April 19, 2026 • 55 mins
We map the shift from chat-based AI to real agentic systems that can run workflows, manage memory, and publish results without constant human babysitting. We break down how to design AI agents like jobs so they stay grounded, cost-controlled, and useful for marketing, SEO, and software work.
• tiering AI use from assistance to automation to full agents with memory and tools
• writing a clear job description with input, process, and output formats
• building an SEO research agent that generates briefs and scores novelty
• grounding agents with state, scripts, and tight context management to cut errors and token waste
• why models ignore guardrails and how system design fixes repeat failures
• plan mode tradeoffs and where to focus review time: planning, code, or tests
• orchestration patterns like dispatchers, specialist sub-agents, and clean bridges between systems
• security and scalability realities for enterprise-ready software built with agents
Guest Contact Information:
More from EWR and Matthew:
Leave us a review wherever you listen: Spotify, Apple Podcasts, or Amazon Podcast
Free SEO Consultation: www.ewrdigital.com/discovery-call
With over 5 million downloads, The Best SEO Podcast has been the go-to show for digital marketers, business owners, and entrepreneurs wanting real-world strategies to grow online.
Now, host Matthew Bertram — creator of the LLM Visibility Stack™, and Lead Strategist at EWR Digital — takes the conversation beyond traditional SEO into the AI era of discoverability.
Each week, Matthew dives into the tactics, frameworks, and insights that matter most in a world where search engines, large language models, and answer engines are reshaping how people find, trust, and choose businesses. From SEO and AI-driven marketing to executive-level growth strategy, you’ll hear expert interviews, deep-dive discussions, and actionable strategies to help you stay ahead of the curve.
Find more episodes here:
Rate the show in your podcast app. Apple ratings are how new listeners decide whether to press play — an honest one takes about thirty seconds. Steps for iPhone, Mac and Spotify:
www.bestseopodcast.com/review
Follow us on:
Facebook: @bestseopodcast
Instagram: @thebestseopodcast
Tiktok: @bestseopodcast
LinkedIn: @bestseopodcast
Connect With Matthew Bertram:
Website: www.matthewbertram.com
Instagram: @matt_bertram_live
LinkedIn: @mattbertramlive
Powered by: ewrdigital.com
Listen
Watch
Mark as Played
Transcript
Episode Transcript
Available transcripts are automatically generated. Complete accuracy is not guaranteed.
SPEAKER_00 (00:02):
This is the unknown secret of internet marketing.
You're inside a guide to thestrategy of top marketing union
to craft the competition, readyto unlock your business full
potential.
Let's get started.
SPEAKER_03 (00:18):
Howdy, welcome to another fun-filled episode of
the unknown secrets of internetmarketing.
I'm your host, Matt Bertram.
Uh, I would tell you, mycomputer might be uh you know
maxing out right now.
I got uh too many thingsrunning.
We got some uh clawed codegoing, we got uh a couple
different agents set up, and soI probably need to close some
(00:38):
browser windows and turn somestuff off.
So uh if there's a little lag,guys, apologies.
Let me try to shut some of thisstuff down.
But as the conversation hasmoved more and more towards the
agentic realm, and we're talkingabout LLM visibility um and just
kind of how to think about allthese things.
I thought I would bring somebodyon that is deep in uh AI agents,
(01:02):
uh setups, audits, uh setting upthat AI agent to help you uh
help with your workflow.
Uh Sebastian, welcome to theshow.
Thank you so much.
SPEAKER_01 (01:13):
Pleasure to be here.
SPEAKER_03 (01:14):
And Sebastian's with uh fountaincity.tech.
Uh you can go check it out anduh I'll let you talk a little
bit more about it at the end.
But you're doing everything AI,Sebastian.
So I thought I'd bring you on.
SPEAKER_01 (01:26):
Yeah, thank you so much.
There's a lot to talk abouthere.
So where would you like tostart?
SPEAKER_03 (01:32):
Well, yeah, the I mean the market's moving really,
really quickly.
I I felt like open claw was uh alittle bit of a distraction.
Um we were setting it up, but itwas having some different kinds
of problems, and we've justpivoted back to quad code.
Um and uh just kind of now theyleaked the harnesses.
So like I think everybody'sgonna have a pretty good
(01:53):
harness.
I know um, you know, whatFacebook's doing is really
interesting.
Like we've been using Mantis alittle bit because you got the
you got the rag recall, quoteunquote, from all the Facebook
ads.
Uh, so that's helpful.
Uh, but it would be good tomaybe talk about how you like a
mental map on how you look atall these different like
(02:14):
frontier models and how you'remaybe using them together.
And like even like uhperplexity, I've been playing
around with uh Comet andComputer, and it's really
agnostic, right?
So it'll just use whatever'sbest to kind of build it.
It's like a yeah, Po 2.0.
Uh, if we're moving from uh thechat bot side to really the the
(02:36):
agent side of things, I wouldlove to just kind of how would
you paint the picture to someonethat is around this, but maybe
is not as deep in it as you are.
SPEAKER_01 (02:47):
Okay, yes.
I think what's coming into myhead as you're talking about all
those different things is firstI would classify the different
usage into tiering as kind of away to kind of focus then in
later on certain aspects of it.
So AI assistance is kind of tierone where you're just using AI
to help you along with whateveryou're doing.
And sometimes that might evenlook at how people are using
(03:10):
cloud code, but certainlyperplexity or a comet browser,
you know, you're and then thenext level up, uh, tier two,
you're using AI to performcertain automations that could
be on a clock, that could betriggered.
And the workflow automationitself might only have AI as one
of those steps in the process,but there's AI, you know,
somewhere in there, otherwiseit's just a normal workflow
(03:32):
automation.
Tier three is where I startthinking about the AI
agentically, meaning it hasmemory tools and agency, meaning
it can take its own actionseither on a schedule or
triggered as well.
And then if you want to gobeyond that, there is levels
above that, but already at thatthird level, you have within
that two different categoriesalready as well, A and B.
(03:54):
A would be it's driving mycomputer, right?
Which is more like co-work orperplexity.
I forget the name of it, whenit's driving your computer
computes, I think is that whatit's called from uh perplexity
computer.
SPEAKER_03 (04:07):
Um, and then like claud co-work is the what's
driving uh is is the shared kindof uh online component.
And then uh claude uh code iswhat at least how I have it set
up, is driving mainly mycomputer.
And so we're starting to putthings in um like uh Apple Minis
(04:27):
where we're we're playing aroundwith like ChemKey 2.5, uh and
and we're we're because thetoken usage, unless you really
manage it, like putting like aslash context or something in
front of it, it will just burnthrough the tokens so fast.
So, like tokage optimization hasbeen the biggest area that we've
(04:48):
been trying to tinker withbecause it, you know, you get it
to do a lot of actions that getexpensive.
SPEAKER_01 (04:53):
Oh, totally, yeah.
And so that that second categoryis more the always-on category
that could be, like you said,like a Mac mini in your house or
something like that.
We, I mean, we're we're buildingthese systems for clients, so
that's not really an option.
So we put them up in the cloud,so it's like AWS EC2 servers or
stuff like that, you know, orGoogle, you know, whatever
(05:14):
flavor of system the client isin, we'll just embed it into
that ecosystem.
Uh within, but yeah, and well,you just talked about cost
management.
I think there's different sotalking about open claw or these
kind of systems that haveagentic capabilities into it.
There's so many different waysto build that system.
(05:35):
I think the first way peoplethink of it is, you know, maybe
they're experimenting with it,they do conversations with the
system, give it some abstractgoal.
The AI goes off and tries toaccomplish that abstract goal,
burns, you know,$50 a day inyour wallet of token usage, and
you're like, oh my god, I don'teven know if I'm getting value
out of this, you know.
So we actually do so we've beenbuilding a lot of these
(05:57):
autonomous workers or autonomousagents.
And there's uh we like to thinkof the agents kind of
categorized also in terms ofthose different tierings that I
just mentioned.
So we might have one agent thatreally has a very specific
workflow.
SPEAKER_03 (06:13):
Are you what are you using for orchestration?
Are you using like Langchain orsomething like that?
Or like I'm trying to understandwhat the tech stack is because
I'm starting to use quad codeand hooks and skills and like um
it's spinning up other agentsunderneath, like whatever it's
trying to do.
I I think I'm probably teeteringbetween like uh if you framed it
up, between two and three,because we're just setting um
(06:36):
triggers like daily or likeautonomous, like okay, go go run
the script.
And then there might be aagentic layer on decision making
at some point.
But like, can you back up andbreak down like what is the tech
stack that you're building on?
And like what are you settingup, or maybe even give an
example of something you'retrying to solve for a client?
SPEAKER_01 (06:58):
For sure.
Yeah, I will add one pre-step towhat I'm about.
I'm gonna answer your question.
I'm gonna say one thing first,which is I think the most
important first step is tofigure out what you're trying to
do and describe that.
I like to think of the coming upwith a job description of what
what is gonna be built.
And then within that, reallythink about the input.
What is the context?
(07:19):
What's gonna be the input foreach job?
What is the overall generalinput, like you know, first day
training on the job, what youneed to know.
Then what is the process?
How does the sausage get made inthe factory?
That's a really important step.
That's where your quality comesin and not just volume of slop.
And then what's the outputformat, right?
And so once you have thosethree, then you can start
thinking about the architectureand how you're going to
(07:41):
accomplish that goal and whattier level you need and
combination and so forth.
So, with that in mind, let's saythat.
So uh so we have one agent, forexample, that does SCOGO
research and builds content thatfeeds into a content pipeline
with ideas and so forth.
And so within that jobdescription, there is a whole
(08:01):
need to understand the contextof the business at heart, but
then also who are thecompetitors, uh, what platforms
do we want it to research everyweek.
So we have it researching allposts from all competitors, plus
Reddit, plus X, plus Substack,probably a few, and then just a
general uh mentions searchacross the whole web of who's
(08:21):
talking about what.
All of that's compiled alongwith you know documents that
describe the objective, and thenall the Google Analytics is
read, the Google Search Console,keyword performance, AI ranking
performance is also tracked,like all of that's input, and
then from that, briefs aregenerated according to a format
of how we want the brief tolook.
(08:42):
Those have actually havestrategies as well of what kinds
of briefs, and then that's theultimate output at the end of
that agent.
Is like, okay, here's the thingsthat should be written.
And we do also score them interms of um I forget the name of
it, but it's a scoring that weuse to determine the tier one is
that it's very novel, uniquecontent, right?
So it's it's it's authoritative.
(09:05):
Tier two is the uh the researchis going to be novel, so no
one's researched this before.
Tier three is the framing isnovel.
So the research and the contentisn't, but the way we're framing
it is, and then tier four iswhat we call commodity content,
and then depending on theclient, that type of content may
never get go further, right?
That's like a barrier, likebecause that client doesn't want
(09:27):
commodity content.
Commodity meaning if itdisappeared from the web, no one
would ever blink an eye, right?
That has no intrinsic.
SPEAKER_03 (09:34):
And so you're you're using you're using like uh APIs,
MPCs.
There's kind of some new stuffthat's coming out, as well as
maybe some like uh uh you know,a data lake or even like a
rudimentary Google Drive.
SPEAKER_01 (09:50):
And so yeah, I can go to the technical stack next,
but I think it's important toframe like what's the job first,
because from that, then you canstart moving more and more into
like okay, the nitty-gritty.
So in order to achieve that,there has to be a memory
management system.
So that could be SQL database orobsidian QMD.
Uh really it doesn't have to beRAG necessarily, depending on
(10:12):
the volume, could be justrelational.
And so there's that piece of it.
We're building it on an EC2,like I said.
You need to have a clock thatgenerates the reoccurrence.
That could be a prompt job,could be open claw, you know,
could I mean we're um uh so fromfrom that clock then that's
creating the occurrence, you'vegot the memory that, then you
(10:33):
have the skills that aredeveloped in order to create
consistency of behavior.
And then we uh there's a blogpost on my website that talks
about four things to really lookat when you're building, which
is if I remember them all offthe front of my head, uh ground
it.
So you want everything to bestate-driven.
You don't want the AI model tojust kind of remember what the
state of something is and trackit in a markdown file, right?
(10:55):
You want to put it either in aJSON or table or something for
state-driven decisions.
You want to script it as much aspossible.
So you really don't want the AIto be thinking through how to do
things over and over again.
Even with a skill, I it's stillnot um, you're still gonna
create error rates because AIsare probability machines, right?
So you want to create scriptsthat the AI actually triggers.
(11:18):
So a skill triggers a script,the script is a set amount of
code, the code runs, it's not AIanymore, and it's also cheaper,
runs an operation, stores statesinto your data store, and the AI
is just taking data back andthen doing the only the step,
the only step that AI should bedoing is the interpretation.
Like, how do I interpret thisinformation?
(11:40):
How do I make a qualitativedecision in this case based on
this data with this context?
You do have to manage alsocontext because like the more
skills you have, the less inputtokens you have.
So that's like an interestingbalance right now in the
ecosystem.
Um but then so you so that's uhthat's the memory agency, and
(12:02):
then the skills and tools.
That's really the maincomponents of the system, right?
And then from there, everythingelse is just driven by the input
output layering.
So for us, those briefs are aninput to another agent, which
then takes it further.
And the ultimate final output isthat it just publishes it on
whatever website or social mediaplatform is connected to the
system or newsletter, whatever.
So that's the ultimate output.
(12:24):
Uh, and then the input layerthat can be that research I
talked about, but it could alsobe transcripts or video
recordings, or it could be evenother AI agents.
So we've coupled also AI fullyagentic systems that build
software, and then those agentsnot only build the software but
write case studies about whatthey did, the problems they
found, how they solved it, andso forth.
(12:44):
All of those lessons learned getfed into the system that create
novel content that is then tierone for novel content, right?
Because like it's act it's notjust research anymore.
It's not, and so you've got thatyou can actually have highly
novel content going through yoursystem end-to-end, which is like
you don't even need a personanymore to do some of that
(13:05):
novelty in the content,especially if it's like you
know, research driven or thingslike that.
SPEAKER_03 (13:09):
Yeah, that that is actually something that we were
talking about earlier today.
Uh, everybody listening, it'sMonday.
And so we were we were talkingabout how to uh pull some of
that in.
Uh and we haven't quite figuredthat piece out yet.
But that would be great to uhtalk to you more about that,
because that's actually whatwe're looking at is like, okay,
the data and the data pointswe're producing, how how do we
(13:31):
turn that into case studies?
Because that's um unique data,right?
Um, it's it's data that we haveaccess to.
And, you know, so that's very,very cool.
Very cool.
Well, that's I think that that'swhat you talked about is where
the market's going.
And I think that uh a lot ofagency owners are trying to kind
(13:52):
of figure out that and are atdifferent places and getting
that done.
Where do you see, like if youtried to project out, because I
know it's hard because thingskeep changing, um, where do you
think everything's ultimatelygoing at at the end of this?
I I've been listening to somedifferent people that's in time
or in 10 years, like what whattime?
Yeah, well, uh I you know,everything's moving pretty
(14:14):
quickly.
Like, let's say like threeyears, like where do you see
everything converging in inthree years?
And like, where's thatsuppression compression
happening in your mind?
And what is the world gonna looklike?
SPEAKER_01 (14:27):
So the compression really is between idea and
delivery.
And another way to look at itwould be problem solution or
opportunity and action.
Whatever you want to frame, youknow, the beginning and the end
point, everything in between iscompressing into a slice that's
gonna get smaller and thinnerand thinner.
(14:48):
There's certain areas right nowalso that people think are
relatively protected, domain,you know, domains that humans
have an advantage over AI.
And uh, I I think though a lotof those are just rugs that
we're standing on that are gonnaget pulled out from underneath
us over and like subsequentlyone after the other.
It's a very similar trend thatwe've seen with computers.
You know, if you look at, youknow, first we thought it could
(15:09):
never beat a human in chess.
That happened with what was hisname, Kasparov.
Then we thought it would neverbeat Go because the domain
spaces were too big, and thenthat happened a few years later.
Then we thought the Turing testwas unbeatable, that was beaten
in 24, right?
So it's like we keep puttinglines there and we think, oh,
it'll never pass that.
I think the really true so Ithink humans really have four
(15:31):
main areas that are kind of themain areas to be at play with
the with the direction thingsare moving, and these the four
areas really to be thinkingabout for how humans are
involved.
The first is direction.
That could be direction of thewhole business, but it could
also be direction within yourdomain, right?
Within marketing or withindesign.
Some people call it taste, butyou know, it's it's ultimately
(15:52):
like what is the differencebetween option A and B?
Like, I guess even moreabstractly, you know, if my kid
came to me and said, Should I bean artist or a doctor?
And it's like, well, what'simportant to you?
You know, like these are soabstract choices that someone
ultimately still needs to makeuh call it a taste decision or
direction within that thing,yeah.
SPEAKER_03 (16:11):
So taste, trust, governance, um, and then there's
like one or two others that uhI've started.
SPEAKER_01 (16:19):
Yeah, trust I would connect to the second quality,
which is the interactions layer.
Okay, people want to interactwith people.
And so whether that's service orsales or support or teaching or
understanding or empathy andtrust and all these kind of
things are kind of on theinteraction layer.
And so that's not going away,you know, that insofar that
(16:43):
people are actually good atthat, right?
Certain types of interactionlayers will eventually not like
people won't even want it, youknow.
The time will come where you goon and you do a phone call for
support, and if you get aperson, you're gonna think, Oh
god, not a person.
I wish it wasn't the guy I wastalking to, because it's gonna
be so much better.
Like, why is this company stillhave a person doing this job,
you know?
So, because you're you know, forthings that are more like
(17:04):
transactional or findinginformation or getting an
answer, right?
Those kind of things again aregonna just compress to the
point.
And I've already experiencedthat, by the way.
Like there are certain companiesout there that have figured this
out.
And when you do support, likeI've done tech support with this
one company, it's all AI, andit's at first I was like, okay,
how's this gonna be?
And it turned out to be amazing,like way better than like super
fast answers, really accurate,and knew every single question I
(17:28):
asked, like instantly, and I wason and off on that support call,
like within minutes.
Normally that'd be like a15-minute call or more.
So me ask my supervisor, right?
Like, is the things that likewould don't need to happen
anymore.
Anyway, and then so those twoare kind of horizontal type
jobs.
And what I mean by that is theycut across disciplines, so
(17:49):
there's compression that's onthat compression that happens
between idea and delivery willbe most experienced by people in
those two horizontal domains.
So right now, you know, if youwant to build a website or
something like that, it's likethings are very staged between
people.
It's like, oh, you have an idea,okay.
Well, let's get the designerinvolved to design it.
Okay, let's get the UX UI study.
(18:09):
Okay, let's let's get theprogrammer involved, figure out
how long it's gonna take.
Okay, let's do who's the testerneeds to come up and come up
with the test.
Like all of that's compressingso that the people in these two
domains of direction andinteraction get a much quicker
experience from beginning toend.
And it's compressing already aswe speak.
The other direction, the othertwo uh human involvements are
(18:31):
what I call vertical instead ofhorizontal.
And they're just they're thebuilders and the supporters, so
they're the ones building andsupporting and maintaining this
infrastructure that is the thingthat is the special sauce,
right?
Like it could be yourdifferentiation, could be uh,
you know, all of your moatthat's created from your data,
your specialized processes, theproof that you've run this a
(18:54):
hundred times and on thisparticular type of problem, and
you've gotten these outcomes,which are much better than
anyone else, you know, whateverthat is, they're the ones
building that secret sauce, thatvalue in there.
And you know, that's all thelike the building of the agents
and the optimizing of them andthe memory management systems
and helping the AIs to replicateand to fix their problems, and
they're the ones behind theglass looking over the machine
(19:17):
floor at the robots that arebuilding the cars, right?
Like when we went from becausewe're kind of transitioning in
that in the agency world fromthe world where you've got 10
people on the floor who are eachputting together a different
part in the car, and right, andsome people are trying to figure
out how they can get a robot topass them a tool on the floor,
right?
That's like AI assistance.
Can you give me the right toolin anticipation of my need?
(19:40):
And then the shift with theagents with agentifying the
whole thing is like, no, no, no,just get off the floor.
We're gonna put you behindglass, and you're gonna look
over the hundred robots.
Five people look after thehundred who are building the
cars.
And there's still work to bedone, there's a lot of work to
be done for those people, butit's a very different kind of
job, is what you know we'reseeing.
It's architecting, uh, you know,translating things that are in
(20:05):
people's heads to now bequantifiable into a process,
right?
So it's a lot of processengineering.
SPEAKER_03 (20:10):
It's it's agent, well, it's human on the loop,
not in the loop anymore.
SPEAKER_01 (20:15):
They're just yeah, or human behind glass is what
it's like this year, right?
It's like it's not even on theloop anymore, it's just behind
the glass and looking at thedata.
SPEAKER_03 (20:23):
Yeah, well, it's just like auto-approve, like set
set the guardrails and likeauto-approve.
SPEAKER_01 (20:29):
It's we have yeah, we use tiering for risks, and
that's how we determineauto-approval.
So things that are low risk, andthat bar keeps going up.
What is low risk?
Yeah, those are auto-approved,medium, it finishes all the way
to the end, and then there's acheckpoint at the end, and then
high risk, it's a gate at thebeginning and the end, right?
So that's how we kind of there'sdifferent ways you can measure
(20:49):
it, but that's how we do it.
SPEAKER_03 (20:50):
And and I, you know, tell you tell me you're like I I
feel like I have a I have anagent set up that's just
watching, like making sure thateven though here's the
guardrails, like sometimes theydo crazy stuff.
Like, why why is that?
Why why would you say that ifyou give it definitive like
guardrails, it sometimes doesn'tfall it.
(21:12):
And I have to have like anobserver uh agent watching that
agent, making sure that it it itdoesn't, you know, hard code in
um you know uh API key or so,you know what I mean?
Like, why does it do it even ifyou tell it not to do it?
Like like I I would I would sayit's like data hygiene, but you
know, if this is a new project,like there it's like it
(21:35):
shouldn't be an issue, butsometimes it is.
SPEAKER_01 (21:37):
There are different reasons that could be happening.
I'm not gonna be able to tellyou one reason without actually
diagnosing your particular case.
But one reason, I'll just gothrough a bullet list and you
can dive into any one of them ifyou want.
So one of them could just be themodel mismatch.
You know, you might have a modelthat's just not good enough for
that.
Like if you're using haiku orsomething equivalent to that,
they're very pattern based.
(21:58):
They're looking for patterns andthey replicate them.
Rather than thinking throughproblems.
And then the mid-tier modelslike Sonnet, they're better, but
they can definitely, I've seenthem definite.
What you're talking about, likehard coding values in, like, oh,
your test case says that in thistest it should come up with this
answer.
So I'll just hard code thatvalue with a magic number in the
code, and all of a sudden yourpat your tests pass.
(22:20):
And it's like, oh my God, Ican't believe you just did that.
You know, that's not the goal ofthe testing, whatever.
But uh another category, so youknow, you might just need a
stronger model, right?
That sometimes that fixesthings.
Uh like Opus is very good atfollowing instructions, and uh,
we've been experimenting a lotalso with GLM5, uh 5.1.
SPEAKER_03 (22:39):
So you just think if they're getting get better, like
they're just gonna like that'sone, that's one vector, right?
SPEAKER_01 (22:44):
Of like possible reasons.
Another is context and memorymanagement, and and you could
also include skill managementand tooling as like all part of
that category.
So if you see, for example, andthis, oh, and then the the next
thing which ties into this ishow do you approach problems
when you see them?
So whenever we see a problem, wealways ask ourselves, you know,
(23:08):
the the question isn't why isthis model making this problem?
It's what can I do to improvethe system to make sure that
this problem doesn't happenagain, right?
And so that framing is reallyimportant to take.
And sometimes I see people justget start writing angry messages
to the AI or tell it moresternly how it should be doing
its job, you know, and like thatmight get you some results, but
(23:30):
it doesn't mean it's it's gonnahappen again to you, right?
Because what you're not doing,you have to really be thinking
about system design.
So, you know, so take a crashcourse on like how to be a good
system designer and look at whatare you know what are the top
five, 10 things you need tolearn to be that to embody those
qualities, and then that's howyou should be approaching the
problems.
So if you have an AI that's notfollowing the instructions
(23:51):
consistently, maybe you need,you know, you can even ask the
AI, do you have a good skill forthis?
What skill are you using?
Or you can look at the log ofhow it did it.
Maybe it tried something and itgot an error, and then it tried
something else, got an error,and then it tried a third thing,
and then it found the answer,and then it took that, and then
you can kind of watch what itdid.
And if you're seeing that it'skind of bumping along the way
(24:12):
like a pinball machine, you cango through that with the AI
afterwards and you canstreamline it with a skill that
tells it immediately the correctway to do it, so it's not having
to bump around like a pinballmachine.
Sometimes people, the prompt orthe system prompt or the thing
that they're writing is notactually phrased in the way that
is the most effective.
As models get stronger too, youreally won't be focusing more on
(24:36):
why you want to be doingsomething, which is often
forgotten, that you need tocommunicate that to the AI and
what the desired end goal of itis, and not just the steps, like
I want you to do step one, steptwo, step three, step four.
That can work, but especially asyou get to stronger models like
Mythos that's coming out, youwant to really be making sure
that you're guiding these AIsystems with goals and purpose
(24:59):
of what they're trying toachieve, because that helps the
AI a lot to make sure that it'snot going to shortcut you with
like a magic number in there.
Because if you're telling itlike we want to make sure that
this works for all possiblevalues of this variable, in
this, you know, here's going tobe some test cases, make sure it
passes those tests.
That's very different thansaying, here's a test, uh the
(25:20):
answer should be 1.5, go testit.
And then the AI is like, oh, wewant it to be 1.5, let's hard
code that thing in.
SPEAKER_03 (25:28):
So so I saw a podcast interview with the
founder, and he was just like, Iuse plan mode like for almost
everything, right?
To to to to give it thoseobjectives to answer those
questions, which I think areextremely helpful.
One of the things you made methink of was like, how are you
solving like persistent memory?
And you you mentioned obsidianearlier, and and we've started
(25:52):
to toy around with that.
I'm not like an advanced user,um, but I I would love to know
kind of like how you're solvingthe persistent memory and kind
of like connecting it alltogether because you only like
you said, you only have so muchin the context window, and when
it compresses it, it kind of itit loses stuff, and you got like
(26:14):
the heartbeat component.
Like there, there's just like uhwith the so many things I can
say now because you keepmentioning new things.
I know, I know, I'm sorry, likeclean it up for me, clean it up
for the listeners of like I do.
SPEAKER_01 (26:26):
I want to also just touch base on plan mode because
plan mode is awesome when youwant to be building things, but
there's a very big differencebetween planning, which is what
to build something, versus usingsomething.
You can't use plan mode to usean identification.
Yeah, so just want to make surethat that's clear.
But in terms of building, Ithere's a lot I could say around
(26:47):
plan mode, how to make that evenmore effective.
Let's pull it out.
I don't know.
We just want to we don't wantto.
SPEAKER_03 (26:53):
No, I don't know what's let's go down some rabbit
holes of like and let's likeactually give some context.
So let's talk about plan modeand we'll move into the okay.
SPEAKER_01 (27:02):
So there's first of all, there is a big debate right
now between where to get themost leverage and plan in in the
planning and execution whenyou're doing agentic coding.
So you're using a team of agentsto write and execute the code
and self-test and all that kindof stuff, right?
So it's not AI assistance whereI'm just having it write one
function at a time, one class.
It's like doing the wholerefactor or the whole project.
(27:24):
Um, those three points ofcontention are should humans be
spending most of their timeduring the planning?
Should they be spending most oftheir time during the code
review?
Like some people are sayingright now.
Some people are changing theirmind too.
They some people used to say,read all the code.
Oh, that's right, sorry, readall the plan, don't read the
code.
And then those same people arenow saying we made a mistake.
(27:46):
You don't don't read all theplan, read all the code.
But there's a third place toalso, there's a third philosophy
or a third camp, which is reviewall the tests, right?
So that's where you really needto be putting all your energies
at the end point, is thetesting.
So you're kind of doing atest-driven setup.
So the I do think personallythat that contention of where to
(28:06):
put the effort between thethree, I think of it almost, you
know how you have those min-maxabsolute points on curves, where
if you had a two-dimensionalgraph and there's like a bell
curve and you want to kind of bein the optimal value with a peak
somewhere around there.
But because it's three points,like time, uh sorry, testing,
code, and planning, it'sactually in like a
(28:27):
three-dimensional space betweenthose three, and like where do
you put most of your effort?
But I do think that that wholequestion will go away as models
get bigger, better and bigger.
Um, there's still so much to bedone around memory and context,
which you're hinting on.
I think that's really a big gapright now that companies will
(28:48):
most likely start having tofocus on a lot more in the
coming year or two.
Um, but in terms of theplanning, when you're doing it,
so the way we build plans isfirst we write up a uh either a
product design requirementdocument or an ADR, an
architecture design requirementdocument.
(29:09):
Depending on the refactor or thecode that we're writing, that
could be somewhere around two tosix thousand lines of planning
that is created.
That's a whole process.
This is where you're putting alot of time in, at least for us,
is the planning.
Our planning time is gone, it'slike way higher than it used to
be, right?
We used to just like write awrite some specs, figure out
some stuff, and start coding,right?
Now we're spending about doublethe time than we used to
(29:31):
historically on all of thisplanning stuff.
And so you so we do that firstplanning up front, then we take
that document.
We're not done with planning,because then we take each phase
or each section of it, and eachof those is then broken down.
Uh, one of the things we tellthe AI is write it as if a
junior was gonna write thiscode, right?
Like how how give the juniorenough information that they
(29:53):
know exactly what they need todo for each of these steps.
There's more to it too, becauseyou really have to think about
edge cases and how you're gonnatest things and validate, and
how are you gonna get the AI todo all of this work without a
human needing to be involvedwith these steps?
Sometimes you have to build abridge.
There's like mini projects thathave to happen to just build
those bridges or thoseself-validation loops.
(30:15):
So you build all of that out,and then from each of those
documents, then you build uhwork order collections, which
are breakdowns of uh contextthat is a living document, plus
a planning, which is a livingdocument, and then actual
individual tasks, which is alsoliving documents too.
Living, meaning that they'reedited as they're being used,
(30:37):
they're not just static.
And then those collections, andthere are different frameworks,
but this is kind of the mostsimple one.
Those are then passed on finallyto a team of agents that then
has their own um state-drivencontrol system that I was
talking about to make sure thatthings are grounded so that and
they have antagonizingobjectives, is also important
(30:59):
too.
Like you want the tester whosegoal is to find as many bugs as
possible, and they get pats onthe back or best fit points, or
however you want to look at it.
And then the developer wants, ofcourse, to code successfully,
and then the task manager wantsto make sure that everyone's
proving that they're doing theirjob so they can mark them all
off.
So every single role in it hastheir own objectives, and some
(31:20):
are antagonizing.
And then they go through all ofthose, self-check and report,
and then you just need to, andthen from there, you as you
sophistic make the system moreand more sophisticated and
resilient, you can accomplishlarger chunks of work at a time
before a human needs to even goand start reviewing it.
So it might run for five minutesin the background when you're
building confidence, but theneventually it's running for like
(31:41):
30 minutes or more in thebackground as it just crunches
through these differentdocuments in series.
Um now, planning one-offplanning mode in claw code, that
I would say I kind of view it aslike it one step above AI
assistance, because what you'redoing is you're basically doing
(32:02):
a one-shot job in in air quoteson a particular function or
feature or mini refactor.
But there's a limit to the scopeand scale that that can
accomplish on its own.
But it is definitely way betterthan just a one-shot prompt in
it.
But even the plan mode needssupporting memory and skills and
(32:23):
things like that in order tomake sure that the plan is
grounded in truth and doesn'tjust have assumptions,
surface-level assumptions in it.
That's that's kind of the devila lot in the detail with a lot
of the plan mode, is that AI hasis sitting on top of a very
large code base, reads threefiles, but doesn't read the
other 10 it needed to read, doesa plan based on assumptions and
(32:46):
what the other files does, andall of a sudden it's completely
gonna just screw you over whenit goes through.
You know, we've seen seen thoseproblems happen.
So, you know, plan mode can beamazing, but it can also be
insufficient for what you'redoing.
SPEAKER_03 (33:00):
No, oh yeah.
No, I I I think it was kind oflike a bare minimum.
Like if you're gonna like just aone-shot prompt, like if you if
it's not really elegantly uhengineered, uh it's not gonna
give you the output you need.
Because there's a lot ofconversations, I guess, on
social media of like all thesetools that are just kind of
wrappers anyway, and it's likebuild me a whole app, and then
(33:23):
it like breaks.
And it's like you can't build awhole app with one line of code
that and then and then you gotto think about well, how many
users are gonna go through thisapp at the same time?
Because you know, if you if youscale it up, you got one one
function processing at a time,like it's not gonna be super
useful.
(33:43):
And so that actually, you know,I know I'm pivoting all around.
I have so many questions foryou, but that's been some of the
debate with a couple clientprojects that I got pulled in
on.
Is okay, like we want to vibecode something, um, and it's
functional, but there's somesecurity concerns, what kind of
data is gonna go through it, howmany users it's gonna be, and
(34:05):
then like, okay, can like base44 or so some of these tools um
uh replaced an enterprisesystem, right?
And it's like, no, use it tocreate the MVP and then at a
certain point switch it over toa developer.
And I'm starting to see withwith these agent systems, it's
(34:27):
like, well, they can code it up,like, you know, like like okay,
maybe you're not using thistool, but these agents, if if,
if they have sufficientinformation, and I get it goes
back to the the memory.
I feel like the memory is thatthat big issue right now, but I
mean they can code it inwhatever you want to code it in.
Like you can build a plugin, youcould, you know, build it with a
(34:49):
similar back end.
Like if you're building aWordPress site, you could build
a similar looking WordPressbackend, or hey, let's just use
um the the the command line orthe terminal to do all the work
and let's not ever touch thesite.
Like, like there's a lot ofthese discussions of like, I
feel like it's when um likedigital cameras switched over to
(35:10):
like a new technology, and it'slike like you you gotta make
like a hard switch at some pointto create different processes
and do things differentlybecause a lot of people are
augmenting the old way, butthere's like a new way of doing
things.
And like, where is that switchover point and also the maturity
level of of what these thingsare doing?
(35:32):
I, you know, like do you have tohand it off to a full stack
developer to build it out?
Or, you know, can these agentsbuild it where it's enterprise
ready?
Like, what's happening withSaaS?
I there's a lot of questions inthere, so feel like to answer
whatever one you want.
SPEAKER_01 (35:50):
Sure.
So what I would say is, I mean,talking about that last thing
you were just mentioning aroundthe full stack developer.
Yeah, the way we've seen so wemade a rule for ourselves first
experimentally back in December,which was can we do this new
project never writing a line ofcode?
And it was a pretty it was ahydraulic system simulation code
(36:14):
refactor, you know, a meatything, not just like something
you could just spin out as afive code thing in it in an
afternoon.
And back then it was alsopre-Opus 4.6.
I think it was 4.5 at the time.
There's something magical thathappened with 4.6, but that's
comes later in my little story.
But man, it was like the firstmonth was like painful because
(36:37):
we were just we had this ruleand we're trying to go through
it, you know, regardless of howto learn how can we get this
thing to to be as good as usdoing it by hand, but not right.
And so, but now with the waythings have progressed and and
all the things that we'velearned, and then also 4.6,
which did like this magical kindof frosting on the top and like
(37:00):
really helped brought it evenmore to reality.
I at least for us, like it'skind of over the whole idea that
you write code by hand, youknow.
Of course, if the client comesover and it's just like, I need
to change something quickly, Imean, maybe you just change the
line of code, or maybe you justone-shot that one little thing.
But any kind of major refactoror major project, it's we're
(37:22):
seeing that it's done way fasterand has less problems than when
people are doing it.
By less problems, I mean lessbugs, you know.
So, I mean, that's kind of a bigkind of milestone.
Is that and then by and lessbugs can compare to like let's
say the average programmer,right?
I mean, of course, there's gonnabe senior coders out there who
are gonna still beat these AIsystems today.
(37:43):
I don't know how long that'sgonna last, honestly, like
probably a year from now.
That might not even be trueanymore.
But at least right now, youknow, certain senior devs can
probably outcompete the AIstill, but probably not on
speed, but more on you know,bugs and broader thinking of
impacts and things like that.
So there's there's still some alittle bit of a moat there.
But um, but yeah, in terms ofenterprise ready software and so
(38:08):
forth, you do need somebody whois that brain who really is
thinking through, you know, whatare the security levels?
How are we gonna test this?
How is it gonna integrate?
When we're done with featureone, what's coming next?
How do we not paint ourselvesinto a corner?
All this kind of like broaderlevel thinking, the AI is not
capable of doing that.
AI is terrible at askingtangential questions.
(38:28):
You know, it's like the AI willtake a task and it will converge
on what is the most likelyoutput or answer for that
particular problem.
In fact, people get reallypissed off if the AI comes back
at them and throws somethingthat they weren't wanting to do.
You know, uh I don't this isjust popping in my head, but I
remember last year, this was awhile ago, but someone posted on
(38:49):
social media that they wereasking the AI to write some code
for them.
And instead of writing the code,the AI responded back, telling
them that they think the AI saidto them, I think instead of me
writing the code, you shouldreally go off and read these
things to become a betterprogrammer, probably because
what they were asking them to dowas just like flat out wrong.
And the person was so pissedoff, and everyone responded to
(39:12):
that social media post saying,like, oh my God, this is the end
of humanity and AI is revoltingagainst us and stuff like that.
But it's like, you know, likepeople don't even want these
systems right now to bethinking, you know, there's this
risk that we don't even wantthem to not obey us, right?
To say to us, like, hey, whatyou're doing is probably you
haven't thought of this thing.
Like, are you sure we should bedoing this?
Like, people would be likelosing their mind if it does
(39:34):
that.
No, it should be.
SPEAKER_03 (39:36):
Well, I think that that's you know, you go back to
like different um of thesedifferent models and like how
they're how they're buildingthem and what they're trying to
achieve.
Because I feel like Claude ispushy, like it's pushy.
It's like, and I'm like, hey,you don't know all the facts,
like let me give you some morefacts, especially kind of more
(39:56):
before we switched over uh tokind of the agentic era.
But but I'll tell you, chat GBT,unless you turn off like nice
guy mode or you, you know, youhave a devil's advocate going
on, it's just gonna tell youwhat it wants.
Like it's and that I think thatthat's the approach.
And people started using chatGBT and they're like, Oh, I'm
(40:16):
always right and I'm amazing andall this kind of stuff.
And then you switch over toClaude and Claude, like, you
know, or Grok will tell you likewhat it is, and then you're
like, Whoa, like I didn't expectyou to say that, right?
SPEAKER_01 (40:28):
You know, and et cetera, yeah.
I mean, it you can take theanalogy also with people when
they hire people, right?
If I hire someone for 15 bucksan hour, 20 bucks an hour,
whatever, and I tell them whatto do.
Most people don't want thatperson saying to you, like, are
you sure this is the job Ishould be doing, right?
Like they just you just want theperson to do the job.
Then you pay someone$200 an hourand you tell them what should be
(40:50):
done, and they come back andsay, actually, you should be
thinking about these things,right?
All of a sudden you really wantto hear that.
And if that person doesn't pushback, you think that that person
doesn't know what the hellthey're talking about, thinking,
you know.
So there this also, I think, istranslating across models and
different price points.
And the only this is gonna thisis a bit controversial, but I
think you could totally takegiving an idea here to OpenAI,
(41:12):
you could totally take yourmodel, the the mass model, and
just only remove one trainingthing, which is that when you're
training the model, don't makeit please the user, but instead
be like more contrarian and andpushing back more, and then make
people pay more money for that,and they will take that model
more seriously because they'repaying more money for a model to
(41:34):
actually um you know push themrather than just please them.
But you know, but most peoplewant cheap and free things to
please them, they don't wantthem to contradict them.
So there's this psychology inthat as well.
SPEAKER_03 (41:46):
No, it's it's super interesting.
Okay, so I we have a few moreminutes.
I would love to talk a littlebit more about kind of how you
think about orchestration,because like a lot of how I was
thinking about orchestrationbefore was like, okay, one agent
specializes in one skill.
And now you can start to add alot of like skills to one agent
(42:08):
and make it like a super agent,right?
But there still is differentroles that you want to define uh
for it to do.
But how I was thinking aboutorchestration before, and even
like some of the books that Ihad bought and read, like one
third of the book is maybeuseful now, and two thirds is
like we've already passed likeeverything it was talking about.
(42:30):
And so where on kind of frontierstuff, when you're setting up uh
these systems, how are youthinking about orchestration
today?
SPEAKER_01 (42:40):
Yeah, so there's two parts of orchestration.
There's the human orchestrationpart of it, and then there's the
agent-to-agent orchestration.
You're talking about more thelayer, right?
SPEAKER_03 (42:48):
Yeah, agent to agent, yeah.
SPEAKER_01 (42:50):
Yeah, so I would say two things.
First is the architecture of it,and the second is the solution
problem-driven part of it.
So if you're gettinginconsistent results from a
process, the process you mightbe putting too much agency on
one agent, and you need to splitthat up into sub-agents or
sub-steps, and you might need toformalize them to become more of
(43:12):
a linear process rather than aqualitative decision matrix,
right?
So, one problem I see often isone agent will be given too much
too much space, too many skills,too many things it could do, and
you're asking the agent to thenmake broad decisions instead of
making it more controlled.
(43:33):
And so, if you can, one quicksolution to that would be you've
got an agent which first does adecision on what type of request
is coming in.
Is it a, you know, uhsecretarial request?
Is it a booking request?
Is it a research request from myhuman, whatever it might be, or
from the clock or from someother info, whatever it is?
(43:54):
And then it dispatches that toanother agent, which then
specializes on just that.
And then Um that that state uhscript driven element to it is
also important there.
Splitting it up can also helpyou with that memory management
context.
I know we probably we may nothave time to get into that.
I know that was one of yourquestions, but memory management
(44:15):
is, of course, really importantright now.
Context management, contextfirewalling as well between
agents.
SPEAKER_03 (44:21):
And then you talked a little bit about the bridge,
like building like littlebridges.
Like, can you give an example ofthat maybe?
SPEAKER_01 (44:29):
So if I'm I hope I'm talking when I talk about stop
me if I'm going down the wrongpath here with the bridge,
you're talking going over thewrong bridge for what you're
talking about.
But I would are you talkingabout like the, for example,
that SEOGO agent who comes upwith a brief and then the
connection to the next agent.
Yeah, yeah.
SPEAKER_03 (44:43):
Like you have to build a little of the brief to
do the bigger project, likegiving people an example of what
that is, you you can start toseparate out what's being done,
and you're connecting like maybetwo systems together, if you
will.
SPEAKER_01 (44:56):
Yeah.
So going back to my analogy ofthinking about jobs.
So if you think about a job andthe domain of what that job is
and the knowledge that's neededfor that, that gives you a
contained space, that job maysometimes need to be split up if
you're putting too many jobsinto it.
You know, I sometimes I'll lookat a job description for a
human, and then when I look atit for an AI, it's actually
three jobs because it makes moresense that way.
They're just trying to hire somekind of unicorn to do these
(45:18):
multiple roles all together.
And then, but then just like inin traditional systems, the some
of the most important parts isthe connectors, the edges
between systems, those bridges,those gaps, or APIs, or whatever
you want to call about it.
And so you need to formalize theoutput to the input of the next
system and how you're going tocontrol for handover and errors
(45:41):
and expectations and checksbetween these systems.
One sec, I have to clear mythroat for a second.
Okay, sorry about that.
And my and my coffee cup'sempty, so I couldn't just liquid
liquidate my throat there.
Liquidate, liquefy, I don'tknow, whatever the word is,
lubricate.
That's what I'm looking for.
(46:02):
Um anyway, so you have you haveto look at that bridge point,
and then there's different waysyou could do that.
You could use an inbox systemwhere you just talk to each
other, you could use afolder-based system where they
scan those folders, you coulduse uh more like a spreadsheet,
right?
Like you drop it on aspreadsheet and it could be a
database.
I don't know, it depends on howsophisticated it needs to be,
how much you need to scale it.
SPEAKER_03 (46:23):
Because I I I was like building stuff that I was
having perplexity create, and Iwanted it to talk to Claud Code.
And the best way I could figureout how to do it is dump
everything into a spreadsheet,and then Claude Code recalled
it.
Like I was trying to get them totalk to each other, uh, but I
just had to like drop everythingum in a document and then you
(46:45):
know have it uh when you saycomplexity, it's just you using
perplexity browser and so comet,uh, you can spin up multiple
agents at a time to do likedifferent kinds of research.
SPEAKER_01 (46:57):
Okay, yeah, yeah, yeah.
SPEAKER_03 (46:58):
And so so, but I was trying to get them to talk to
each other, and I I was justdropping in a spreadsheet and
then recalling when it wasupdated so it would check it,
and if new information waschecked, it would pull it.
Uh, it's kind of how I got itset up, but it's pretty rude for
it.
SPEAKER_01 (47:13):
When we're using yeah, when we're using
perplexity, because it's greatfor certain, it's really good
for certain things.
We're using it as an API toolfor an agent that itself has
access to databases or memory orwhatever.
So we don't need to that wholekind of what you're talking
about, we don't end up having todeal with it because agent, I
don't know, Bob.
That's called agent Bob.
(47:34):
But Bob has the API perplexity,does research with there, gets
data back, stores it either in aJSON file or a table or
superbase or whatever kind ofneeded.
And then the next agent can justpull straight from that system.
So it's just all within the sameecosystem.
Yeah, not having to translateacross a third medium.
Got it.
SPEAKER_03 (47:53):
Yeah, yeah, awesome.
So I I I do have a hard stop onthe hour.
I would love to kind of I know II jumped around.
I'll I'll I'll definitely haveto have you back.
I I really enjoyed ourconversation.
I know we were talking somephilosophical stuff.
Um, yeah, no, you try to talk tosomebody on the street about
this.
Sometimes uh people are like,what are you talking about?
(48:15):
Oh, totally.
Um, no, maybe uh is thereanything in the course of this
conversation that you thoughtmight be really valuable to just
add to to this conversationhere?
And then if you could share alittle bit more about what uh
Fountain City is doing and howto get in touch with you.
Uh, you talked about some blogsyou're writing, uh, and that
(48:37):
would be phenomenal.
SPEAKER_01 (48:40):
Things that weren't discussed.
I mean, there's so many thingsthat could be talked about, you
know, the security and how areyou handling that?
SPEAKER_03 (48:46):
There's scalability, there's don't drop your don't
drop your API keys right in theprop, guys.
Like, don't do that because it'sgonna get processed.
Like there, you gotta you gottaset up a file or something like
that where it pulls from.
SPEAKER_01 (48:58):
So yeah.
Yeah.
So there's there's a lot to talkabout there.
Um I I yeah, I think everyeverything I can think of is
just rabbit holes.
So I'm I won't bring up like toomany things that haven't been
talked about.
But in terms of me, us, fountaincity, so fountaincity.tech is
the website, and our clientsright now are actually a lot of
(49:20):
agencies are hiring us to buildeither systems for themselves
and or for their clients.
Sometimes uh one person alsowe're talking to right now is to
build a vertical stack system.
So it's they have lots of reallytiny clients, like 80 clients in
one vertical.
Uh, I won't name it, but youknow, imagine it could be like,
(49:42):
I don't know, plumbing orservices, yeah.
SPEAKER_02 (49:45):
So we've got a service.
SPEAKER_01 (49:45):
So some kind of services industry or some kind
of industry.
And so then you you get thiseconomy scale because you can
have one research pipeline, andthen it generates a whole bunch
of topics, which then getinterpreted and resegmented for
different writers for each um ofthe different subdomains.
So we're doing a lot, yeah, likeI said, with agencies, direct to
(50:05):
client, and then these these uhintegrated verticals or
multi-site hosted stuff.
So that's kind of one part ofour business.
And then the other part is doingwhat I described with the agenda
coding, because we're able to,and then all those projects tend
to be building systems that haveAI in it.
So, like a uh right now we'vegot two biggest projects we're
running is one is a voice systemwhere it listens into enterprise
conversations and then givesvery and then listens into these
(50:29):
enterprise leaders speaking andthen gives very thoughtful input
to kind of help them to thinkbroader or think deeper or think
in a new way about the problemsthey're bringing up.
It's a very interesting project.
And then the other one we'redoing right now is on a data
intelligence layer.
So we're connecting together.
People come to the website fromall different possible paths,
(50:51):
everything they interact on thewebsite, and then everything
they've purchased or thinkingabout as they touch all these
different systems, and thenbuilding that out into on-demand
charts, uh, being able topredict, you know, who should
they go talk to first, and thenbeing able to AI talk to the
data and stuff like that.
So that's another cool kind ofproject we're doing.
SPEAKER_03 (51:10):
So I'm using a tool called uh E-R A F A N A.
Are you familiar?
I don't even know how to saythat.
SPEAKER_01 (51:18):
Can you pronounce that?
Sorry, Grafna?
Is that what you mean?
SPEAKER_03 (51:21):
Grafna, yeah, Grafna.
Um, basically to to to take likethat data intelligence layer
that you're talking about, yeah,to throw it into these kind of
dashboards.
You can look it up.
You would I don't know whatyou're using.
I would love to have a toolsconversation with you.
SPEAKER_01 (51:36):
Um we're using grist on that one, okay, which is
another G letter.
Yeah.
It's not grapna, it's grist.
It's close.
I was almost wondering if you'retalking about that.
But yeah, we are using thatright now as kind of a it's it's
kind of like a it's like aspreadsheet in a database had a
baby kind of thing.
(51:57):
So it's a bit like a spread, orit's like a almost like a uh
what was that thing called?
There was the spreadsheet thingthat was popular with project
managers a few years ago, whereit's like super spreadsheets.
So it's kind of like thatbecause they this client really
likes spreadsheets, so we'regiving them that as like a
low-cost kind of middleware toall of the data collection, and
then from there, because it'sessentially database, we can
(52:19):
connect it to anything we want,visualization, AI model context
and stuff like that.
SPEAKER_03 (52:23):
Yeah, this would be like the next generation of like
BI or looker or something likethat that's more interactive and
you can pull in more data.
No, I uh Sebastian, thisconversation's been been
awesome.
So, where are you putting outnew new information?
You talked about your blog.
Is there like do you havesocials stuff?
SPEAKER_01 (52:43):
Yeah, so LinkedIn, my account's probably the best.
Uh so just my name on LinkedInand um YouTube.
I think it's been a bit sloweron YouTube recently, but I I it
used to be once a week, now it'skind of like once a month.
But I'll probably pick it backup a little bit more.
I just have some speaking eventscoming up, so I've been and this
(53:04):
launch of this kind of virtualdigital agency product has been
very time consuming.
So I've been more focused onthat recently, this kind of
autonomous agent system.
But anyway, YouTube, LinkedIn,the blog on our website is
almost daily blog posts oneverything related to this, like
you know, from security to uhimplementate implement
(53:25):
implementation.
I'm stumbling over my words, andthen um what else?
There is a newsletter also onFountain City.
It's pretty, it's not very oftenthat we newsletter out, so you
won't get overloaded, but youcan sign up for that too on our
website.
SPEAKER_03 (53:38):
Awesome.
Well, we'll definitely have tohave you back on.
Um, I I have so many morequestions for you, and and uh I
like all the stuff you're doing.
If anybody's listening, this iskind of where where it's all
going.
Um, and so you need to I I thinkreally understand the um taste.
Is that what we said?
Is that what we said on like thetaste of kind of where you what
(54:03):
is your secret sauce?
Yeah, like why people shouldhire you.
What are what are you doing?
I think that that pin pendulumis going to kind of swing back
because a lot of these thingsare gonna be compressed and um
you know, uh programmatic isjust, you know, like the world's
changing.
Um and so uh it's hard to stayon top of all this stuff.
(54:27):
Um, Sebastian, it's great tohave you on.
Um, thank you for being here,guys.
Go check out what he's doing.
Um, I think uh, you know, thisis the future.
Um until the next time, if youwant to grow your business with
the largest, most powerful toolon the planet, which I would
have said is the internet, butyou know, I would say it's now
(54:49):
agentic agents.
Uh keep listening to thispodcast.
We'll keep bringing on um someof the people in the industry
that are moving it forward.
Uh, thank you so much for yourtime.
Uh until next time.
My name is Bat Bertram.
Bye bye for now.
This is the unknown secret of internet marketing.
You're inside a guide to thestrategy of top marketing union
to craft the competition, readyto unlock your business full
potential.
Let's get started.
SPEAKER_03 (00:18):
Howdy, welcome to another fun-filled episode of
the unknown secrets of internetmarketing.
I'm your host, Matt Bertram.
Uh, I would tell you, mycomputer might be uh you know
maxing out right now.
I got uh too many thingsrunning.
We got some uh clawed codegoing, we got uh a couple
different agents set up, and soI probably need to close some
(00:38):
browser windows and turn somestuff off.
So uh if there's a little lag,guys, apologies.
Let me try to shut some of thisstuff down.
But as the conversation hasmoved more and more towards the
agentic realm, and we're talkingabout LLM visibility um and just
kind of how to think about allthese things.
I thought I would bring somebodyon that is deep in uh AI agents,
(01:02):
uh setups, audits, uh setting upthat AI agent to help you uh
help with your workflow.
Uh Sebastian, welcome to theshow.
Thank you so much.
SPEAKER_01 (01:13):
Pleasure to be here.
SPEAKER_03 (01:14):
And Sebastian's with uh fountaincity.tech.
Uh you can go check it out anduh I'll let you talk a little
bit more about it at the end.
But you're doing everything AI,Sebastian.
So I thought I'd bring you on.
SPEAKER_01 (01:26):
Yeah, thank you so much.
There's a lot to talk abouthere.
So where would you like tostart?
SPEAKER_03 (01:32):
Well, yeah, the I mean the market's moving really,
really quickly.
I I felt like open claw was uh alittle bit of a distraction.
Um we were setting it up, but itwas having some different kinds
of problems, and we've justpivoted back to quad code.
Um and uh just kind of now theyleaked the harnesses.
So like I think everybody'sgonna have a pretty good
(01:53):
harness.
I know um, you know, whatFacebook's doing is really
interesting.
Like we've been using Mantis alittle bit because you got the
you got the rag recall, quoteunquote, from all the Facebook
ads.
Uh, so that's helpful.
Uh, but it would be good tomaybe talk about how you like a
mental map on how you look atall these different like
(02:14):
frontier models and how you'remaybe using them together.
And like even like uhperplexity, I've been playing
around with uh Comet andComputer, and it's really
agnostic, right?
So it'll just use whatever'sbest to kind of build it.
It's like a yeah, Po 2.0.
Uh, if we're moving from uh thechat bot side to really the the
(02:36):
agent side of things, I wouldlove to just kind of how would
you paint the picture to someonethat is around this, but maybe
is not as deep in it as you are.
SPEAKER_01 (02:47):
Okay, yes.
I think what's coming into myhead as you're talking about all
those different things is firstI would classify the different
usage into tiering as kind of away to kind of focus then in
later on certain aspects of it.
So AI assistance is kind of tierone where you're just using AI
to help you along with whateveryou're doing.
And sometimes that might evenlook at how people are using
(03:10):
cloud code, but certainlyperplexity or a comet browser,
you know, you're and then thenext level up, uh, tier two,
you're using AI to performcertain automations that could
be on a clock, that could betriggered.
And the workflow automationitself might only have AI as one
of those steps in the process,but there's AI, you know,
somewhere in there, otherwiseit's just a normal workflow
(03:32):
automation.
Tier three is where I startthinking about the AI
agentically, meaning it hasmemory tools and agency, meaning
it can take its own actionseither on a schedule or
triggered as well.
And then if you want to gobeyond that, there is levels
above that, but already at thatthird level, you have within
that two different categoriesalready as well, A and B.
(03:54):
A would be it's driving mycomputer, right?
Which is more like co-work orperplexity.
I forget the name of it, whenit's driving your computer
computes, I think is that whatit's called from uh perplexity
computer.
SPEAKER_03 (04:07):
Um, and then like claud co-work is the what's
driving uh is is the shared kindof uh online component.
And then uh claude uh code iswhat at least how I have it set
up, is driving mainly mycomputer.
And so we're starting to putthings in um like uh Apple Minis
(04:27):
where we're we're playing aroundwith like ChemKey 2.5, uh and
and we're we're because thetoken usage, unless you really
manage it, like putting like aslash context or something in
front of it, it will just burnthrough the tokens so fast.
So, like tokage optimization hasbeen the biggest area that we've
(04:48):
been trying to tinker withbecause it, you know, you get it
to do a lot of actions that getexpensive.
SPEAKER_01 (04:53):
Oh, totally, yeah.
And so that that second categoryis more the always-on category
that could be, like you said,like a Mac mini in your house or
something like that.
We, I mean, we're we're buildingthese systems for clients, so
that's not really an option.
So we put them up in the cloud,so it's like AWS EC2 servers or
stuff like that, you know, orGoogle, you know, whatever
(05:14):
flavor of system the client isin, we'll just embed it into
that ecosystem.
Uh within, but yeah, and well,you just talked about cost
management.
I think there's different sotalking about open claw or these
kind of systems that haveagentic capabilities into it.
There's so many different waysto build that system.
(05:35):
I think the first way peoplethink of it is, you know, maybe
they're experimenting with it,they do conversations with the
system, give it some abstractgoal.
The AI goes off and tries toaccomplish that abstract goal,
burns, you know,$50 a day inyour wallet of token usage, and
you're like, oh my god, I don'teven know if I'm getting value
out of this, you know.
So we actually do so we've beenbuilding a lot of these
(05:57):
autonomous workers or autonomousagents.
And there's uh we like to thinkof the agents kind of
categorized also in terms ofthose different tierings that I
just mentioned.
So we might have one agent thatreally has a very specific
workflow.
SPEAKER_03 (06:13):
Are you what are you using for orchestration?
Are you using like Langchain orsomething like that?
Or like I'm trying to understandwhat the tech stack is because
I'm starting to use quad codeand hooks and skills and like um
it's spinning up other agentsunderneath, like whatever it's
trying to do.
I I think I'm probably teeteringbetween like uh if you framed it
up, between two and three,because we're just setting um
(06:36):
triggers like daily or likeautonomous, like okay, go go run
the script.
And then there might be aagentic layer on decision making
at some point.
But like, can you back up andbreak down like what is the tech
stack that you're building on?
And like what are you settingup, or maybe even give an
example of something you'retrying to solve for a client?
SPEAKER_01 (06:58):
For sure.
Yeah, I will add one pre-step towhat I'm about.
I'm gonna answer your question.
I'm gonna say one thing first,which is I think the most
important first step is tofigure out what you're trying to
do and describe that.
I like to think of the coming upwith a job description of what
what is gonna be built.
And then within that, reallythink about the input.
What is the context?
(07:19):
What's gonna be the input foreach job?
What is the overall generalinput, like you know, first day
training on the job, what youneed to know.
Then what is the process?
How does the sausage get made inthe factory?
That's a really important step.
That's where your quality comesin and not just volume of slop.
And then what's the outputformat, right?
And so once you have thosethree, then you can start
thinking about the architectureand how you're going to
(07:41):
accomplish that goal and whattier level you need and
combination and so forth.
So, with that in mind, let's saythat.
So uh so we have one agent, forexample, that does SCOGO
research and builds content thatfeeds into a content pipeline
with ideas and so forth.
And so within that jobdescription, there is a whole
(08:01):
need to understand the contextof the business at heart, but
then also who are thecompetitors, uh, what platforms
do we want it to research everyweek.
So we have it researching allposts from all competitors, plus
Reddit, plus X, plus Substack,probably a few, and then just a
general uh mentions searchacross the whole web of who's
(08:21):
talking about what.
All of that's compiled alongwith you know documents that
describe the objective, and thenall the Google Analytics is
read, the Google Search Console,keyword performance, AI ranking
performance is also tracked,like all of that's input, and
then from that, briefs aregenerated according to a format
of how we want the brief tolook.
(08:42):
Those have actually havestrategies as well of what kinds
of briefs, and then that's theultimate output at the end of
that agent.
Is like, okay, here's the thingsthat should be written.
And we do also score them interms of um I forget the name of
it, but it's a scoring that weuse to determine the tier one is
that it's very novel, uniquecontent, right?
So it's it's it's authoritative.
(09:05):
Tier two is the uh the researchis going to be novel, so no
one's researched this before.
Tier three is the framing isnovel.
So the research and the contentisn't, but the way we're framing
it is, and then tier four iswhat we call commodity content,
and then depending on theclient, that type of content may
never get go further, right?
That's like a barrier, likebecause that client doesn't want
(09:27):
commodity content.
Commodity meaning if itdisappeared from the web, no one
would ever blink an eye, right?
That has no intrinsic.
SPEAKER_03 (09:34):
And so you're you're using you're using like uh APIs,
MPCs.
There's kind of some new stuffthat's coming out, as well as
maybe some like uh uh you know,a data lake or even like a
rudimentary Google Drive.
SPEAKER_01 (09:50):
And so yeah, I can go to the technical stack next,
but I think it's important toframe like what's the job first,
because from that, then you canstart moving more and more into
like okay, the nitty-gritty.
So in order to achieve that,there has to be a memory
management system.
So that could be SQL database orobsidian QMD.
Uh really it doesn't have to beRAG necessarily, depending on
(10:12):
the volume, could be justrelational.
And so there's that piece of it.
We're building it on an EC2,like I said.
You need to have a clock thatgenerates the reoccurrence.
That could be a prompt job,could be open claw, you know,
could I mean we're um uh so fromfrom that clock then that's
creating the occurrence, you'vegot the memory that, then you
(10:33):
have the skills that aredeveloped in order to create
consistency of behavior.
And then we uh there's a blogpost on my website that talks
about four things to really lookat when you're building, which
is if I remember them all offthe front of my head, uh ground
it.
So you want everything to bestate-driven.
You don't want the AI model tojust kind of remember what the
state of something is and trackit in a markdown file, right?
(10:55):
You want to put it either in aJSON or table or something for
state-driven decisions.
You want to script it as much aspossible.
So you really don't want the AIto be thinking through how to do
things over and over again.
Even with a skill, I it's stillnot um, you're still gonna
create error rates because AIsare probability machines, right?
So you want to create scriptsthat the AI actually triggers.
(11:18):
So a skill triggers a script,the script is a set amount of
code, the code runs, it's not AIanymore, and it's also cheaper,
runs an operation, stores statesinto your data store, and the AI
is just taking data back andthen doing the only the step,
the only step that AI should bedoing is the interpretation.
Like, how do I interpret thisinformation?
(11:40):
How do I make a qualitativedecision in this case based on
this data with this context?
You do have to manage alsocontext because like the more
skills you have, the less inputtokens you have.
So that's like an interestingbalance right now in the
ecosystem.
Um but then so you so that's uhthat's the memory agency, and
(12:02):
then the skills and tools.
That's really the maincomponents of the system, right?
And then from there, everythingelse is just driven by the input
output layering.
So for us, those briefs are aninput to another agent, which
then takes it further.
And the ultimate final output isthat it just publishes it on
whatever website or social mediaplatform is connected to the
system or newsletter, whatever.
So that's the ultimate output.
(12:24):
Uh, and then the input layerthat can be that research I
talked about, but it could alsobe transcripts or video
recordings, or it could be evenother AI agents.
So we've coupled also AI fullyagentic systems that build
software, and then those agentsnot only build the software but
write case studies about whatthey did, the problems they
found, how they solved it, andso forth.
(12:44):
All of those lessons learned getfed into the system that create
novel content that is then tierone for novel content, right?
Because like it's act it's notjust research anymore.
It's not, and so you've got thatyou can actually have highly
novel content going through yoursystem end-to-end, which is like
you don't even need a personanymore to do some of that
(13:05):
novelty in the content,especially if it's like you
know, research driven or thingslike that.
SPEAKER_03 (13:09):
Yeah, that that is actually something that we were
talking about earlier today.
Uh, everybody listening, it'sMonday.
And so we were we were talkingabout how to uh pull some of
that in.
Uh and we haven't quite figuredthat piece out yet.
But that would be great to uhtalk to you more about that,
because that's actually whatwe're looking at is like, okay,
the data and the data pointswe're producing, how how do we
(13:31):
turn that into case studies?
Because that's um unique data,right?
Um, it's it's data that we haveaccess to.
And, you know, so that's very,very cool.
Very cool.
Well, that's I think that that'swhat you talked about is where
the market's going.
And I think that uh a lot ofagency owners are trying to kind
(13:52):
of figure out that and are atdifferent places and getting
that done.
Where do you see, like if youtried to project out, because I
know it's hard because thingskeep changing, um, where do you
think everything's ultimatelygoing at at the end of this?
I I've been listening to somedifferent people that's in time
or in 10 years, like what whattime?
Yeah, well, uh I you know,everything's moving pretty
(14:14):
quickly.
Like, let's say like threeyears, like where do you see
everything converging in inthree years?
And like, where's thatsuppression compression
happening in your mind?
And what is the world gonna looklike?
SPEAKER_01 (14:27):
So the compression really is between idea and
delivery.
And another way to look at itwould be problem solution or
opportunity and action.
Whatever you want to frame, youknow, the beginning and the end
point, everything in between iscompressing into a slice that's
gonna get smaller and thinnerand thinner.
(14:48):
There's certain areas right nowalso that people think are
relatively protected, domain,you know, domains that humans
have an advantage over AI.
And uh, I I think though a lotof those are just rugs that
we're standing on that are gonnaget pulled out from underneath
us over and like subsequentlyone after the other.
It's a very similar trend thatwe've seen with computers.
You know, if you look at, youknow, first we thought it could
(15:09):
never beat a human in chess.
That happened with what was hisname, Kasparov.
Then we thought it would neverbeat Go because the domain
spaces were too big, and thenthat happened a few years later.
Then we thought the Turing testwas unbeatable, that was beaten
in 24, right?
So it's like we keep puttinglines there and we think, oh,
it'll never pass that.
I think the really true so Ithink humans really have four
(15:31):
main areas that are kind of themain areas to be at play with
the with the direction thingsare moving, and these the four
areas really to be thinkingabout for how humans are
involved.
The first is direction.
That could be direction of thewhole business, but it could
also be direction within yourdomain, right?
Within marketing or withindesign.
Some people call it taste, butyou know, it's it's ultimately
(15:52):
like what is the differencebetween option A and B?
Like, I guess even moreabstractly, you know, if my kid
came to me and said, Should I bean artist or a doctor?
And it's like, well, what'simportant to you?
You know, like these are soabstract choices that someone
ultimately still needs to makeuh call it a taste decision or
direction within that thing,yeah.
SPEAKER_03 (16:11):
So taste, trust, governance, um, and then there's
like one or two others that uhI've started.
SPEAKER_01 (16:19):
Yeah, trust I would connect to the second quality,
which is the interactions layer.
Okay, people want to interactwith people.
And so whether that's service orsales or support or teaching or
understanding or empathy andtrust and all these kind of
things are kind of on theinteraction layer.
And so that's not going away,you know, that insofar that
(16:43):
people are actually good atthat, right?
Certain types of interactionlayers will eventually not like
people won't even want it, youknow.
The time will come where you goon and you do a phone call for
support, and if you get aperson, you're gonna think, Oh
god, not a person.
I wish it wasn't the guy I wastalking to, because it's gonna
be so much better.
Like, why is this company stillhave a person doing this job,
you know?
So, because you're you know, forthings that are more like
(17:04):
transactional or findinginformation or getting an
answer, right?
Those kind of things again aregonna just compress to the
point.
And I've already experiencedthat, by the way.
Like there are certain companiesout there that have figured this
out.
And when you do support, likeI've done tech support with this
one company, it's all AI, andit's at first I was like, okay,
how's this gonna be?
And it turned out to be amazing,like way better than like super
fast answers, really accurate,and knew every single question I
(17:28):
asked, like instantly, and I wason and off on that support call,
like within minutes.
Normally that'd be like a15-minute call or more.
So me ask my supervisor, right?
Like, is the things that likewould don't need to happen
anymore.
Anyway, and then so those twoare kind of horizontal type
jobs.
And what I mean by that is theycut across disciplines, so
(17:49):
there's compression that's onthat compression that happens
between idea and delivery willbe most experienced by people in
those two horizontal domains.
So right now, you know, if youwant to build a website or
something like that, it's likethings are very staged between
people.
It's like, oh, you have an idea,okay.
Well, let's get the designerinvolved to design it.
Okay, let's get the UX UI study.
(18:09):
Okay, let's let's get theprogrammer involved, figure out
how long it's gonna take.
Okay, let's do who's the testerneeds to come up and come up
with the test.
Like all of that's compressingso that the people in these two
domains of direction andinteraction get a much quicker
experience from beginning toend.
And it's compressing already aswe speak.
The other direction, the othertwo uh human involvements are
(18:31):
what I call vertical instead ofhorizontal.
And they're just they're thebuilders and the supporters, so
they're the ones building andsupporting and maintaining this
infrastructure that is the thingthat is the special sauce,
right?
Like it could be yourdifferentiation, could be uh,
you know, all of your moatthat's created from your data,
your specialized processes, theproof that you've run this a
(18:54):
hundred times and on thisparticular type of problem, and
you've gotten these outcomes,which are much better than
anyone else, you know, whateverthat is, they're the ones
building that secret sauce, thatvalue in there.
And you know, that's all thelike the building of the agents
and the optimizing of them andthe memory management systems
and helping the AIs to replicateand to fix their problems, and
they're the ones behind theglass looking over the machine
(19:17):
floor at the robots that arebuilding the cars, right?
Like when we went from becausewe're kind of transitioning in
that in the agency world fromthe world where you've got 10
people on the floor who are eachputting together a different
part in the car, and right, andsome people are trying to figure
out how they can get a robot topass them a tool on the floor,
right?
That's like AI assistance.
Can you give me the right toolin anticipation of my need?
(19:40):
And then the shift with theagents with agentifying the
whole thing is like, no, no, no,just get off the floor.
We're gonna put you behindglass, and you're gonna look
over the hundred robots.
Five people look after thehundred who are building the
cars.
And there's still work to bedone, there's a lot of work to
be done for those people, butit's a very different kind of
job, is what you know we'reseeing.
It's architecting, uh, you know,translating things that are in
(20:05):
people's heads to now bequantifiable into a process,
right?
So it's a lot of processengineering.
SPEAKER_03 (20:10):
It's it's agent, well, it's human on the loop,
not in the loop anymore.
SPEAKER_01 (20:15):
They're just yeah, or human behind glass is what
it's like this year, right?
It's like it's not even on theloop anymore, it's just behind
the glass and looking at thedata.
SPEAKER_03 (20:23):
Yeah, well, it's just like auto-approve, like set
set the guardrails and likeauto-approve.
SPEAKER_01 (20:29):
It's we have yeah, we use tiering for risks, and
that's how we determineauto-approval.
So things that are low risk, andthat bar keeps going up.
What is low risk?
Yeah, those are auto-approved,medium, it finishes all the way
to the end, and then there's acheckpoint at the end, and then
high risk, it's a gate at thebeginning and the end, right?
So that's how we kind of there'sdifferent ways you can measure
(20:49):
it, but that's how we do it.
SPEAKER_03 (20:50):
And and I, you know, tell you tell me you're like I I
feel like I have a I have anagent set up that's just
watching, like making sure thateven though here's the
guardrails, like sometimes theydo crazy stuff.
Like, why why is that?
Why why would you say that ifyou give it definitive like
guardrails, it sometimes doesn'tfall it.
(21:12):
And I have to have like anobserver uh agent watching that
agent, making sure that it it itdoesn't, you know, hard code in
um you know uh API key or so,you know what I mean?
Like, why does it do it even ifyou tell it not to do it?
Like like I I would I would sayit's like data hygiene, but you
know, if this is a new project,like there it's like it
(21:35):
shouldn't be an issue, butsometimes it is.
SPEAKER_01 (21:37):
There are different reasons that could be happening.
I'm not gonna be able to tellyou one reason without actually
diagnosing your particular case.
But one reason, I'll just gothrough a bullet list and you
can dive into any one of them ifyou want.
So one of them could just be themodel mismatch.
You know, you might have a modelthat's just not good enough for
that.
Like if you're using haiku orsomething equivalent to that,
they're very pattern based.
(21:58):
They're looking for patterns andthey replicate them.
Rather than thinking throughproblems.
And then the mid-tier modelslike Sonnet, they're better, but
they can definitely, I've seenthem definite.
What you're talking about, likehard coding values in, like, oh,
your test case says that in thistest it should come up with this
answer.
So I'll just hard code thatvalue with a magic number in the
code, and all of a sudden yourpat your tests pass.
(22:20):
And it's like, oh my God, Ican't believe you just did that.
You know, that's not the goal ofthe testing, whatever.
But uh another category, so youknow, you might just need a
stronger model, right?
That sometimes that fixesthings.
Uh like Opus is very good atfollowing instructions, and uh,
we've been experimenting a lotalso with GLM5, uh 5.1.
SPEAKER_03 (22:39):
So you just think if they're getting get better, like
they're just gonna like that'sone, that's one vector, right?
SPEAKER_01 (22:44):
Of like possible reasons.
Another is context and memorymanagement, and and you could
also include skill managementand tooling as like all part of
that category.
So if you see, for example, andthis, oh, and then the the next
thing which ties into this ishow do you approach problems
when you see them?
So whenever we see a problem, wealways ask ourselves, you know,
(23:08):
the the question isn't why isthis model making this problem?
It's what can I do to improvethe system to make sure that
this problem doesn't happenagain, right?
And so that framing is reallyimportant to take.
And sometimes I see people justget start writing angry messages
to the AI or tell it moresternly how it should be doing
its job, you know, and like thatmight get you some results, but
(23:30):
it doesn't mean it's it's gonnahappen again to you, right?
Because what you're not doing,you have to really be thinking
about system design.
So, you know, so take a crashcourse on like how to be a good
system designer and look at whatare you know what are the top
five, 10 things you need tolearn to be that to embody those
qualities, and then that's howyou should be approaching the
problems.
So if you have an AI that's notfollowing the instructions
(23:51):
consistently, maybe you need,you know, you can even ask the
AI, do you have a good skill forthis?
What skill are you using?
Or you can look at the log ofhow it did it.
Maybe it tried something and itgot an error, and then it tried
something else, got an error,and then it tried a third thing,
and then it found the answer,and then it took that, and then
you can kind of watch what itdid.
And if you're seeing that it'skind of bumping along the way
(24:12):
like a pinball machine, you cango through that with the AI
afterwards and you canstreamline it with a skill that
tells it immediately the correctway to do it, so it's not having
to bump around like a pinballmachine.
Sometimes people, the prompt orthe system prompt or the thing
that they're writing is notactually phrased in the way that
is the most effective.
As models get stronger too, youreally won't be focusing more on
(24:36):
why you want to be doingsomething, which is often
forgotten, that you need tocommunicate that to the AI and
what the desired end goal of itis, and not just the steps, like
I want you to do step one, steptwo, step three, step four.
That can work, but especially asyou get to stronger models like
Mythos that's coming out, youwant to really be making sure
that you're guiding these AIsystems with goals and purpose
(24:59):
of what they're trying toachieve, because that helps the
AI a lot to make sure that it'snot going to shortcut you with
like a magic number in there.
Because if you're telling itlike we want to make sure that
this works for all possiblevalues of this variable, in
this, you know, here's going tobe some test cases, make sure it
passes those tests.
That's very different thansaying, here's a test, uh the
(25:20):
answer should be 1.5, go testit.
And then the AI is like, oh, wewant it to be 1.5, let's hard
code that thing in.
SPEAKER_03 (25:28):
So so I saw a podcast interview with the
founder, and he was just like, Iuse plan mode like for almost
everything, right?
To to to to give it thoseobjectives to answer those
questions, which I think areextremely helpful.
One of the things you made methink of was like, how are you
solving like persistent memory?
And you you mentioned obsidianearlier, and and we've started
(25:52):
to toy around with that.
I'm not like an advanced user,um, but I I would love to know
kind of like how you're solvingthe persistent memory and kind
of like connecting it alltogether because you only like
you said, you only have so muchin the context window, and when
it compresses it, it kind of itit loses stuff, and you got like
(26:14):
the heartbeat component.
Like there, there's just like uhwith the so many things I can
say now because you keepmentioning new things.
I know, I know, I'm sorry, likeclean it up for me, clean it up
for the listeners of like I do.
SPEAKER_01 (26:26):
I want to also just touch base on plan mode because
plan mode is awesome when youwant to be building things, but
there's a very big differencebetween planning, which is what
to build something, versus usingsomething.
You can't use plan mode to usean identification.
Yeah, so just want to make surethat that's clear.
But in terms of building, Ithere's a lot I could say around
(26:47):
plan mode, how to make that evenmore effective.
Let's pull it out.
I don't know.
We just want to we don't wantto.
SPEAKER_03 (26:53):
No, I don't know what's let's go down some rabbit
holes of like and let's likeactually give some context.
So let's talk about plan modeand we'll move into the okay.
SPEAKER_01 (27:02):
So there's first of all, there is a big debate right
now between where to get themost leverage and plan in in the
planning and execution whenyou're doing agentic coding.
So you're using a team of agentsto write and execute the code
and self-test and all that kindof stuff, right?
So it's not AI assistance whereI'm just having it write one
function at a time, one class.
It's like doing the wholerefactor or the whole project.
(27:24):
Um, those three points ofcontention are should humans be
spending most of their timeduring the planning?
Should they be spending most oftheir time during the code
review?
Like some people are sayingright now.
Some people are changing theirmind too.
They some people used to say,read all the code.
Oh, that's right, sorry, readall the plan, don't read the
code.
And then those same people arenow saying we made a mistake.
(27:46):
You don't don't read all theplan, read all the code.
But there's a third place toalso, there's a third philosophy
or a third camp, which is reviewall the tests, right?
So that's where you really needto be putting all your energies
at the end point, is thetesting.
So you're kind of doing atest-driven setup.
So the I do think personallythat that contention of where to
(28:06):
put the effort between thethree, I think of it almost, you
know how you have those min-maxabsolute points on curves, where
if you had a two-dimensionalgraph and there's like a bell
curve and you want to kind of bein the optimal value with a peak
somewhere around there.
But because it's three points,like time, uh sorry, testing,
code, and planning, it'sactually in like a
(28:27):
three-dimensional space betweenthose three, and like where do
you put most of your effort?
But I do think that that wholequestion will go away as models
get bigger, better and bigger.
Um, there's still so much to bedone around memory and context,
which you're hinting on.
I think that's really a big gapright now that companies will
(28:48):
most likely start having tofocus on a lot more in the
coming year or two.
Um, but in terms of theplanning, when you're doing it,
so the way we build plans isfirst we write up a uh either a
product design requirementdocument or an ADR, an
architecture design requirementdocument.
(29:09):
Depending on the refactor or thecode that we're writing, that
could be somewhere around two tosix thousand lines of planning
that is created.
That's a whole process.
This is where you're putting alot of time in, at least for us,
is the planning.
Our planning time is gone, it'slike way higher than it used to
be, right?
We used to just like write awrite some specs, figure out
some stuff, and start coding,right?
Now we're spending about doublethe time than we used to
(29:31):
historically on all of thisplanning stuff.
And so you so we do that firstplanning up front, then we take
that document.
We're not done with planning,because then we take each phase
or each section of it, and eachof those is then broken down.
Uh, one of the things we tellthe AI is write it as if a
junior was gonna write thiscode, right?
Like how how give the juniorenough information that they
(29:53):
know exactly what they need todo for each of these steps.
There's more to it too, becauseyou really have to think about
edge cases and how you're gonnatest things and validate, and
how are you gonna get the AI todo all of this work without a
human needing to be involvedwith these steps?
Sometimes you have to build abridge.
There's like mini projects thathave to happen to just build
those bridges or thoseself-validation loops.
(30:15):
So you build all of that out,and then from each of those
documents, then you build uhwork order collections, which
are breakdowns of uh contextthat is a living document, plus
a planning, which is a livingdocument, and then actual
individual tasks, which is alsoliving documents too.
Living, meaning that they'reedited as they're being used,
(30:37):
they're not just static.
And then those collections, andthere are different frameworks,
but this is kind of the mostsimple one.
Those are then passed on finallyto a team of agents that then
has their own um state-drivencontrol system that I was
talking about to make sure thatthings are grounded so that and
they have antagonizingobjectives, is also important
(30:59):
too.
Like you want the tester whosegoal is to find as many bugs as
possible, and they get pats onthe back or best fit points, or
however you want to look at it.
And then the developer wants, ofcourse, to code successfully,
and then the task manager wantsto make sure that everyone's
proving that they're doing theirjob so they can mark them all
off.
So every single role in it hastheir own objectives, and some
(31:20):
are antagonizing.
And then they go through all ofthose, self-check and report,
and then you just need to, andthen from there, you as you
sophistic make the system moreand more sophisticated and
resilient, you can accomplishlarger chunks of work at a time
before a human needs to even goand start reviewing it.
So it might run for five minutesin the background when you're
building confidence, but theneventually it's running for like
(31:41):
30 minutes or more in thebackground as it just crunches
through these differentdocuments in series.
Um now, planning one-offplanning mode in claw code, that
I would say I kind of view it aslike it one step above AI
assistance, because what you'redoing is you're basically doing
(32:02):
a one-shot job in in air quoteson a particular function or
feature or mini refactor.
But there's a limit to the scopeand scale that that can
accomplish on its own.
But it is definitely way betterthan just a one-shot prompt in
it.
But even the plan mode needssupporting memory and skills and
(32:23):
things like that in order tomake sure that the plan is
grounded in truth and doesn'tjust have assumptions,
surface-level assumptions in it.
That's that's kind of the devila lot in the detail with a lot
of the plan mode, is that AI hasis sitting on top of a very
large code base, reads threefiles, but doesn't read the
other 10 it needed to read, doesa plan based on assumptions and
(32:46):
what the other files does, andall of a sudden it's completely
gonna just screw you over whenit goes through.
You know, we've seen seen thoseproblems happen.
So, you know, plan mode can beamazing, but it can also be
insufficient for what you'redoing.
SPEAKER_03 (33:00):
No, oh yeah.
No, I I I think it was kind oflike a bare minimum.
Like if you're gonna like just aone-shot prompt, like if you if
it's not really elegantly uhengineered, uh it's not gonna
give you the output you need.
Because there's a lot ofconversations, I guess, on
social media of like all thesetools that are just kind of
wrappers anyway, and it's likebuild me a whole app, and then
(33:23):
it like breaks.
And it's like you can't build awhole app with one line of code
that and then and then you gotto think about well, how many
users are gonna go through thisapp at the same time?
Because you know, if you if youscale it up, you got one one
function processing at a time,like it's not gonna be super
useful.
(33:43):
And so that actually, you know,I know I'm pivoting all around.
I have so many questions foryou, but that's been some of the
debate with a couple clientprojects that I got pulled in
on.
Is okay, like we want to vibecode something, um, and it's
functional, but there's somesecurity concerns, what kind of
data is gonna go through it, howmany users it's gonna be, and
(34:05):
then like, okay, can like base44 or so some of these tools um
uh replaced an enterprisesystem, right?
And it's like, no, use it tocreate the MVP and then at a
certain point switch it over toa developer.
And I'm starting to see withwith these agent systems, it's
(34:27):
like, well, they can code it up,like, you know, like like okay,
maybe you're not using thistool, but these agents, if if,
if they have sufficientinformation, and I get it goes
back to the the memory.
I feel like the memory is thatthat big issue right now, but I
mean they can code it inwhatever you want to code it in.
Like you can build a plugin, youcould, you know, build it with a
(34:49):
similar back end.
Like if you're building aWordPress site, you could build
a similar looking WordPressbackend, or hey, let's just use
um the the the command line orthe terminal to do all the work
and let's not ever touch thesite.
Like, like there's a lot ofthese discussions of like, I
feel like it's when um likedigital cameras switched over to
(35:10):
like a new technology, and it'slike like you you gotta make
like a hard switch at some pointto create different processes
and do things differentlybecause a lot of people are
augmenting the old way, butthere's like a new way of doing
things.
And like, where is that switchover point and also the maturity
level of of what these thingsare doing?
(35:32):
I, you know, like do you have tohand it off to a full stack
developer to build it out?
Or, you know, can these agentsbuild it where it's enterprise
ready?
Like, what's happening withSaaS?
I there's a lot of questions inthere, so feel like to answer
whatever one you want.
SPEAKER_01 (35:50):
Sure.
So what I would say is, I mean,talking about that last thing
you were just mentioning aroundthe full stack developer.
Yeah, the way we've seen so wemade a rule for ourselves first
experimentally back in December,which was can we do this new
project never writing a line ofcode?
And it was a pretty it was ahydraulic system simulation code
(36:14):
refactor, you know, a meatything, not just like something
you could just spin out as afive code thing in it in an
afternoon.
And back then it was alsopre-Opus 4.6.
I think it was 4.5 at the time.
There's something magical thathappened with 4.6, but that's
comes later in my little story.
But man, it was like the firstmonth was like painful because
(36:37):
we were just we had this ruleand we're trying to go through
it, you know, regardless of howto learn how can we get this
thing to to be as good as usdoing it by hand, but not right.
And so, but now with the waythings have progressed and and
all the things that we'velearned, and then also 4.6,
which did like this magical kindof frosting on the top and like
(37:00):
really helped brought it evenmore to reality.
I at least for us, like it'skind of over the whole idea that
you write code by hand, youknow.
Of course, if the client comesover and it's just like, I need
to change something quickly, Imean, maybe you just change the
line of code, or maybe you justone-shot that one little thing.
But any kind of major refactoror major project, it's we're
(37:22):
seeing that it's done way fasterand has less problems than when
people are doing it.
By less problems, I mean lessbugs, you know.
So, I mean, that's kind of a bigkind of milestone.
Is that and then by and lessbugs can compare to like let's
say the average programmer,right?
I mean, of course, there's gonnabe senior coders out there who
are gonna still beat these AIsystems today.
(37:43):
I don't know how long that'sgonna last, honestly, like
probably a year from now.
That might not even be trueanymore.
But at least right now, youknow, certain senior devs can
probably outcompete the AIstill, but probably not on
speed, but more on you know,bugs and broader thinking of
impacts and things like that.
So there's there's still some alittle bit of a moat there.
But um, but yeah, in terms ofenterprise ready software and so
(38:08):
forth, you do need somebody whois that brain who really is
thinking through, you know, whatare the security levels?
How are we gonna test this?
How is it gonna integrate?
When we're done with featureone, what's coming next?
How do we not paint ourselvesinto a corner?
All this kind of like broaderlevel thinking, the AI is not
capable of doing that.
AI is terrible at askingtangential questions.
(38:28):
You know, it's like the AI willtake a task and it will converge
on what is the most likelyoutput or answer for that
particular problem.
In fact, people get reallypissed off if the AI comes back
at them and throws somethingthat they weren't wanting to do.
You know, uh I don't this isjust popping in my head, but I
remember last year, this was awhile ago, but someone posted on
(38:49):
social media that they wereasking the AI to write some code
for them.
And instead of writing the code,the AI responded back, telling
them that they think the AI saidto them, I think instead of me
writing the code, you shouldreally go off and read these
things to become a betterprogrammer, probably because
what they were asking them to dowas just like flat out wrong.
And the person was so pissedoff, and everyone responded to
(39:12):
that social media post saying,like, oh my God, this is the end
of humanity and AI is revoltingagainst us and stuff like that.
But it's like, you know, likepeople don't even want these
systems right now to bethinking, you know, there's this
risk that we don't even wantthem to not obey us, right?
To say to us, like, hey, whatyou're doing is probably you
haven't thought of this thing.
Like, are you sure we should bedoing this?
Like, people would be likelosing their mind if it does
(39:34):
that.
No, it should be.
SPEAKER_03 (39:36):
Well, I think that that's you know, you go back to
like different um of thesedifferent models and like how
they're how they're buildingthem and what they're trying to
achieve.
Because I feel like Claude ispushy, like it's pushy.
It's like, and I'm like, hey,you don't know all the facts,
like let me give you some morefacts, especially kind of more
(39:56):
before we switched over uh tokind of the agentic era.
But but I'll tell you, chat GBT,unless you turn off like nice
guy mode or you, you know, youhave a devil's advocate going
on, it's just gonna tell youwhat it wants.
Like it's and that I think thatthat's the approach.
And people started using chatGBT and they're like, Oh, I'm
(40:16):
always right and I'm amazing andall this kind of stuff.
And then you switch over toClaude and Claude, like, you
know, or Grok will tell you likewhat it is, and then you're
like, Whoa, like I didn't expectyou to say that, right?
SPEAKER_01 (40:28):
You know, and et cetera, yeah.
I mean, it you can take theanalogy also with people when
they hire people, right?
If I hire someone for 15 bucksan hour, 20 bucks an hour,
whatever, and I tell them whatto do.
Most people don't want thatperson saying to you, like, are
you sure this is the job Ishould be doing, right?
Like they just you just want theperson to do the job.
Then you pay someone$200 an hourand you tell them what should be
(40:50):
done, and they come back andsay, actually, you should be
thinking about these things,right?
All of a sudden you really wantto hear that.
And if that person doesn't pushback, you think that that person
doesn't know what the hellthey're talking about, thinking,
you know.
So there this also, I think, istranslating across models and
different price points.
And the only this is gonna thisis a bit controversial, but I
think you could totally takegiving an idea here to OpenAI,
(41:12):
you could totally take yourmodel, the the mass model, and
just only remove one trainingthing, which is that when you're
training the model, don't makeit please the user, but instead
be like more contrarian and andpushing back more, and then make
people pay more money for that,and they will take that model
more seriously because they'repaying more money for a model to
(41:34):
actually um you know push themrather than just please them.
But you know, but most peoplewant cheap and free things to
please them, they don't wantthem to contradict them.
So there's this psychology inthat as well.
SPEAKER_03 (41:46):
No, it's it's super interesting.
Okay, so I we have a few moreminutes.
I would love to talk a littlebit more about kind of how you
think about orchestration,because like a lot of how I was
thinking about orchestrationbefore was like, okay, one agent
specializes in one skill.
And now you can start to add alot of like skills to one agent
(42:08):
and make it like a super agent,right?
But there still is differentroles that you want to define uh
for it to do.
But how I was thinking aboutorchestration before, and even
like some of the books that Ihad bought and read, like one
third of the book is maybeuseful now, and two thirds is
like we've already passed likeeverything it was talking about.
(42:30):
And so where on kind of frontierstuff, when you're setting up uh
these systems, how are youthinking about orchestration
today?
SPEAKER_01 (42:40):
Yeah, so there's two parts of orchestration.
There's the human orchestrationpart of it, and then there's the
agent-to-agent orchestration.
You're talking about more thelayer, right?
SPEAKER_03 (42:48):
Yeah, agent to agent, yeah.
SPEAKER_01 (42:50):
Yeah, so I would say two things.
First is the architecture of it,and the second is the solution
problem-driven part of it.
So if you're gettinginconsistent results from a
process, the process you mightbe putting too much agency on
one agent, and you need to splitthat up into sub-agents or
sub-steps, and you might need toformalize them to become more of
(43:12):
a linear process rather than aqualitative decision matrix,
right?
So, one problem I see often isone agent will be given too much
too much space, too many skills,too many things it could do, and
you're asking the agent to thenmake broad decisions instead of
making it more controlled.
(43:33):
And so, if you can, one quicksolution to that would be you've
got an agent which first does adecision on what type of request
is coming in.
Is it a, you know, uhsecretarial request?
Is it a booking request?
Is it a research request from myhuman, whatever it might be, or
from the clock or from someother info, whatever it is?
(43:54):
And then it dispatches that toanother agent, which then
specializes on just that.
And then Um that that state uhscript driven element to it is
also important there.
Splitting it up can also helpyou with that memory management
context.
I know we probably we may nothave time to get into that.
I know that was one of yourquestions, but memory management
(44:15):
is, of course, really importantright now.
Context management, contextfirewalling as well between
agents.
SPEAKER_03 (44:21):
And then you talked a little bit about the bridge,
like building like littlebridges.
Like, can you give an example ofthat maybe?
SPEAKER_01 (44:29):
So if I'm I hope I'm talking when I talk about stop
me if I'm going down the wrongpath here with the bridge,
you're talking going over thewrong bridge for what you're
talking about.
But I would are you talkingabout like the, for example,
that SEOGO agent who comes upwith a brief and then the
connection to the next agent.
Yeah, yeah.
SPEAKER_03 (44:43):
Like you have to build a little of the brief to
do the bigger project, likegiving people an example of what
that is, you you can start toseparate out what's being done,
and you're connecting like maybetwo systems together, if you
will.
SPEAKER_01 (44:56):
Yeah.
So going back to my analogy ofthinking about jobs.
So if you think about a job andthe domain of what that job is
and the knowledge that's neededfor that, that gives you a
contained space, that job maysometimes need to be split up if
you're putting too many jobsinto it.
You know, I sometimes I'll lookat a job description for a
human, and then when I look atit for an AI, it's actually
three jobs because it makes moresense that way.
They're just trying to hire somekind of unicorn to do these
(45:18):
multiple roles all together.
And then, but then just like inin traditional systems, the some
of the most important parts isthe connectors, the edges
between systems, those bridges,those gaps, or APIs, or whatever
you want to call about it.
And so you need to formalize theoutput to the input of the next
system and how you're going tocontrol for handover and errors
(45:41):
and expectations and checksbetween these systems.
One sec, I have to clear mythroat for a second.
Okay, sorry about that.
And my and my coffee cup'sempty, so I couldn't just liquid
liquidate my throat there.
Liquidate, liquefy, I don'tknow, whatever the word is,
lubricate.
That's what I'm looking for.
(46:02):
Um anyway, so you have you haveto look at that bridge point,
and then there's different waysyou could do that.
You could use an inbox systemwhere you just talk to each
other, you could use afolder-based system where they
scan those folders, you coulduse uh more like a spreadsheet,
right?
Like you drop it on aspreadsheet and it could be a
database.
I don't know, it depends on howsophisticated it needs to be,
how much you need to scale it.
SPEAKER_03 (46:23):
Because I I I was like building stuff that I was
having perplexity create, and Iwanted it to talk to Claud Code.
And the best way I could figureout how to do it is dump
everything into a spreadsheet,and then Claude Code recalled
it.
Like I was trying to get them totalk to each other, uh, but I
just had to like drop everythingum in a document and then you
(46:45):
know have it uh when you saycomplexity, it's just you using
perplexity browser and so comet,uh, you can spin up multiple
agents at a time to do likedifferent kinds of research.
SPEAKER_01 (46:57):
Okay, yeah, yeah, yeah.
SPEAKER_03 (46:58):
And so so, but I was trying to get them to talk to
each other, and I I was justdropping in a spreadsheet and
then recalling when it wasupdated so it would check it,
and if new information waschecked, it would pull it.
Uh, it's kind of how I got itset up, but it's pretty rude for
it.
SPEAKER_01 (47:13):
When we're using yeah, when we're using
perplexity, because it's greatfor certain, it's really good
for certain things.
We're using it as an API toolfor an agent that itself has
access to databases or memory orwhatever.
So we don't need to that wholekind of what you're talking
about, we don't end up having todeal with it because agent, I
don't know, Bob.
That's called agent Bob.
(47:34):
But Bob has the API perplexity,does research with there, gets
data back, stores it either in aJSON file or a table or
superbase or whatever kind ofneeded.
And then the next agent can justpull straight from that system.
So it's just all within the sameecosystem.
Yeah, not having to translateacross a third medium.
Got it.
SPEAKER_03 (47:53):
Yeah, yeah, awesome.
So I I I do have a hard stop onthe hour.
I would love to kind of I know II jumped around.
I'll I'll I'll definitely haveto have you back.
I I really enjoyed ourconversation.
I know we were talking somephilosophical stuff.
Um, yeah, no, you try to talk tosomebody on the street about
this.
Sometimes uh people are like,what are you talking about?
(48:15):
Oh, totally.
Um, no, maybe uh is thereanything in the course of this
conversation that you thoughtmight be really valuable to just
add to to this conversationhere?
And then if you could share alittle bit more about what uh
Fountain City is doing and howto get in touch with you.
Uh, you talked about some blogsyou're writing, uh, and that
(48:37):
would be phenomenal.
SPEAKER_01 (48:40):
Things that weren't discussed.
I mean, there's so many thingsthat could be talked about, you
know, the security and how areyou handling that?
SPEAKER_03 (48:46):
There's scalability, there's don't drop your don't
drop your API keys right in theprop, guys.
Like, don't do that because it'sgonna get processed.
Like there, you gotta you gottaset up a file or something like
that where it pulls from.
SPEAKER_01 (48:58):
So yeah.
Yeah.
So there's there's a lot to talkabout there.
Um I I yeah, I think everyeverything I can think of is
just rabbit holes.
So I'm I won't bring up like toomany things that haven't been
talked about.
But in terms of me, us, fountaincity, so fountaincity.tech is
the website, and our clientsright now are actually a lot of
(49:20):
agencies are hiring us to buildeither systems for themselves
and or for their clients.
Sometimes uh one person alsowe're talking to right now is to
build a vertical stack system.
So it's they have lots of reallytiny clients, like 80 clients in
one vertical.
Uh, I won't name it, but youknow, imagine it could be like,
(49:42):
I don't know, plumbing orservices, yeah.
SPEAKER_02 (49:45):
So we've got a service.
SPEAKER_01 (49:45):
So some kind of services industry or some kind
of industry.
And so then you you get thiseconomy scale because you can
have one research pipeline, andthen it generates a whole bunch
of topics, which then getinterpreted and resegmented for
different writers for each um ofthe different subdomains.
So we're doing a lot, yeah, likeI said, with agencies, direct to
(50:05):
client, and then these these uhintegrated verticals or
multi-site hosted stuff.
So that's kind of one part ofour business.
And then the other part is doingwhat I described with the agenda
coding, because we're able to,and then all those projects tend
to be building systems that haveAI in it.
So, like a uh right now we'vegot two biggest projects we're
running is one is a voice systemwhere it listens into enterprise
conversations and then givesvery and then listens into these
(50:29):
enterprise leaders speaking andthen gives very thoughtful input
to kind of help them to thinkbroader or think deeper or think
in a new way about the problemsthey're bringing up.
It's a very interesting project.
And then the other one we'redoing right now is on a data
intelligence layer.
So we're connecting together.
People come to the website fromall different possible paths,
(50:51):
everything they interact on thewebsite, and then everything
they've purchased or thinkingabout as they touch all these
different systems, and thenbuilding that out into on-demand
charts, uh, being able topredict, you know, who should
they go talk to first, and thenbeing able to AI talk to the
data and stuff like that.
So that's another cool kind ofproject we're doing.
SPEAKER_03 (51:10):
So I'm using a tool called uh E-R A F A N A.
Are you familiar?
I don't even know how to saythat.
SPEAKER_01 (51:18):
Can you pronounce that?
Sorry, Grafna?
Is that what you mean?
SPEAKER_03 (51:21):
Grafna, yeah, Grafna.
Um, basically to to to take likethat data intelligence layer
that you're talking about, yeah,to throw it into these kind of
dashboards.
You can look it up.
You would I don't know whatyou're using.
I would love to have a toolsconversation with you.
SPEAKER_01 (51:36):
Um we're using grist on that one, okay, which is
another G letter.
Yeah.
It's not grapna, it's grist.
It's close.
I was almost wondering if you'retalking about that.
But yeah, we are using thatright now as kind of a it's it's
kind of like a it's like aspreadsheet in a database had a
baby kind of thing.
(51:57):
So it's a bit like a spread, orit's like a almost like a uh
what was that thing called?
There was the spreadsheet thingthat was popular with project
managers a few years ago, whereit's like super spreadsheets.
So it's kind of like thatbecause they this client really
likes spreadsheets, so we'regiving them that as like a
low-cost kind of middleware toall of the data collection, and
then from there, because it'sessentially database, we can
(52:19):
connect it to anything we want,visualization, AI model context
and stuff like that.
SPEAKER_03 (52:23):
Yeah, this would be like the next generation of like
BI or looker or something likethat that's more interactive and
you can pull in more data.
No, I uh Sebastian, thisconversation's been been
awesome.
So, where are you putting outnew new information?
You talked about your blog.
Is there like do you havesocials stuff?
SPEAKER_01 (52:43):
Yeah, so LinkedIn, my account's probably the best.
Uh so just my name on LinkedInand um YouTube.
I think it's been a bit sloweron YouTube recently, but I I it
used to be once a week, now it'skind of like once a month.
But I'll probably pick it backup a little bit more.
I just have some speaking eventscoming up, so I've been and this
(53:04):
launch of this kind of virtualdigital agency product has been
very time consuming.
So I've been more focused onthat recently, this kind of
autonomous agent system.
But anyway, YouTube, LinkedIn,the blog on our website is
almost daily blog posts oneverything related to this, like
you know, from security to uhimplementate implement
(53:25):
implementation.
I'm stumbling over my words, andthen um what else?
There is a newsletter also onFountain City.
It's pretty, it's not very oftenthat we newsletter out, so you
won't get overloaded, but youcan sign up for that too on our
website.
SPEAKER_03 (53:38):
Awesome.
Well, we'll definitely have tohave you back on.
Um, I I have so many morequestions for you, and and uh I
like all the stuff you're doing.
If anybody's listening, this iskind of where where it's all
going.
Um, and so you need to I I thinkreally understand the um taste.
Is that what we said?
Is that what we said on like thetaste of kind of where you what
(54:03):
is your secret sauce?
Yeah, like why people shouldhire you.
What are what are you doing?
I think that that pin pendulumis going to kind of swing back
because a lot of these thingsare gonna be compressed and um
you know, uh programmatic isjust, you know, like the world's
changing.
Um and so uh it's hard to stayon top of all this stuff.
(54:27):
Um, Sebastian, it's great tohave you on.
Um, thank you for being here,guys.
Go check out what he's doing.
Um, I think uh, you know, thisis the future.
Um until the next time, if youwant to grow your business with
the largest, most powerful toolon the planet, which I would
have said is the internet, butyou know, I would say it's now
(54:49):
agentic agents.
Uh keep listening to thispodcast.
We'll keep bringing on um someof the people in the industry
that are moving it forward.
Uh, thank you so much for yourtime.
Uh until next time.
My name is Bat Bertram.
Bye bye for now.
Popular Podcasts
Betrayal Weekly
Betrayal Weekly is back for a new season. Every Thursday, Betrayal Weekly shares first-hand accounts of broken trust, shocking deceptions, and the trail of destruction they leave behind. Hosted by Andrea Gunning, this weekly ongoing series digs into real-life stories of betrayal and the aftermath. From stories of double lives to dark discoveries, these are cautionary tales and accounts of resilience against all odds. From the producers of the critically acclaimed Betrayal series, Betrayal Weekly drops new episodes every Thursday. If you would like to share your story, you can reach out to the Betrayal Team by emailing them at betrayalpod@gmail.com and follow us on Instagram at @betrayalpod and @glasspodcasts. Please join our Substack for additional exclusive content, curated book recommendations, and community discussions. Sign up FREE by clicking this link Beyond Betrayal Substack. Join our community dedicated to truth, resilience, and healing. Your voice matters! Be a part of our Betrayal journey on Substack.
Dateline NBC
Current and classic episodes, featuring compelling true-crime mysteries, powerful documentaries and in-depth investigations. Follow now to get the latest episodes of Dateline NBC completely free, or subscribe to Dateline Premium for ad-free listening and exclusive bonus content: DatelinePremium.com
Stuff You Should Know
If you've ever wanted to know about champagne, satanism, the Stonewall Uprising, chaos theory, LSD, El Nino, true crime and Rosa Parks, then look no further. Josh and Chuck have you covered.