I keep having to say my old friend, Ian
Landsman, a builder and a business leader, the founder of Help Spot.

Ian, would you mind saying hi and telling everybody who you are and what you do?

Hey, yeah, Ian Landsman, yeah, 20 years ago founded Help Spot, which is help desk
application, and got involved with Laravel.

Obviously, Taylor, Eric Barnes, Fidao worked at Userscape at various points, yeah, run
Lara jobs in terms of the Laravel world.

So, fan, friend of the show.

uh

Been around for some stuff.

So yeah, happy to come on.

very wonderful to have you.

And I would say about two thirds to three quarters of the listeners of the show are
intimately familiar with who Ian is.

And then a third of them, who I've been trying to kind of expand the audience to be
outside of just our little tech world, might not know.

So I will give a little bit more context.

So my day to day job is in an environment called Laravel.

You've probably heard various guests mention Laravel in the past.

And it is a particular set of framework is what we call it, a chunk of technology.

that allows us to build applications a certain way.

And Ian built a successful SaaS, if you listened to the last episode, you know what a SaaS
is, but it's a successful web-based software platform 20 years ago, back when it was

really manual labor.

It was hard work, you know, writing every freaking line of code, nothing helping you, not
just no AI, but not even like pre-made packages to do a lot of the

A lot of the stuff that today we just say, I just install a package to do that.

Ian wrote by hand when he was building Help Spot and stuff like that.

And Ian had a vision early on and employed people, including Taylor, the guy who founded
Laravel.

And he said this Laravel thing is it.

And said, I'm going to pay you to both work at Help Spot and integrate Laravel in there.

Also, I'm going to pay you to keep developing Laravel.

So I have been calling him the godfather of Laravel for ages, both because he loves the
Godfather movies and all things movies and, know, mob Italian.

But also because I don't think this entire section of the industry would have existed,
maybe not at all, because Taylor's amazingly brilliant, but it certainly would not existed

at the same timeline, at the same everything like that without Ian's support.

So this guy is an OG of the internet.

Also, you might have heard previous podcasts talking about bootstrapped SaaS, people
coming up with a software as a service without any app exterior funding.

Ian also had a podcast for probably over 10 years talking about what it looks like to...

make your own kind of bootstrap sass.

So one of the reasons I'm giving all those context is I want you all to know where the
voice, the importance of his voice is coming from.

But part of that is because I want to understand that I'm one of the large reasons I'm
excited to hear Ian's voice about AI is because he's been on the internet in technology

building real software that real people use for decades.

This is not Silicon Valley hype boy, you know, I just get excited about things because

I'm excited.

like Ian's excitement to me has an importance that other people's excitement doesn't
because Ian actually ships.

Does that make sense?

So anyway, Ian, sorry to kind of put all that on your shoulders, but I really do want to
say like my first question for you is just like as someone who's been building for 20

years, I feel like you could be the first person I would expect to say, you know, this is
not how we did it in my day, right?

Like, you know, like I know how to do it the right way, whatever.

And you

but you end up being a very pro AI voice.

So could you tell us a little bit about like, was your journey?

Like, did you start critical and get hyper?

Like, what's your journey been of your relationship with AI?

no, I was all in from the very beginning as soon as chat, chat, GPT three or whatever the
one, you know, November, a couple of years ago, they launched, kind of got everything

rolling.

I was like, this is it.

Uh, this is what, this is what we need.

Um, yeah, I think part of it is like, I come from a different background than like the
average software developer.

And so I've always been more on like the business mindset with it.

And

I programmed both, you I enjoyed programming, but it's also a means to an end.

And, you know, I wasn't trained as a programmer.

I was an accounting major.

So to me, it's like the code mostly has gotten in the way of achieving things in the sense
of like, if I, if I couldn't learn it or I didn't know it well enough, or it just is

slower for me to, to produce then.

that's delaying me or causing me to not be able to do things.

And with the AI that removes a fair chunk of that, not all of it, but it removes a lot of
that and uh kind of shifts the work more into an area that I prefer, which is like

thinking about the product itself, how it works, how the interface works, what it does as
opposed to the actual syntax.

What is the actual language, statements required to do this?

Do I remember how that all works together?

What the edge cases are, blah, blah, blah.

Yeah, and I think for founders with somewhat successful businesses, also think there's
just, once you're at that point where you have paying customers, there's just so many

distractions that keep you from developing software yourself.

And...

because it's just stuff to do.

There's just always a customer with an issue or some business thing or some bill or some
problem.

And so now when you don't have to be all the way into the zone to get something done
coding, you can kind of like pop in and out of it.

Yeah, it just feels so much easier.

I mean, I think David Kramer is as named the Sentry CEO, um which is like a bug logging
tool.

Like he's coding a ton.

Like I follow him on Twitter and he's always like, I'm coding a ton, right?

Cause he's like,

Now I can get back in there.

He's always posting his GitHub graphs of his productivity and it's like, yeah, how much
code he's shipping.

Because you can kind of hop in and hop out without needing to be all the way mentally back
into the actual code in every class and every function and what's going on.

And that takes a long time to kind of get into before you can even write code.

But the AI gets right back into that quick.

It knows where all the classes are and where all the functions are.

and just start building features.

Yeah, and that makes a ton of sense.

You know, one of the things that I've said previously in the podcast when your podcast
co-host Aaron Francis was on is that I think that AI is especially beloved by the, and I

don't attribute ADHD to you, but like the founder with a million things going on or with a
million ideas who does not have time to just sit in front of an IDE or a code editor for

eight hours a day.

you now have the ability to say, in between meetings, I'm going to go write out a prompt,
and then I'm going to go do the other responsibilities of my job.

And then you come back and you review the code in five minutes, and then you ship it
versus that would have taken like a half day of work in order to do that.

And so it's making breathing that the breathing room that we have as founders that is not
enough to write apps and to live in app code and live in that is enough to build spec, uh

send prompts, review code, review functionality, ship, and do that in little bits and
pieces.

um

Yep.

someone told you uh the end result of AI's growth means you're never gonna write a line of
code ever again, is that a disappointment to you or is that sort of like, yeah, whatever,

I don't care?

Okay, tell us more.

So you're not, do you ever open the IDE at all, the code editor?

I never opened an IDE.

have this whole mem, not never, but very rarely open an IDE.

I my whole mem where I like to pass around Luke Skywalker with the blast shields down when
he's on the Millennium Falcon.

That's me.

I don't want to see what's going on.

No, it's not 100 % true and it depends a little bit on exactly what we're talking about.

But like for my side project stuff, I've I haven't looked at the code at all.

um For Help Spot, which is our main product,

I do review it a little more closely, but it's still even then I'm not like making
decisions too often unless I see something it does really wacky.

um But even then I'm not usually, I'm almost never in the IDE.

I'll just be in GitHub desktop because it gives me like a nice diff.

Yeah, and I'll just like review what it did and make sure it seems like fairly reasonable.

um But like I already feel comfortable that the AI is a much better coder than I am uh
where the AI isn't yet.

Um, like we just had a thing the other day where like, uh, we're doing this whole project,
which we can talk about in a bit, but anyway, I made a PR that the AI generated to fix a

security issue and, we still review each other's code.

So I like submitted the PR and then like a human, our CTO reviewed it and found, uh, where
we should be, have been using this setting that help spot has, um, as part of this.

and I won't get too deep into that.

But this is where the AI is still not very good, is like the holistic view of the
application, right?

So it's like, it doesn't know, it didn't go look for if there was a setting that might
impact it.

It's just like, oh yes, I see the bug, I will fix the bug this way.

And it did fix the bug.

Right.

So it's very myopic in that way.

then like, and the way it did it, was fine.

And implementing the setting is in a sense,

wouldn't have been the end of the world to not do it this way, but it just makes more
sense in how the application works.

The setting exists, it should probably factor this setting in, and now it does.

So there's still, know, it's not perfect yet, but in terms of like the raw coding, yeah, I
mean, it's definitely better than I am.

So I'm fine with that.

That doesn't bother me.

there's one thing you kind of like already suggested here was you said kind of in my side
projects versus my day job.

And that's something that keeps coming up in the podcast is people saying, I'm going to
treat it differently based on, ya know, is this going to be used by five people or 5,000

people?

Is this going to be used internally versus something we're selling to people?

So there's clearly like a line we're all drawing.

um Do you...

If you discovered that somebody on your team was using AI for code in exactly the same way
that you do in your side projects for Help Spot, would that be a warning sign?

You're like, no, like we have to have different approaches to it.

I mean, I guess in exactly the same way, but I really think the majority of the problem is
not just for me, but in general is to getting software developers to use it enough.

I feel like there's still inclination to not use it as much as I think you should use it.

so, ah so I would be okay with that generally speaking, cause I would prefer people are
using it more.

Yeah, like, and we're overusing it and now we need to correct back as opposed to we're
underusing it.

uh Because there's just so much to do and it's just making more work in many different
ways and I've and like the only solution to the more work in my opinion is like the AI is

going to have to help us to achieve all the work that's coming out of it.

so yeah, so in that regard I would be okay and I'd rather they're on that leading that
direction with it.

Sure, and one of the things that I've, you know, like I try not to inject my opinion too
much, but one of the opinions I have failed to not inject is ah I don't think anybody

should ever ship AI generated code that they have not read and understood and can own.

And I'm just like, cause there's a lot of people out there saying, I don't even read it.

If there's a bug later, the AI is going to fix it anyway.

what, and I'm like, nah, that's, I do not accept that.

I understand and I do not accept it.

But you've already said like you're reading, you said I don't write code.

but you didn't say I don't read code.

You're reading everything, you're understanding it, you're catching mistakes, whatever.

So yeah, I think that's one of the reasons why I would feel comfortable similarly to you
if I'm just like, hey look, if you read every single line of code, you believe it's the

way it should have been written, you own it, you understand it, and you believe it's the
right thing, I still, personally because of the way my company runs, I don't think it's

gonna be the right thing for us.

But if someone were to tell me I can deliver the same quality output as I did before I
used AI, as I do afterwards, and I own it and I read it, I'm like, okay, that's what I

care about, Go ahead.

yeah, I mean, I think, uh, you know, I don't know.

Do I?

It's a tough one.

I don't know if I literally read every line, even in Help Spot.

um I'm definitely reading a lot of it.

uh I think that this is like a temporary state, the having to read it.

I think the reading it is going away pretty soon.

Yeah, I just think the reading it's going away pretty soon.

At least you reading it.

Now, I think there's going to be some of what we've been putting in place recently is like
other AI that's reading the

the code, right, and reviewing it and like multi-stage processes of that.

But you know, I think there's a little bit of a false premise to some degree of like, it's
not like there was no bugs when the humans were writing the code and now there's AI code

and there's bugs.

Like there was tons of bugs before.

yeah, um I just think that there's...

going to be bugs in the AI code, but there was bugs in the human code too.

And the human code has a lot of bugs that the AI tends to make less of, which is like, I
forgot this was used in this other spot type of issues where like the AI is really good at

being able to search through the whole code base and be like, yeah, it's used in these
four places.

This one hasn't been touched in 10 years.

You totally forgot about this, didn't you?

And like, yeah, I would have forgot about that, right?

But.

So I just think the AI is gonna be better at it than we are, like very soon.

I have this on the to-do list, but we're getting to it now.

You and I were both on a live stream with the University of Connecticut talking about
hiring.

We found a place where, I mean, because Ian and I have been friends for decades at this
point, is one of my favorite people in the entire internet.

Like, you know, like I love Ian Landsman, like you wouldn't believe, but that doesn't mean
we always agree.

And we were on this thing together.

And I think one of the statements that you made was something along the lines of it was
more important.

for developers to understand how to work with AI than it was for them to understand the
code.

And I don't wanna misrepresent you, so kind of correct me if I'm wrong there.

But it does kind of point to this right here, because what you just described, I'm like,
no, no, I hate that.

And the main place I'm kind of leaning in there is I'm thinking about, there's a couple
articles that have come out recently that I really liked about people saying,

we've gotten to a point where nobody fully understands the system, nobody understands the
interlocking pieces.

And so our ability to debug, our ability to understand if a new change is or is not the
right thing is decreasing as more and more of the code and the architecture and the pieces

as they work together, we're all decided on by an AI.

And so we get to the point where something goes wrong and nobody actually has ever
actually touched or even fully read that code.

And so nobody knows what to do.

So you really hope that the AI can debug it.

And if you can't, you're sort of like,

Well, I have to approach this as if it's a brand new code base I've never seen before.

And I'm like, man, if I were full-time responsible for an application and a bug happened
and I'd be like, I don't know, let me go dig in and learn, that's terrifying for me.

you, you, is it just you have the confidence that the AI is gonna get so much better where
that's not gonna be an issue?

Like they will always be able to find it and solve it for us.

I mean, yeah, think that, so this was, especially on that program, right, we're talking to
like people coming out of school.

so I think there is definitely an element of like skating to where the puck is going.

And so while yes, right today, you need to know the code well of the app that is very
useful.

You wanna be able to review the code that AI created.

ah I just think in like a couple of years, it's just not gonna be true.

just think like the AI can be so much better at coding than you that.

Like you would never debug it on your own anyway.

I I don't debug anything on my own now.

Like if there's a bug, I'm like, okay, AI, like here you go.

Like here's the scenario, here's what's going on.

Like go in there and take a first pass at it and then we'll see what's going on and if
that works, if it's solution seems reasonable.

Yeah.

often have you gotten to a point though where you do that and you're like, yeah, it just
doesn't get it.

it did, cause that's happened to me a bunch of times, right?

Like a bunch of times I'm like, nope, that didn't work.

Try it again.

Nope, that didn't work.

Try it again.

Cause you know, I've, it's for me, it only happens on the apps where the AI has generated
the whole thing, right?

Because if I, if I, I have like, if it's a app for work or something I've written in the
past,

Not once have I hit a point where I'm constantly cycling through these bug things, because
at some point I'm just like, I know the problem, I'm just gonna go fix the freaking

problem.

But if I have a couple apps where I'm trying to go blast shields down in parlance, so I'm
just sort of like, I don't really know what could be causing this because you implemented

some tool here that I don't actually fully understand, and I'll just get in these loops
and you'll see my commit history and it's just like.

try the new thing that the AI suggested that probably won't work.

And I'm just doing that for hours.

And obviously it's hours of one minute prompt, go let it run, go do something else.

But still I'm just like, I've never in my life just constantly thrown stuff at the wall
hoping something's gonna stick until this.

Does that happen to you?

that doesn't happen to me too much anymore.

Definitely used to more.

As the models have gotten better, I don't have that problem that often.

I do think, I do hit stuff more to me is like, it's not very good still at like user
interface.

That's where I ended up spending still most of my time, I would say.

And maybe it's just the type of apps I'm building, which I mean, most of the apps are like
cruddy, pretty standard type stuff.

But the UI, it's just bad at the UI.

Especially when you get to subtle, really how it works and how it feels type of stuff,
it's still just not very good at that.

And some of it, yes, if you're in a framework, that will help.

But it's still, when it puts all the parts together, sometimes it's just not doing a very
good job of that.

there's that.

I mean, I think what it means to be a software developer is probably gonna change.

And it's hard for us to dissect that, right?

Because right now we do know.

I think you definitely know way better than me, but even I know the code and what's
supposed to be going on.

So what level is that gonna be important at?

I think probably for quite a while it'll be important to have an understanding of
programming languages and if you're in the web, the HTTP stack and what happens in a web

request or whatever.

I don't know.

I don't know if that means you really need to have

fully memorized and internalized the languages and frameworks the way we had to, because I
just think the AI is gonna be really good at that part.

And so then it's more about shaping it and helping it when it doesn't totally get the
bigger context it's working in of the application and its purpose.

Or yeah, it's stuck on fixing a bug because it's taking a like kind of big picture
approach and it's more like you gotta shift it.

to a different direction.

But I don't know, a lot of those things I find useful to use multiple AIs at once.

That can break through those things a lot.

Where like, okay, I'll send four AI agents from different models.

So you have Gemini and Claude and Codex all go look at this problem and they will then
hash it out amongst each other.

yeah.

And I told you before we recorded it, I'm like, one of the things is, of the non-Hype
boys, you use more crazy amounts of things interacting than anybody else I know.

So talk to me, you have more than one subscription, probably pretty high level
subscriptions to different AI tools, and you are having them all talk to each other?

Like, what's that process look like for somebody who doesn't understand how that works?

everybody should do this.

I think this is like, especially if you're a programmer, but even for non-programmers, I
think it's pretty useful.

And you actually don't need the highest level subscription necessarily in each of them
because you're going to them, you're going to the whole panel of AIs when you have a

harder problem.

And so it's not actually that, it's not like all day every day.

And so yeah, if you have like a hard problem or something you wanna research, yeah, I have
like.

Claude is my main one I use.

That's like my main AI I'm in all day is Claude.

But then I have a Codex account, know, ChatGPT, and Gemini, which is Google's.

And then I used to have AMP, but I don't use that one as much, but I still have it.

And so yeah, so like the thing is they're all trained slightly differently.

They all have different instruction sets when they're built as to what they're emphasizing
and things like that.

And so, I mean, it's just...

It's very like humans.

If I sent the bug to four humans and was like, hey, everybody look at this and give me
your thoughts, like they're gonna have, take different angles, they're gonna search

through the code base differently.

And so it's the same thing.

that by sending a bunch of them out, you just have more opportunity for one of them to
like do the search that finds the thing that was actually the issue or for one of them to

understand architecture a little better than the other ones.

and then they'll make their case to each other.

So you have like one AI that like is the orchestrator and then, you know, and Aaron
Francis kind of developed some tools around this that I use.

And yeah, it's great because they'll just go out, you send them all out, you say the
orchestrator spin up one of each of these, give them the problem, come back with the

results and summarize everything for me.

And then it'll do that and it'll summarize it.

And most of the time they'll agree, but then sometimes they don't agree, which is

great, so you'll have the detractor be like, actually be the one who finds the issue.

So yeah, so just having your stuff, it's a little bit of work to get that set up, and that
was still a little bit of expense, but the expense is nothing.

Compared to five years ago even, compared to hiring four people, it's not expensive.

And this is where we can get into jobs and stuff like that, but it's not taking away those
four people's jobs, because I would never have hired those four people.

We are a small company.

Yeah.

people, so I have four extra people to chat with about when we have a bug once in a while,
right?

So that's not a thing that's being removed.

It's just now, it's like I can have four extra people that I cannot otherwise afford, but
now I can afford them because they're 20 bucks a month for three of them and $200 a month

for the one of them.

So great.

So I have that, the suite of tools, and they go out and they do stuff.

I'm not all the way to like full automation of my whole life.

Obviously there's people out there.

It would be interesting for you to have on who are fully on that.

I'm not all way to that.

But when it comes to this stuff, I think it's super useful.

I think, cause yeah, I don't know.

just, I don't know everything, you know?

And like the AI knows a lot more than me in some ways.

And so putting a bunch of them on it, obviously you can go even farther.

I don't usually tend to have to do this, but you can give them different instructions.

So like you're really.

Optimistic about this you're really pessimistic about this like you're you know and like
let them battle it out and come up with different ideas for it and So yeah, I mean

obviously the AIs in the current state tend to like yes, man you a bit so like uh That you
have to watch out for which again when you have a panel of them I feel often is better

because you don't get quite as much of that um But yeah, I think people just go the tools
are out there they're useful

no matter how much you use them in terms of what level, even if you're coding, even if you
wanna still do the code, I still think this process would be quite useful because even if

you leave it at the level of like give me feedback on this idea, and then it's gonna go
through and be like, I see this edge case and this other thing, and then you can factor

that into how you actually implement the features and code them up and design them and all
those things, so.

Well, I know that I told you ahead of time, I want to make sure that we're making space
for people who aren't programmers, but I don't want the programmers to miss out on kind of

some of the nerd talk.

So apologies, non-programmers.

We're going to go a little bit deeper here.

uh If somebody were to hear what you're talking about, and I imagine you might even be
able to use this tooling for not code, but if someone was like, I need that setup, what is

the thing you just described?

I assume it's Aaron's counselor's thing, right?

So how does it work?

so there's a couple different ones.

would say, yeah, Counselors is like a command line tool.

So for a non-programmer, it's gonna be a little bit harder, but it's possible.

Aaron also has another tool called Solo, which would be a little bit easier for
non-programmer, which is like a desktop application.

It kind of runs your terminals for you.

um And from there, Solo on its own now um can orchestrate

across different AI command line tools.

So yeah.

Yeah, I do, right?

Yeah, it's coming out soon for everybody else.

So yeah, you'll be able to do that with Solo.

to be able to have multiple things running that can all talk to each other, right?

Yeah.

can, just to say geeky for one second, it's controlling all your terminals and it's
controlling all the command line clients of the agents, it knows their agents.

And so if you've configured correctly Gemini and Claude and Codex, then it can just, it
creates an internal MCP server so that one agent can talk to another agent internally

across your terminal.

And so you can say, hey Claude, spin up five Codexes and do XYZ and it'll do that.

That's the geeky way, but I think even for the non-geeks, um this is, because a of these
tools have this internally and it's still almost as useful.

So that's like the extreme of like you have actual different companies, right?

Like we have ChatGPT, OpenAI, have Claude, Anthropic, whatever.

But even if you are like a desktop AI user of like Claude Cowork and the Claude desktop
app, like you can tell it to spin up sub-agents, I'm pretty sure.

And...

So then those would just be additional Claudes They wouldn't be going out to OpenAI, but
you still get like 90 % of the benefit when you do that.

So you can say, okay, spin up five subagents to all look at the same problem in slightly
different ways, and then aggregate the results of what they returned to you into a summary

and tell me what they found.

And you're still gonna get a lot of the benefit of by sending five different ones out,
they're gonna search differently, they're gonna come up with different little angles on

it, and.

then you're gonna see those results.

So yeah, if you have something hard, and I totally don't think it's just code at all.

I think it's for all kinds of stuff.

uh To me, is like the biggest, kind of bigger than the code to me is the idea of writer's
block.

This impacted me all the time, in coding.

Like I've had tons of side projects, like I never got anywhere near finishing, because
it's just like, get started, and then you hit some bug or some weird issue.

And you just, that's it.

Like then once you stop, then like a week goes by, then a month goes by, then like, you
know, you're done.

And that could be in coding, but it could be in like literal writing.

It could be in all kinds of things where you're just like thought processes is just
stopped for whatever reason.

And so I think by being able to just send those things out to the AI and be like, Hey, go
think about this.

And then you come back 10 minutes later, an hour later, day later, whatever it is.

And it has a bunch of ideas for you.

You're like, oh yeah, that's it.

And now you're through the writer's block and now you're going again.

um So to me, that's like the biggest thing.

Where like in the past, like from a code perspective again, like I'd be coding something
and then there's a bug and I can't figure it out and it's something stupid, but I don't

know what it is.

And it's very disheartening.

And you're just like, oh what's the point?

And now it's just like, nope, the AI just fixes that.

Even if it doesn't, this is like where it doesn't fix it optimally.

Who cares?

Yeah.

the problem is I wasn't gonna do it at all and now it's getting done.

you know what, someday when there's a bug or an issue that comes out of it, we will fix it
in the future.

ah Which again, like people kind of get caught up in this idea of like, well, I don't like
that mindset.

But to me, that's like how it's always worked.

I'm still fixing things I did 20 years ago.

Like, so.

been a thing since the very beginning.

So what's the difference?

Like there's no difference.

Like, okay, it's not optimal.

If anything, the difference now is like, if you tell the AI to go in there and fix a bunch
of stuff, like it's good at refactoring, large refactors in a way that humans again are

not the best at.

So telling it to go in there and hey, let's rebuild this thing and move a bunch of files
around.

Like again, another point where like in the past I get into something and like, I should
have architected this differently.

And now it's like I have to restart it from the beginning or I have to like try to do it
in place, just a big nightmare.

But yeah, it's just like AI, go in there and move everything around.

And it'll go in there and it'll move everything around.

It'll take two hours and it'll do it.

And it'll come out and everything's moved around.

And you're like, man, that was amazing.

And not always perfect, perfect.

But it's now much better than it was, right?

And just amazing tool that if you had to do that yourself, it would have taken you two
weeks and you still screwed up the way the AI's gonna screw it up too.

So again, I don't.

The difference to me is negligible in those ways.

love this writer's block idea because I uh referenced Justin on the last three episodes of
this podcast, so clearly I need to get him on.

But Justin just shared a video with Justin Jackson, shared a video with me yesterday of
him working on a page on his site, and he talked to the Ahrefs, which for those who don't

know, it's like a search engine optimization tool that tells you, here's what you're
ranking for.

And he's like, hey, there's this thing I want to rank for well.

ah

how well am I doing and what pages are ranking?

And it was like, here's the pages.

And he's like, I expected this page to be ranking really well.

And Ahrefs was like, no, definitely not.

It's terrible because the content is not good.

And so he's getting the combined knowledge of Ahrefs through its MCP server.

And then the reasoning of Claude through it's just general being an LLM.

It's like what you really want to rank is for it to be like this.

And I don't remember exactly what his process was like, but basically he said, well,

just throw together something that gives me that basic concept and then spent what seemed
to me, I only saw a quick video, but like probably hours of work taking it from that

initial thing to like what would actually be Justin Jackson quality content.

And some Claude help, a lot of work on his own.

But as someone who's had to write content, create YouTube videos, create podcasts, saying
here's a page and saying, it's good in this way, it's bad in that way, here's what I want,

blah, blah, versus you should make a, here's a blank screen, like do something with it.

Yep.

It's a transformational difference, both from a code perspective, from a content
perspective, from plans of something you're going to do perspective.

Giving me six hotels that the family can visit in Punta Cana is very different than
saying, go find Punta Cana hotels on your own on Google, try to figure out.

It's just like, so yeah, I love the um writer's block and whatever the equivalent writer's
block would be outside the context of writing.

It gives you something to respond to versus a blank slate.

Yeah, when we were building new website for Help Spot, it's like for 20 years, I've never
had the versus pages or switching pages of like switch from Zendesk to Help Spot or

whatever, all these things.

And it's like, such a heavy lift.

You have to research all the details and then build the pages and all this stuff.

And it's just like never happened.

But it's like, okay, I just use the AI to do like deep research on each one, sent it out
there and had it do like nice summaries of everything.

And then...

Check that, but just checking what it found is like one 100th the work of actually going
out and digging around for every answer to these things, right?

So check it, found some errors, fixed them, and then had it build the pages.

And now we have like 15 of these pages when we had none.

And it's like, if I was rebuilding, I've rebuilt the website over the years, I've never
done these pages, because it's like, by the time you get through rebuilding the actual

content of the websites, well, I don't have any time and bandwidth, I've already worked on
this for two months, like I'm not gonna go through and like.

launch the thing.

Yeah.

And then I get busy and I never come back to it, right?

So like all that kind of stuff is just so much more helpful.

um just, yeah, this is where it just makes, it makes more work in a sense, right?

Because it's like, now I can do more, so I am doing more.

um I could just do the same amount.

Like I could have spent less time on the website by having it do the same amount of
website we did before.

But instead I did more websites.

So it still took.

Yeah.

Yeah.

didn't take the time savings.

I put it back in to make a bigger website.

Right, so yeah, there's just more possible.

lines up with so many things.

was like, there's so many directions.

But one of the things that's making me think about is we've talked at Tighten often about
the fact that people are shipping so much faster than they were before that people are

getting to the point where they don't always have the ability to actually stop and ask, is
this what I should be building?

Right.

It's the, the Malcolm, you know, in Malcolm question, what is his name?

The guy from Jurassic park.

i wouldn't stop and ask if we we could we do stop as we should uh...

tons of that out there

apps and features and content.

It sort of like, yeah, you did.

You you shipped that in the same amount of time.

It would have made you to ship less.

But like, is it actually better?

Have you found that you have to adapt your working processes or your feature definition
for Help Spot or whatever else now that the pace of shipping is different?

Yeah, I think we're still working through that stuff.

It's still early in a sense, but there's like, yeah, we're still kind of working through
that.

We definitely need to ship faster and more, right?

Customers have ever increasing needs.

like software's never done.

we didn't early, the first couple of years of this stuff, we didn't get hardly any
requests for AI stuff in Help Spot.

And we've been chipping away at adding little things, but we haven't gone all the way in.

But now we're definitely getting more feature requests for MCP server, which will be
coming soon, and AI auto responding.

And it's still a crazy world because half the demos I do for potential customers are like,
we don't want AI, no AI, can you turn it off?

And they're like, yes, we can turn it off, don't worry about it.

And then the other half are like, what do you mean it doesn't already auto answer
everything for me?

So it's a little bit of a crazy space right now.

uh

I don't know, it's gonna be interesting to see the winners and losers and things like that
because I don't feel like, don't use, do I use any just AI built application, like that

are new applications that have come up in this AI era.

I don't think so.

So it's letting existing good ideas iterate faster.

Like the apps I already used are getting better, right?

Internal tooling, like kind of what you're saying with Justin did, like recently I...

built a tool just for myself to interact with Google AdWords, which is something I've
dabbled in over time, but the AdWords interface is so bad.

And I just feel like I'm getting screwed, and we're in a very competitive space where
every click is a minimum of $10 to $15, and a lot of them can be over $100.

And so it's not just like I can put five grand in there and like, oh, I get 1,000 clicks.

It's like, no, I can get five grand with 50 clicks.

So that's not great.

So it's hard to experiment.

So AI has just revolutionized it for me and it's so much better to work with AdWords
through this basically professional oh AdWords, an AdWord professional essentially, right?

Which I couldn't afford to pay again one of those because they're not gonna wanna work
with my $5,000 a month budget because they get a percentage of that and they're like,

well, I'm not gonna work for you for 500 bucks, right?

But now I can have one for my budget because I can just do it with Claude and...

We fixed all kinds of stuff and we'll see how much it actually works, but I feel like it's
already vastly superior set up, at least as a potential to be successful, versus my own

just like cobbling around in there.

I have no idea what I'm doing.

And so.

being able to pay attention to it for 30 minutes every month because you got too many
other things going on.

Yeah.

and I just use the Claude desktop app, but I have a scheduled task, and every morning it's
running a report, so I'm actually looking at it, which is another thing, in the past I'm

like, okay, well, I'll throw five grand in there and see what happens.

And then I come back and the money's gone, yeah, nothing's happened, the money's gone, but
I don't see any other differences.

So it's like, can stay on it every day.

um Yeah, so just stuff like that, really great.

But I have a question for you, because this kind of came up the other day.

You actually brought up earlier on this podcast too, but a lot of what Tighten does, I
believe, is you are embedded with other teams.

So with your customers, you have employees that are just in there all day every day.

They're really working with, I presume they're actually interacting with other developers
in your customers.

so what have you seen from them wanting to use AI tools?

with this philosophy you currently have, like, well, we need to have human review
everything.

Like, when they are like, that is slowing us down, We have to, like, automate more of
this.

I'm curious if you've seen that and what your plan is when I think inevitably you see some
of that

Yeah, so there's a couple answers there.

The simplest answer is we have never in the history of Tighten had a client who was
unhappy with our pace unless it was too fast.

So we have a consistent problem where we have spent months to years building up the
relationship, thinking about the timing, warning them about how much review they're gonna

need and how much planning they're gonna need.

And we get in there and we're four weeks in and they're like, we're out of tasks.

Alright.

can move so fast, right?

And so that's what in my business development meetings I tell people I'm like, hey, we're
not the cheapest in the world and I'm never gonna want to be the cheapest, know race to

the cheapest gives you know, but The amount of value you're gonna get for your dollar
spend like you're not prepared for how much we can deliver So AI has not changed that like

we are still able to deliver more functional working software and it's not like our devs
don't use AI they use AI but they don't use AI like I think a lot of AI maximalists do

because

At the current moment, least Tighten developer knows all the software Tighten developers
working in significantly more than AI.

Like we're not talking about you and me where we kind of like we're coders who are also
entrepreneurs who are also doing business calls who are also managing people.

We're talking full time in the IDE experts in their craft, their usage of AI, the vast
majority of the time, almost always what AI outputs is going to be worse than what a

Tighten developer outputs in the same amount of time.

Yeah.

then it's sort of like, well, what can I do that allows me to free up my brain to think
about things?

So they're using their brain for architecture, but maybe the AI builds out the template or
something like that.

Like I don't need to be paying somebody Tighten developer hours, know, dollars per hour to
go build out, you know, freaking flex box or whatever, right?

So like they're still finding ways.

So anyway, the simplest answer is we're not even close to a point where anybody would say
I want you to be faster.

However, we have plenty of clients who are using AI for their developers and

The vast majority of time what's happening is the output of the code into their code bases
is diminishing.

The quality of the output of the code in their code base is diminishing.

Their comprehension of what's happening in their code bases is diminishing.

The number of times production has gone down inexplicably has gone up.

That's pretty universal across all of the clients who are really heavily relying on it.

There's one client we're working with who

knows the pros and cons of AI.

So this guy will go to a client and he knows his client wants a new feature.

So he will spike out this feature in Claude or Devin in a week.

He'll sell it to the client.

And then he's like, all right, guys, we have two months to build this feature between now
and when the contract actually starts.

And he'll hand it off to us and he'll say, I know this is garbage.

Make this scalable, secure, or whatever.

And then we'll run with it.

So that's the dream, right?

Like find the things that AI is specifically good at.

and he's a CTO, right?

So he's not supposed to be in the code day to day.

He doesn't want to care about arc.

And so he's like you, right?

I don't want to sit and write code all day long.

So when those folks get access to AI, it's really fun.

We have another client who dreamed about this project for over a year.

He and I are meeting all the time, but it's just like, it's a six figures budget and he
just didn't have it.

He's completely bootstrapped.

But then, we define an MVP.

He builds the MVP in lovable.

ah You know spends under a thousand dollars in credits in Loveable define it hands that
MVP off to us and says hey Can you build that and I say because a you've already built us

this feature definition and B we're gonna lean really heavily into AI Let's see what we
can do and significantly less money, right?

So we build him this this thing and it launched in and it was you know I'll just say like
well, no, I can't say the dollar amounts It was like a fifth the cost it would have been

if he had not done that definition up front if we hadn't leaned really heavily

So there are clients using AI in really creative and innovative ways that tend to be
non-technical people using AI in a way that they don't expect that code is going to

production without our review.

And that's going great.

But when clients, developers, usually the more junior developers are leaning more heavily
on the AI, basically when there's no senior level person saying, I need to ensure the code

quality of what was AI generated here, um that's...

causing all sorts of problems.

Features shipping that shouldn't be shipped, code being delivered that doesn't actually do
what they think it is, more more spaghettification, more production going down, more

untested code being delivered.

uh So we're trying to build harnesses and frameworks to enable them to do, you know,
increase their speed of output using AI without all those downsides.

um But a large portion of people making those changes are trying to do it so that they can
save money, so they can move faster, so they're not

kind of slowing down and asking the questions of how to do this responsibly.

So that's not true for all of them.

There are some clients who are using AI and they're they're doing it responsibly and
they're seeing kind of a slow increase in their still quality output.

But nine times out of 10, it's not for the long-term benefit of the company.

And I'm worried about a reckoning coming for a lot of people in the near future.

Interesting.

Okay.

with these, we're talking about code bases that have been around for a decade.

We're not talking about a brand new app with an entrepreneur.

We're talking about more junior level developers adopting AI and now more confidently
shipping code that this shouldn't be confident because there was nobody looking at this

code to say, does that have an implication on our code?

And then when something breaks, nobody's ever seen the freaking code that's breaking
things.

So.

right, yeah.

Interesting.

Now what do you think about like, cause one of the things I think is gonna be happening in
the next year or so is like, just gonna be, the reckoning is gonna be in like security

because it's just so easy to have the AI spin up like and just attack everything.

And it's gonna be really good at that kind of stuff.

And it already is pretty good at it.

So like that's like a specialty.

that your team doesn't necessarily have.

Do you know what mean?

Like they know the best practices for building a secure Laravel application in terms of a
traditional sense of that kind of thing, but that's not really quite the same thing as

like, let me think about this.

Right, exactly.

Like how do all these three things work together that actually allow me to get in here and
execute some code that I shouldn't?

Like, um.

So I think for those kinds of specialties, like this is an area we're using it a lot more
of in terms of like just get in there and try to make stuff break and try to hack the

system.

so we can fix those obviously like urgent issues we fix right away, but even the like,
yeah, this isn't an issue on the surface, but if they get this thing to happen and that

thing to happen, then this actually is an issue.

um

So I don't know if you've incorporated any of that stuff in yet, because that's like lot
of what we've been, that's been keeping me busy the last few weeks is we've gone like

really heavy into that and it's, it finds stuff, it finds stuff.

eh

for me is that I'm finding that um there are, if we blanket, say, usage of AI is
benefiting us in all these ways, period.

We can trust that it's smarter than us, period.

We're going to end up shipping code that's not as good.

But what we can find is that prompting or choosing to use tools at specific angles allows
it to shine.

So, for example, if I were to say,

Hey AI, make sure that this feature you're writing is secure.

Or even if I were to say, hey AI, please audit this application for security issues.

I'm only gonna get so much value out of it.

But if like you said, you say, hey, here's a production application that is set up with a
particular configuration of NGINX and Laravel and whatever, I want you to try and hack it.

I want you to try and find holes in it or use this particular set of guidelines, OWSP or
whatever it is and try and...

kind of get in there, you get a lot more beneficial output than if you were to just look
at a code base and say, find bugs in this code base.

So I think that one of the things I've been saying very, very often since the very
beginning is AI is a tool.

And if we treat it like a tool, you have to understand where to use a tool and what a
tool's good at and what a tool's not good at.

You don't use a hammer for everything.

You use a hammer for, you know, like hard attack thing.

You use a screwdriver for a different thing.

And so AI is a complicated tool in that

based on which AI you're using, in which context, and with which prompting, it becomes
different tools, right?

But the thing is, it's still good at some things and not good at other things.

And if you just say, hey, you know what, just do it.

It's not gonna just do it.

You have to know how to prompt.

You have to know what it's good at, what it's not good at, and you have to use it
creatively there.

to answer your question, we don't have all the answers of like, what should you do on
every single application?

But we're starting to think through, hey, like you just said, like, hey,

Security has always been a concern and one of the ways that security has, we've all
benefited is the smaller the app, the less likely that a determined person with technical

knowledge is going to choose to sit down and throw their energy at this app for Bob's
magnets, right?

Like nobody cares about Bob's magnets.

So we can build an app with the best security practices.

And we've often told people who are like, we have real security issues.

Like every app is hackable at some level.

There's you know, like if you freaking Pentagon gets hacked, right?

Like so everything so it's not about making unhackable applications It's making
applications that are reasonably unhackable with the level of kind of like security

issues.

You're likely to have Right and to your point it has become significantly cheaper to throw
senior level hacking at an application because the senior level hackers could get access

to

Mythos or whatever, know, the latest, greatest, you know, whatever.

So yes, I think that there's a baseline concern where we have to harden a generic
application in a way we didn't in the past, because in the past, the worst thing they were

gonna get is some automated bot is gonna try and hit WordPress, unsecured WordPress
endpoints.

Who cares?

And now it's a bigger concern.

No, I don't have the answers.

Like the long answers, no, I don't have the answers.

um I do think that...

um

It is if I were to be an owner of an application, I would be more concerned with spending
time and energy at the end, making sure before we're officially launched to the Internet

that we are throwing AI at things and saying, just try and break it, just try and hack it,
just try whatever.

So I'd love to that you're doing that because you got to leave our way here.

Yeah, no, we've gone like really all in on it because it's just, yeah, they're just gonna
be doing the same thing, right?

So it's like, they're gonna be doing it against us.

And I feel like you wanna be ahead of that.

Because the good news is we have access to the same tool they have access to.

And, you know, for most of us, we're not doing stuff that's necessarily that financially
rewarding for them to attack, like you're mentioning, but.

when the cost is very low to attack and if they can just even get access to the server
which they can use for other nefarious things against more profitable targets, right?

So like there's all the different levels there of like, yes, it's worth hacking you
sometimes just to get access to your server to then do other bad stuff with that server,

not necessarily to get your stuff and sometimes to get your stuff, right?

So yeah, there's this interesting talk.

I don't know, I might have posted it to you in one of the chats we're in, but uh we could
put it in the show notes, but it's by an anthropic security researcher and how they go

about it.

that's like, we built an internal tool that's kind of based on the same concept of like,
go through all the files in the app, know, tell the AI it's in like a security game where

it has to find security issues and say, start at this file.

And we just do that over and over, but for every file in the application.

And...

It goes through there and it finds all kinds of stuff.

And then now then then there's like layers of this where it's still AI, right?

It's like, well, it gets a little myopic and it's like, well, this is an issue, but it's
really that issue has already been mitigated higher up the chain.

And so it didn't actually look high enough up the chain to determine that.

So there's like another pass that's like actually verifying.

And so that at least when we get down to what the humans are going to now spend human time
on, because it's a security thing and we want to like

have the humans give a real check in the context of the whole application.

So we're not all the way to all AI everything yet.

um But the human time is precious and we don't want to spend it ideally on a bunch of like
false positives and things.

So this application gets more complicated as we build it out.

Kind of interestingly, Codex and OpenAI has the Codex programming tool which just launched
a security feature.

Okay.

works exactly like the tool we built in-house, which is kind interesting, because it's
just an obvious approach, I would say, to this.

ah So we ran our stuff through that, which is kind interesting.

It found some different things.

um So yeah, I would say use all the tools that are out there, ah because everybody else is
going to be using them against you.

you have to do it.

And then obviously we'll be back to the game of fishing and humans being the weak link.

Yes.

is often the case.

em But that's a bit of a separate issue and there's some stuff you can do there to help
people too, but that also at least tends to be much more isolated as opposed to like,

well, I can just run a script and hack every instance of your application is much
different than, yes, I can trick somebody into one account into giving me their stuff,

which is.

which people are doing and they are using um AI to be better at it.

Like I literally just read a thing this morning where someone was like, hey, I went
through multiple steps of this job application process and I thought I was on calls with

people and they had all these indicators of being the real people and it turns out they
were using AI to generate every aspect of the entire thing.

So even just like our trust of what's real on the internet, like the signifiers of that
are changing.

I...

emails have gotten so much better.

Like, they used to be terrible, and now they're like, this is an app I actually use.

And it says, my billing payment thing is no good anymore.

Click here, and it looks totally legit.

And you're like, it's not legit, though, man.

It's not legit.

Yeah, the scammers are, they're getting good.

It's quite annoying.

Yeah.

So we are 52 minutes in and we haven't talked to Outro.

We haven't even named Outro yet.

So we've got to give a minute on Outro.

So you are building um a software as a service tool for podcasters.

I am so excited to use it because I have, I think, three podcasts running right now and
two that are on hiatus but may come back one day.

Maybe Outro will enable me to do it because I'll have time for it.

um But you're on, I think, the...

Third rebuild of this, and so this is one of the really interesting things to me is
watching you build this thing in public.

And ah I'm very curious, if you kind of step back from where outro is on about to launch
or whatever, and kind of look at it from like a high level perspective, you have rebuilt

the same application three different times.

Has the change...

how has AI helped and hurt your ability to actually launch this thing?

Because I know you rebuilt it and Aaron kind of gives you crap in the podcast about, no,
you're rebuilding it from scratch again.

Why don't you just launch the thing?

But each time you're sort of like talking about why you're rebuilding it and some of it is
code related, some of it is user perspective.

So can you talk us through like a little bit of like building a SaaS?

using almost exclusively AI to build a thing, but this is not just a fun little internal
tool.

This is a production level thing that you're going to have people paying money to use.

So you really want a certain level of quality.

What has the story been of like not shipping at the timeline you have, but are you happier
with the output?

know, if you reflect, like, has it gone well?

Were the rebuilds the right decision?

Has AI helped and hurt you through this?

Like kind of just talk us through like what that process has been like.

Well, we'll find out if it's the right decision.

I feel like it is.

Yeah, so here's the thing.

Definitely in the past, I would have just never finished this.

I wouldn't have finished it at all, much less we built it three times.

Yeah, I would have started it, and then I wouldn't have finished it, because I would have
been like, oh, I took the wrong path here, and then my choice is gonna be in practice to

actually ship this.

I just have to keep going down the wrong path, and kind of hackily.

make it the not totally wrong path, but like it's gonna be kinda not ideal, cause if I
start from scratch again, it's like, oh that's like terrible.

Like it sucks to start from scratch as a human just typing every character.

So, in practice maybe I would've got the shipping, but it would've been like, actually I
tore a bunch of stuff out and I added a bunch of stuff in, and it's all hacky and crappy,

it's all messy, I'm super sick of it, and.

that would have been, I think, bad.

And so this way it's like, no, I'm just gonna restart it.

And we restart it and I give the AI some initial big prompts, because now I know even more
what I'm looking for and we make a lot of progress pretty quickly.

And then it inevitably kind of bogs down, as I said before, and kind of like the actual UI
and what it actually does and the AI's not so good at building that part of it out.

So that's where it takes more time.

And I tend to work...

I've changed this a little bit, but mostly I tend to work in pretty tight commands to the
AI, so it's a pretty small scope of what I have it doing.

I have, as I've gained more tools, found that can let it be a little broader now than I
used to let it be and bite off bigger chunks than it used to.

But yeah, the rebuilding it, think, I mean, it's not my main product.

I already have a product out there.

If I had no product out there and I was like, this is my shot and I quit my job and I
gotta get to market, then I might have still made a different decision because it's like,

well, even in two or three weeks to reset is like not, that's like, I don't know if I can
afford that and like, we're just gonna push ahead.

But that wasn't my case for me.

So it's like, let's just rebuild it.

have a little bit, the vision of the product itself has changed.

It wasn't really so much like the code.

It was just like the vision of what I want the product to be changed.

And so it was like, well.

kind of like bike shedding constantly changing your technical architecture because you're
like, no, as I built I realized it should be a different product.

Yeah, yeah.

And to me, it's like the technical architecture for this app I was never too worried
about, because it's it's not a hot, it's never gonna be, even if it's wildly successful,

it's not gonna be like, well, it's gotta handle 100 million requests a second.

It's like just not that kind of thing.

So nobody's finances are in it.

So I'm not too much worried about that as long as the code is reasonable, it's fine.

So it's more just about what the product actually does.

And so yeah, it's been.

Part of it too is this was a whole experiment for me to learn AI programming better on.

so the restarting was an advantage, because it's like, well, now I know how to work with
AI better than I did two months ago.

And so it becomes a reboot, where now I'm doing things differently and I think better.

um So yeah, the number of hours, it's so tricky, this last reboot, which I started, it
wasn't a full reboot, but it was a mostly reboot.

was like in January, I think, something like that, December, January.

And like the number of hours I actually have into it at this point are probably like not
that many, because I've probably only worked a couple of weeks on it total since January,

because it's just been super busy with help spot stuff and other things.

But it's like almost ready.

Hopefully it'll be out soon.

it's, yeah, I mean, it's a tool to manage your podcast.

mean, you have been incredibly, you don't even know, there's like one of these things
where like,

the person you ever crush on doesn't know.

This is our relationship with Outro because you were kind enough to take me behind the
curtain, which this was part of one of my re-dos was after I talked to you and, what's

your producer's name?

Kayla.

And you guys took me through that, your process for managing the podcast.

And I was like, yeah, I had.

a good chunk of these ideas, but then to see them with somebody who's really very serious
about it and like really planning the podcast to very meticulous levels.

I was like, oh, okay.

Like, yeah, we gotta have like to-dos.

We gotta have like the organization of the to-dos.

ah Yeah, just a bunch of different things there.

We gotta handle multiple podcasts better because like the first version of this, like
you'd really have to like kind of.

Switch like let's say let's call it log in and log out It wasn't exactly that but like you
had to like really switch context to manage multiple podcasts I was like that's not gonna

work like I want to be able to see all my different podcasts together in one spot So like
now you can do that and like so a bunch of different things But yeah, your guys

organizational aspects really influenced the like organizational aspects of outro So when
you get in there, you're gonna be like, yeah, like this is kind of like how we do it

because it's based on what you do uh So that was really

helpful and fascinating.

But yeah, like what's it take to run multiple podcasts and how do you organize that and
how do you not miss things and how do you make sure the tweets go out and how do you make

sure the cover art is right and how do you make sure the guest, know, guest management is
something I haven't built yet but is a big part of this app too.

Cause like what you guys do with that is, you know, inviting the guests and scheduling the
times and asking questions to get how do they want to be introduced and what are the links

for the show notes and all those things.

I chose a tool to manage all of that stuff.

so, yeah, seeing how you guys do that is super useful and yeah, it's gonna be.

you know, so when I started this podcast, one of the things I told Kayla was like, I've
been paying VAs to help me with my podcast since my friend Jason Langstorf first

introduced me to the concept years ago.

I was like, Jason, I've got one podcast about to start a second.

I'm exhausted.

And he's like, VAs, you got to use VAs for it.

And I was like, brilliant.

He even introduced me to his VA.

So very grateful to Jason.

And since then have built out a system that I'm pretty happy with, but still.

Every new podcast episode is a pretty crazy amount of work on Kayla's plate and my plate
and there's a good amount of it that's really repetitive.

And so one of the things I told her with this podcast, I was like, hey, it's an AI
podcast.

Let's see if we can kind of get some AI help on not the creative human things, but it just
on like the really repetitive things.

And it's been helpful, but it definitely tells me, we need to build harnesses and systems
and structures.

And so I've spent the last year being like, nah, we'll just wait for outro.

We'll just wait for outro.

uh So no pressure.

uh

But yeah, like the AI is a huge part of it.

Like the gopher, who's the AI assistant in outro, like he knows your podcast, right?

And he has access to all your episodes and ah he can make anything you need.

So he can make to-dos and he can make segments and he can move things around.

like, so that's gonna be like phase one.

I think some of the repetitive stuff I wanted to handle longer term is even like.

The things like posting to the social media for you or all that stuff, right?

Like that would be nice to just have in there and be like, yeah, like we upload the show
audio and it does this already, but like it makes the show notes and it makes the tweets

and it makes the Instagram and TikTok the text of those things.

right phase one, it's not going to actually post them.

It could of course actually post them.

It could take video and make video.

It could do all that stuff, which is not going to be.

phase one stuff, is the framework is there for it to potentially do all that stuff.

so, you know, this is the problem though, when you actually know what's involved with
these things, that it keeps you from like, take a pan, like, launch it, like, I know once

I launch it, people start using it, right?

Then you have the crush of, oh, it doesn't do this, I need to do that.

Right.

Right, all that stuff you get into and it's like, well, if I'm not launched yet, I don't
have any of that stuff to deal with.

So.

for sure.

wanna get it as far as I can.

Because then once you just have the customers asking you for stuff, it just inevitably
eats up all the time.

yeah, but at some point it does have to launch.

Aaron's on me every day.

So we gotta get it launched.

But I was hoping to have it launched today, but it's probably not gonna, not launched
publicly today, but me and Aaron are gonna start using it today.

Yeah, but I don't know, probably not gonna happen.

But soon.

This next week me and Aaron should be using it.

That'll be phase.

That's phase one of using on my own podcast is phase one and then we'll get get some more
people in there pretty soon

Okay, and I mean, so one of the things you said here straight up was you're just like,
would not have launched without AI.

I would have started it and I wouldn't have finished it.

And I definitely hear a lot of people saying like, look, I've got this long list of ideas,
especially entrepreneurs and creatives, got this long list of ideas that never go anywhere

because the the cost to benefit ratio is just not good enough.

You have built some side projects over the years and some of them are still running like
Larajobs It's just wildly successful.

And some of them, you I think it was snappy.

You made another one.

You get at some point, you're like, you know what?

This kind of isn't worth the investment to keep it running.

Is there a world where AI could get you producing things fast enough where you just
basically as a creative, like build out a portfolio of little sites?

Because I remember one of the things Aaron talked about in his podcast was I'm going to
ship seven pieces of software using AI.

And your first thing was like, no, you're not.

You're going to ship one.

It's going to be exhausted.

And that's exactly what happened.

Right.

He shipped solo.

He's not making anymore, right?

Exactly.

like a ceiling, regardless of how much AI is doing for us, I feel like there's like a
ceiling of not just our creativity, but also our ability to handle the human aspects of

these and the relational.

Or do you have a vision for a world where that ceiling even goes higher with AI?

I personally, I feel like if whatever the reverse hockey stick is, I feel like a reverse
hockey stick of the usefulness of AI in my life.

Like it's great, it's wonderful, but there are...

their ability to do human level stuff is we're reaching, what is it, diminishing returns?

Do you feel that way or do you like, no, man, we're still hockey sticking up and it's just
gonna get better and better and better better?

So I'm kind of in the middle.

I don't think we're getting to AGI where it's like, it is a human, superhuman.

Like I don't think we're ever getting there or at least not in the near term with how
things currently work.

Maybe there'll be some big breakthrough with some other way, who knows?

ah So I don't think that, but I'm definitely also not on the we are.

dropping now and things are super slow.

I think there's still a fair amount of runway of stuff getting better.

Not so much even um like the models necessarily getting super better, but I think as the
structures around the models get better of like really being able to have agentic

workflows where like, like what I was talking about before, like imagine what every time
you ask the AI something, it just automatically goes out to seven copies of itself and it

just gets the best answer.

and now it's doing even better with that kind of stuff.

It's connected to a bunch of different systems and so it can get all the context it needs.

And as that stuff gets a little bit better about making sure it only gets the data it
needs to know and blah, blah, blah.

So I think there's a lot of stuff that's gonna be built around it that's gonna help you do
stuff better, even if the model itself doesn't get all the way to super smart.

But yeah, I think that, I I love to say that people are going nowhere fast, right?

So there is like,

because you could build so much faster, even if it's not magically fast, it's still much
faster than you could, especially if you're willing to forego looking at the code and all

those things, you can build much faster than you could before.

But ultimately, you still have to have a good idea.

And this is where I think people are like, look at, I built 10,000 lines of code and I
built seven apps and nobody's using any of them.

But I'm a little torn on the idea that you brought up of like,

I mean, listen, I have 10 ideas I would like to build right now.

Is it smart to build them?

For who I am, I feel like the amount of attention and effort I prefer to give to
something, I wouldn't be able to do that.

It just wouldn't be able to happen.

And so I'd have to be okay with like, hey, AI answers all the customer support, even if
it's crappy, they'll be doing it.

Hey, AI is just building everything.

I'll worry about it if stuff goes down, but otherwise I'm not really paying attention to
it.

Yeah.

I don't care about anything until it gets to like $5,000 in revenue and I'm just gonna
kill it off on the people using it if it doesn't reach that.

So if you're willing to be totally ruthless about it, and see like I'm gonna build 20
things, just giving myself more shots and hopefully one of them works.

At the same time, I know very few people that's ever worked for it and it's like the
people who worked harder and smarter usually are the people that came out on top in my

experience.

Yeah.

Yeah.

it so that if I have a good idea and I work harder at it, like now I can execute faster.

I can even more than execute faster, which I'm not too keen on faster being that important
either, but I can do more, which I think can be useful.

So now this product can be broader and maybe touch.

Right, yeah, now I can ship a more fully featured app to begin with.

Now I can.

Expand it into more markets more quickly than I would have been able to otherwise Maybe I
could do more marketing in different ways than I could have in the past.

So like I think all that stuff is There so yeah, I don't know.

It feels like you should be able to build 20 apps at once and Take that approach but then
I see people out there doing it and I don't see any of them who've been like, yeah I'm

like now I'm a millionaire and like I've got four of these that just took off like

It's only so much time and attention from customers available.

I think that's the part everybody's missing, It's like, you could build faster, but the
people who want to buy this product still have limited time, limited attention.

They still have all kinds of concerns that like, because, you another big thing of mine is
like, why people buy software.

I feel like a lot of founders don't think about this and not just software, like in a lot
of products.

Like, why am I buying something?

What is my goal to acquiring it?

And people don't really have a good construct of why people do things they do.

It's like a whole psychological thing you really should get into.

Like you should talk to the AI about this.

Because like if you're a middle manager, like so most of my experience is more B2B.

B2C is a little bit different.

But B2B, like yes, I want this software to accomplish this task, but.

I also don't want to get fired because of it, which is a huge one.

It might be number one, right?

And I want it to be reliable.

And if there's a problem, I want to be able to talk to somebody.

I don't care how much it costs.

This is another concept people, like money might be like the fifth item.

Not for some people, it's the first item.

But for a lot of people, it's not the first item.

Like don't get me fired is the first item, right?

Like don't get me fired is always the first item.

And so...

Again, how do I convince people?

It's all the marketing.

Yes, you've vibe-coded this app into existence, but how do I convince people that it's not
going to get them fired?

And you being totally distracted, obviously not engaged in it, having bad support for it,
that's not going to give people the impression that they are not gonna get fired by

choosing your application.

So that's what it is.

and so much software that's used by especially larger businesses is the junkiest,
nastiest, most out of date stuff ever.

But they've got service level contracts to guarantee certain output uptime.

And they've got salespeople who know how to sell the pitch that it's going to meet your
specific needs.

And so you can, one of the things I see software makers do all the time is I made better
software, right?

This is obviously so much better than X legacy app that I didn't wanna use.

They use it, they hate it, they make an alternative, but they don't understand those
needs.

They don't understand how to sell those needs and they don't go anywhere.

And so the AI, feel like just like you're saying, just, it makes more of that software
where people don't actually understand it.

don't, you know, it's just sort of like, yeah, yeah, yeah, okay.

it's like if I'm, I, like we deal a lot of like accounts payable, because we, not just
credit cards you can pay for help spot with, but we'll invoice you and then we'll accept

checks or ACH or all these things.

And so a lot of these bigger companies want you to go through like their accounts payable
portal.

And these are 100 % of the time, absolutely the worst software you ever used.

The worst software you've ever used.

But it's Cisco or it's SaaS or it's whatever, like it's some big company.

And the person who makes this decision is like, well, if I go with StartupX and their
account payable software, am I gonna get fired if it goes poorly?

And the answer is yes, you will get fired.

Whereas if I just choose Cisco, everybody knows Cisco, everybody's heard of Cisco, yes,
like it's ugly and the people we need to use it have a hard time using it.

But at the end of the day, like I'm not gonna get fired if there's a problem, it's a
gigantic public company that's gonna help me.

And so.

Yeah, I don't know.

This is where I have some concerns for myself in that does the AI just make it so that big
companies suck less, which hurts the small software companies and small companies in

general, because it's like some percentage of the people using the bad accounts payable
software do peel off and say, all right, I'm gonna take a flyer on this company that,

they're way smaller, but I got a good vibe from them.

I had a good demo, like whatever.

Let's take a risk.

But if Cisco sucks less, like maybe they never look for that alternative, right?

And then that leaves the smaller startups in the bed spot.

Do you actually see the big businesses making their software any better with AI?

Because I have not, but I don't live in that world, so...

not I guess in that sense, but I feel like it's still kind of early I don't know I feel
like the stuff's not all there yet, you know, like this is still Very leading edge and I

think these big companies take much longer time to turn the ship a little bit, right?

So it's like Once they adopt these things will they be able to move much faster?

Maybe they won't right?

We'll see like

Right now they're in the old world and they are not there.

So it still takes them a long time to adopt these things.

Now once they adopt them, the AI gets better.

Who knows, maybe it gets so weighed down in all those structural elements of like people
don't wanna get fired and mistakes are unacceptable.

If there are mistakes, I need someone to blame.

Like blaming Claude, nobody wants to be able to take, you come into the meeting with the
CEO and say, well, it's Claude's fault.

Like that's not gonna work, right?

not gonna fly.

Yeah.

maybe those things are so structural to big businesses that like you actually can't power
through them.

And so like the AI is gonna have so much restrictions on it and so many checks on the
checks on the checks that it actually doesn't let them move that much faster.

I don't know.

But I do think it's a concern from a small business.

me right now, when I see big and even medium, like big B2B and even medium B2C companies
adopting AI, it's not even that they haven't gotten to this level yet, it's that their

understanding of the adoption of AI is different.

What they think adopting AI means is putting in unasked for AI-based features that are AI
chat bots and AI whatever, and then charging more money for it.

Right.

are unhappy and resentful and truly I'm the same way.

like, I don't want this freaking AI integration.

I want to put all my personal data in.

I don't want you to replace the ability for me to get real answers from real human with
this stupid AI based feature.

Like literally last week, Imani and I applied for updated passports.

And the last time I did that, um I used a service that made it much easier for me.

You pay them a little bit of money and then they track your passport application the whole
way through.

They review your application to make sure it's all legit.

They understand passport world in a way that humans don't.

I'm like, yeah, I'm gonna pay for that.

Well, I did the one that Clear really recommends and it's called hellogov.ai and I'm like,
I'm sure they just put AI in his branding.

No, it actually is using AI and it's garbage.

I made a typo somewhere.

I think I don't even make a typo.

I think that their system built it wrong, but whatever.

And I was like, oh, I'll go in and add the application and then regenerate a PDF.

There's no way to do that.

Once you've generated the PDF, that's it.

I've been on multiple calls with their customer support.

We're a week into this process and I still can't get them to just click the button to
regenerate the freaking PDF because nobody understands how it works because it's all AI,

right?

That's what enterprise and large business adoption of AI looks like.

Not thoughtful, considerate, how do we build better user interactions for our users using
AI as a part of development workflow, but how do we put .AI at the end of our thing?

How do we rebrand as not calendars, but calendars for AI-enabled companies, right?

They're so, it's not even that they're behind.

It's like they don't even understand the story, right?

Like they don't even understand the benefit they can provide.

So I'm not worried for you at all, Ian.

They don't get it.

I have two notes on that.

One is that totally unrelated to our main discussion, but just so you're aware, the actual
government process for renewing your passport is actually super easy, and you should have

just gone direct, and next time just go direct.

Because actually, we've done it a few times here recently, and it's like absolutely
nothing, it's been great.

um But yeah, then the other thing, the other side is like people looking too far out, and
it's like, like I just saw a thing.

a couple days ago where Meta is now requiring every employee to have tracking software
that tracks every mouse click and keystroke.

All that's gonna be fed to the AI and screenshots their screen so that the AI can learn
what they do at their job.

And I'm just like, I don't think where I click, that is part of the job at a big company,
but there's the meetings.

Are you just gonna be able, I assume, wholeheartedly replace everybody you think?

I don't know, it's very weird.

That seems like the super ultra futuristic.

yes, the robots will do every job and that'll be it.

And I'm like, that doesn't seem like it's gonna work either.

So it's like, yeah, they haven't right-sized totally.

do think that, like I think Salesforce is kind of doing an interesting approach.

I don't know if you saw what they're doing, like there's now like Salesforce has the
option to be totally headless.

And so everything is accessible via MCP.

so I think there's like stuff like that where like, hey, we're just gonna like open
everything up to other AI.

can use the data that we have.

And even though our whole model is based on like humans in seats, like over time,
presumably there will be a lot of like just data interaction between different systems,

which is already true, but it just couldn't do as much with it, because there wasn't
anything in between that could like analyze.

So I think that's kind of interesting.

definitely, that's the most AI thing we've had direct requests for and which we're, like I
said, shipping soon is the MCP server, because people want to be able to like,

Interactive to help those tickets and do other stuff and other systems with that and
whatever So there is this thing happening I think is really happening of like the data

being able to move between places more easily and what that impacts but Yeah, I mean
nobody knows where it's gonna end up right who knows it's hard to tell how far it's gonna

be and all that uh But yeah, I don't know small business will hopefully survive probably
survive

It'll surprise.

Yeah, it'll be good ah Gosh, it sucks cuz I was like, we're gonna do an hour and half and
I will not run out of topics and Alright, we'll finally run out of topics.

I would just that topic alone I could talk to you for another 20 minutes and we have to
start wrapping it up.

Dang it though I'll try just do one last little note on that which is that uh The I had
Kent C.

Dodds on a couple episodes ago.

It hasn't released yet.

So you haven't heard it

but he talked about MCP and MCP being, I was like, how is MCP any different from API?

And he's like, MCP is different than API because all the APIs in the world are like,
they're spilt in a million different ways and you have no idea if it's gonna be REST or

whatever.

MCP is a much stricter definition and yeah, it is for AIs, but it's also just much
stricter.

there, you can assume that all MCP servers, which are basically just a subset of APIs are
all shaped a specific way.

If someone were to come along and say, we are going to somehow force

all of the websites on the internet to expose their data for free through a specific,
really narrowly defined API.

Even if it wasn't in the world of AI and MCP, I feel like that would have a massive impact
on how good do uh their UIs need to be, right?

Like how many barriers are there between me actually building my own tools integrating the
thing?

And like we had to move towards that a long time ago that has all kind of stopped, you
know, Twitter had a great API and then they...

Right?

it and also put it behind paid thing.

You know, I've been building lots of internal tools at Tighten and the number one thing
that keeps me from being able to do it is because the bigger the company is, the less

likely they are to have a free API that allows me to reasonably get data out of it.

So it's curiously if the number one thing that AI does to all these big enterprise tools
is forces them to put their data out in a freely accessible, consistent way.

On the one hand, it's definitely to the benefit of all of us.

On the other hand, it definitely changes the shape of our offerings as small businesses,
because now maybe we can build things to integrate with them that we couldn't before.

But like you said, now maybe they're here feeling the hit of having a really crappy UI
that nobody likes working with less because nobody's using their stupid UI anymore anyway.

Right.

So it's weird.

It's sort of one of those like maybe one business goes out of business, but now it's made
opportunity for another business to go into business.

So, yeah, who knows who knows where it takes us.

hard to know, right?

If you could just access the data across everything in a more efficient manner.

Like I said, with AdWords, it's totally, I'm now spending thousands of dollars a month in
AdWords I wasn't spending because I can avoid their user interface, was, I'm sure makes

sense in some world for you're spending billions of dollars through it, or millions of
dollars through it, you're Proctor & Gamble, and you need every one of these little dials

and whatever.

But for like the guy who just wants to spend five grand a month, like it doesn't make any
sense most of those dials.

um But now I can avoid all the dials and I can just spend more money with Google than I
have in the past, right?

So.

Is that an MCP server or is that Google AdWords API or what are you interacting with?

Okay.

and it's like weird, have to like do a thing with it.

You have to become like a developer.

You get approval through a developer account.

So this is not even, this is not easy yet.

This is not what you're talking about.

Yes, exactly.

Yeah, so it is interesting though that like the MCP makes it all easier, but like a lot of
it's still possible as long as there's some API.

Cause like once I got access to the API, then the AI just wrote a bunch of Python scripts
to interact with the API.

And now it just,

does that and just interacts directly with the API, um which it's good at that stuff too,
but the MCP is just nicer.

The authentication's like integrated as part of it and just cleaner and simpler.

yeah.

But yeah, I mean, there's gonna be a lot of opportunity there for like data stuff, right?

To put it roughly, the data stuff is gonna be an interesting area.

I don't have any ideas in that area, unfortunately, but I feel like everything's gonna
have an...

API MCP of some sort and then you're gonna be able to do stuff because of that and There's
gonna be a lot of money made there.

I just I don't know what the money is, but there's gonna be something there for sure

who expected the the world of AI to actually potentially open up the internet again?

Right?

Because like the the early days of the web were all about open data and open everything
and then the last, know, whatever 10, 15 years has been all the big companies owning it

and then shutting it down.

And like now they're kind of being told, like, if you want to hang in the world of AI, you
got to open your data up again.

I'm like, well, I'll take that as a win.

So um so I know that we're almost out of time.

have to.

kind of read one thing, but before I read the one thing at the end, is there anything that
you hope we were gonna talk to or any topics we didn't get to finish that we didn't get to

get to today?

Um, no, think we did a pretty good coverage.

Yeah, we did a lot.

We hit a lot of areas.

I'm pretty happy.

All right.

Well, let me read our one community suggestion.

So Patrick Amara said, recently had a puppy join our family, Lola.

ChatGPT wrote out a crate and toilet training plan, gave recommendations on shoe toys,
food, teething, vaccinations, training schedules, et cetera.

And I keep telling people, I'm like, of the benefits of AI is it's a better Google, right?

Because you probably would have Googled every single one of those things in the past.

You would have read 17 websites and then you eventually would have put together a plan.

You're like, go do the Googling and reading and looking at the signals to figure out which
of these websites is more or less good and find which recommendations are consistent

across every website versus which ones aren't and just do all that work for me, right?

Like it's like there's no, it's not necessarily taking the place of a human, but it is
taking the place of an existing technological tool.

One thing I didn't ask you Ian was like, you know,

Do you use AI to plan all your personal life things?

Do you use stuff for your pets, for your kids, for your travel, or are you mainly using it
in business contexts?

no, I'm all in, all in on the personal stuff.

Again, I'm not all the way to like, it's fully automated everything for me, but it's
definitely, uh replace Google, it's kind of interesting right now.

Google was very smart, I to say, by like very early on, just having those AI summaries of,
if you ask a non-commercial question, you just get the answer there.

And so that's actually kept me on Google for like quick things, because I'm just like
autopilot, like in the,

in there, okay, I just do it.

But if I have something a little more, it's gonna be a longer reply or I want a better
answer, because that's like a light Google model, then I definitely go to Claude or

ChatGPT and like even all kinds of interesting stuff, like if I'm going to a new city and
I'm planning on doing photography there, like I'll be like, okay, what are like the, I'm

interested in doing minimalist photography, XYZ, like what are areas of the city that are
good for that?

What are specific buildings?

And it'll have, now some of that's gonna be stuff that's very popular with other people.

Maybe I don't wanna just do the same picture as a million other people have done.

But it gives you like that kickoff point of like, yeah, and like to find that on your own
would actually be quite hard even with old school Google.

um So yeah, so it's like so cool for just stuff like that to be able to throw stuff on a
map or build out, um yeah, just general questions about daily life stuff.

I'm very excited for it to replace doctors for regular doctor stuff, which I know might
not be your thing given our current discussion.

But doctors are so bad and useless now.

They're so useless.

It's like, oh, you have three minutes to describe everything that's ever happened in your
life as well as what's wrong right now.

And so you just give them what's wrong right now without any context.

it's like, just to be able to be like, I feel like there's gonna be, I mean, just today,
ChatGPT launched an AI specific,

health-specific model for doctors, specifically for doctors.

And I was like, here we go, yes, let the doctor, let the AI summarize my whole history for
the doctor so the doctor can spend a minute before they come in and be like, I know you've

had this and that in the past.

your issue is this now, that's probably related to the other thing.

Whereas they never know about the other thing because they never look at any of it.

If you didn't pre-research it to tell them about the other thing, they're never gonna
know.

Whole thing is bad.

I can be on board as long as the GPT, as long as the AI is not the one making the final
health decisions for it.

Same thing with programming, right?

Like if it makes the actual human act more informed, more blah blah blah, cool, I'm here
for a bit.

But I know people who are like using chat GPT for therapy and for diagnosis.

I'm like, no, no, please, I can't go that far.

Yeah, yeah.

But I don't know, for the regular doctor stuff, like hey, what's this thing?

What about that thing?

I feel like AI's gonna be pretty good at that.

um So we'll see.

Yeah.

Well, like I said, I could talk to you for hours more, but thank you for hanging out for
so long.

ah Like I said, of the people who are very, very, very pro-AI, I think you're one of the
ones I trust the most.

So was really fun kind of getting to nerd out a little bit with you today.

And yeah, thanks for hanging out, man.

Thanks for having me on always love being on your podcast the best run podcast out there
Be excited for outro to be part of that someday, but Yeah, thanks for having me on this is

great always a big fan of all your stuff of course so keep at it

All right, appreciate it, you too.

And for the rest of you, we will see y'all next time.