September 24, 2025 • 34 mins
What if AI could be more than just a tool—what if it could become your thought companion, helping to extract and refine the ideas already in your head? In this fascinating conversation, Brad Moss, founder of Enhanced AI, reveals how his journey from video game development led to creating a platform that empowers anyone to build sophisticated AI agents without coding skills.
Moss introduces a powerful concept: AI as "the right brain of technology." While traditional technology demands precision and operates on exact data, AI excels at pattern recognition and creative generation. This distinction helps explain why many misunderstand AI's capabilities and limitations. Through his platform, Flux Prompt, Moss aims to help users orchestrate multiple AI systems to accomplish complex tasks that would typically require engineering teams.
The discussion explores how AI has evolved from a simple tool into an extension of human thought processes. Rather than replacing human creativity, well-designed AI systems amplify it by handling routine tasks and helping organize ideas. Moss shares practical examples of professionals using his platform to reverse-engineer competitor brand profiles and develop marketing materials that outperform traditional approaches.
For students and professionals navigating this AI-first landscape, Moss emphasizes understanding patterns and frameworks. Since AI fundamentally operates on pattern recognition, those who can effectively communicate frameworks will excel at guiding AI to produce meaningful results. The conversation also addresses privacy concerns, distinguishing between consumer-facing platforms and more secure API implementations.
Ready to build your own AI agents? Try FluxPrompt and discover how orchestrating AI systems can transform your productivity and creative output. Special discount available for Mediascape listeners!
https://enhanced.ai/fluxprompt/
This podcast is proudly sponsored by USC Annenberg’s Master of Science in Digital Media Management (MSDMM) program. An online master’s designed to prepare practitioners to understand the evolving media landscape, make data-driven and ethical decisions, and build a more equitable future by leading diverse teams with the technical, artistic, analytical, and production skills needed to create engaging content and technologies for the global marketplace. Learn more or apply today at https://dmm.usc.edu.
Mark as Played
Transcript
Episode Transcript
Available transcripts are automatically generated. Complete accuracy is not guaranteed.
Speaker 1 (00:01):
Welcome to Mediascape insights from digital
changemakers, a speaker seriesand podcast brought to you by
USC Annenberg's Digital MediaManagement Program.
Join us as we unlock thesecrets to success in an
increasingly digital world.
Speaker 2 (00:22):
We just had an entire podcast interview before we
jumped on Brad Moss.
You have done so much and youcontinue to do so much, and I'm
so thrilled to have you onMediascape Insights from Digital
Changemakers today.
Speaker 3 (00:36):
Oh, thank you, Annika .
I'm thrilled to be here,thrilled to be here.
Speaker 2 (00:39):
We didn't even realize we had this video game
connection.
I used to work in video gamemagazines and launched the
official Xbox magazine when thatcame out.
Speaker 3 (00:47):
And you have you've created how many Over 60 video
games in my career.
Speaker 2 (00:52):
Okay, amazing, but we're really here to talk about
AI that does tie in, so yeah itdoes.
Speaker 3 (00:59):
I can tell you a little bit of the story of that.
That's actually how I got intoAI.
Speaker 2 (01:03):
Let's do it.
Yeah, please share.
Speaker 3 (01:10):
Yeah, so I mean my background.
After video games I went to bigtech and I worked inside of
Amazon.
I ran the third party sellingplatform the third party
marketplace platform thateveryone uses to sell on Amazon
and then built a new businessfor Amazon.
Then I left after a while aftermaking lots of money for Amazon
, their shareholders, and then Iwent.
I'm a little bit of anentrepreneur and I went and
built some e-commerce companiesand whatnot, but fast forward
like seven years.
(01:31):
I love video games and I'dbuilt a lot of games.
But the game business is reallytough.
It's really brutal and it evenis right now.
There's a lot of layoffs withsome of a lot of my friends
who've been hit by them, butit's.
The process of building games isjust like so cumbersome and
we've seen it from inside theindustry too.
So about three and a half yearsago I got into AI.
Before ChatGPT like exploded,it was GPT-3.
(01:54):
We even started hitting two atthe time from time to time.
But when I discovered that itexisted generative AI I mean
I've been coding regular AI invideo games.
You know, 20 years ago Irealized this generative AI was
like really powerful and that Icould actually accomplish a lot
of things and the whole process.
The whole goal was tostreamline operations of game
(02:15):
development, of like, hey, howdo you burn through iteration
cycles and make sure this is fun?
Because games is all aboutfinding the fun and you got to
go through iteration afteriteration until you finally find
it and you refine it more andmore and more.
And I was trying to automatesome of that process and
discover generative AI, and sothat's how I stepped into.
The whole world is trying tojust refine and make game
(02:37):
industry, the game business,work better, because making
games is super fun.
Speaker 2 (02:42):
Yeah, well, and you just used AI to help you create
a new game.
You want to talk?
Speaker 3 (02:48):
a little about that.
Yeah, this is kind of a randomthing so I could grab it.
I've been working on a boardgame idea I've had for about six
years and it's all about thenot nearly what this whole
podcast is going to be about,but this board game idea I've
had for about six years and Ithink the point of it is that AI
is just empowering and I've hadthis idea for a long time.
(03:08):
I prototyped it with my sonwith old card deck of cards, but
then over time, you know, Istarted using AI and I generated
like these, you know this, thiswhole card game.
You can't really see it with my, with the blur on it
(03:44):
no-transcript, don't go on thatdirection, go in this direction.
But it's kind of this thoughtcompanionship that we do or that
I've been doing both that likewriting executive briefs to my
board, you know, and to myinvestors.
Using AI is almost a companionand a lot of things to get out
of my mind, things I have but Ineed to kind of pull them out a
(04:05):
little bit more.
Speaker 2 (04:07):
You are preaching to the choir.
This isn't AI related.
But a couple of years ago I waswriting a book chapter for a
book about women and businessand business on purpose and
having really purposeful lives,and I was having writer's block,
not because I didn't know whatI wanted to say, but because I
don't have time to sit down andtype it all out.
So I started just voice memosto myself.
(04:30):
Now this is before Gen AI iswhat it is today, so I don't
know if that was an option, butnow that's absolutely something
I would do, and I speak toClaude every single day as my
thought companion my favoriteone, right now at least and so
it's really exciting to see howwe can really utilize this
technology in the best examples.
(04:50):
Right, it's not about writing apaper for us, it is about
taking our thoughts, ouroriginal thoughts, and putting
them in order and helping usthink through what are we
missing.
So I really love that, andthat's where Enhanced AI really
comes in.
I'm in your agent buildingclass right now, just started
this week really exciting and wemet at an AI conference earlier
(05:14):
this year, so I'd love to talkabout that.
So you discovered AI and thenwhy did you decide to create
this brand, this company.
Speaker 3 (05:24):
Yeah.
So when I discovered AI and thepower of it, I realized pretty
quickly so you got to rememberthis is before ChatGPT existed.
This genre of technology wasreally fascinating, but to me it
could do, and I actually bringthis up in the class.
I consider, and I came to thisconclusion after about a year of
thinking through it and workingwith it.
Ai feels like the right brainof technology where we're used
(05:48):
to dealing with the left brainof technology.
Right, like technology, is veryprecise, exact.
It's like a database.
You can get exact things out.
That's very much like the leftbrain.
Ai is like the right brain whereit's really creative and it's
generative and it's inspiring,where it's really creative and
it's generative and it'sinspiring.
And one reason we misunderstandit at first is we're not used
to dealing with a technologythat can do these things, but
(06:10):
we're also expecting the rightbrain to do things that the left
brain can do.
Right, we're expecting it to beexact and precise.
It's like guess what?
It doesn't.
It actually knows nothing, AI,it only knows patterns.
It doesn't know any actualknowledge.
That's what's stored in adatabase and so actually, when
you're communicating with it,you got to be able to connect
the right and the left braintogether the right way and this
(06:32):
is a very long-winded way aboutsaying why I got into this, but
starting to realize that Irealized that generative AI just
needed structure, and if yougive it the right structure and
you break it up enough, then youcan get really, really accurate
responses and results for whatyou want.
And so, but as my thought, myAmazon brain was kicking in of
like, well, how does this scale?
(06:52):
I realized quickly I was goingto have more ideas and more
things that I would want tobuild than any engineering team
could ever build for me, and sowhat I needed was a whole layer
that was like a business, userfriendly layer to be able to
configure AI prompts andresponses talking to my Excel
(07:14):
sheets and sending reports andconnecting to the internet,
various websites, etc.
And I would need a kind of alayer that I could just
configure all this and create asystem that would do accomplish
whatever goal it would be andthat would take an engineer I
could write out the the detailsand take an engineering team
time to do that but I realizedthere's way too many of these
(07:34):
that I would want to do and AIcould do them so quickly.
So I built, I started enhancedAI just all on my own and built
my first prototype myself, andthen it started growing because
it's this whole platform, it'sthis Flux Prompt platform, it's
what we call the platform insideof Enhanced AI, and Flux Prompt
is where you know, Flux pullsthings, pulls it all together
(07:56):
into one place and you'reprompting these various
different components and it putsit in one place and makes you
be able to orchestrate that'sthe best word for it Like you're
the conductor of the orchestra,right?
You're orchestrating these AIagents to do, or these AI calls
to do all sorts of variousthings and tasks and, at the end
of the day, I can now build.
I could build five or 10 thingsin a day, versus, like what I
(08:20):
could have done before with anengineering team would have
taken me, you know, five timesas long to accomplish that, and
so that's the ultimate reason Ijust want to build more and
build more faster, and that'sreally why we started it.
Speaker 2 (08:32):
Yeah, that's amazing because you thought like, if I
have this and I can create thisfor myself, other people
probably have the same thing.
And that's an area, that whereyou and I really connected at AI
Mavericks Conference, because Ihave a million and one things,
I'm doing a million and oneideas, but I'm one person and
finding the right people toscale with you can be difficult.
And, yes, we still need to hirepeople.
Speaker 3 (09:11):
If we can use technology better to do a lot of
the tasks, management and theideation with us and be that
thought partner, then that takes.
I mean, our time to market isso much faster.
Yeah, it also.
It improves the quality.
At the end of the day, when yousay, find the right people, you
find the people.
And now, instead of your timebeing spent on saying, you know,
rewording a paragraph or asentence and now maybe you still
need to do that in some casesbut instead of spending a lot of
time on doing that over andover again, you're spending a
lot of time on just a littlehigher level, thinking of like
well, should we even be talkingabout this paragraph?
Like maybe we should throw thisout, right, and you know
(09:34):
there's people who do that nowin layers, but you can
consolidate to one person.
So I think it's actually thequality of the work actually
improves.
Also with what we're doing, andpeople have said this before
right, we're all.
Ai is kind of up-levelingeverything in society and right
now it's a shake-up and so it'sa little bit nerve-wracking, but
it's the same.
As you know, moving, as peoplehave said, moving from the plow
(09:56):
into the tractor, right, andwhere you're.
Everyone can do more higherlevel work and move on.
And once you learn to use thesesystems, that's actually the
best part is, once you learn todo these, it's not engineer only
, and that's the best part,right, you don't have to be
actually an engineer to bebuilding these things, and you
can build amazing, fascinatingsystems that accomplish things
(10:16):
that no one's ever thought ofbefore.
It's such an exciting time tome.
Speaker 2 (10:20):
Yeah, so what prompted you to have these
classes, these courses thatpeople can take?
And because a lot of peopleoffer a product to the market,
people can buy the subscription,they can maybe look at the
tutorials and they'll use yourproduct and it's fine.
But you actually are investingin people's capacity to utilize
your tools to the fullestadvantage.
Speaker 3 (10:43):
Yeah, yeah, what inspired.
So I think about nine monthsago we would get a lot of
questions like, oh, that's soamazing, I just don't know if I
can do it, or like I just don'tknow if I can, and realizing
that people were just feelingthey weren't feeling empowered
the way they should.
And for me, I'm just kind of adetermined entrepreneur, so I
(11:03):
just like figure it out Right.
But and I'm fortunate that I'vehad been able to have time to
do that A lot of people arereally busy.
They have a lot going, a lotgoing on, a lot of stresses,
pressures from every parts oflife, and so what?
What we thought was like hey,let's try, just as an experiment
, to teach business executivesand business folks how to use or
build the agents on their ownand make it kind of low impact
(11:27):
where you could do it once a.
You know, the current course isset up where it's once a week
for one hour, so it's easy to do, for anyone can spare an hour
and we got some homework, so wemake sure people are learning
and doing things inside theclass and that way it's, it
becomes a less, moreapproachable thing for the very
busy person who actually reallywants to get into this and maybe
(11:47):
a little scared to get in,doesn't know where to start, and
we've had some people who are,who are definitely scared to
start.
They've been consultants or hadother jobs in various spots,
never deep in technology.
There was always a differenttech team.
There was always a differenttech team.
They were never on that.
But they jumped in and nowthey're building chatbots and
they're building sophisticated,automated forms that are doing
(12:09):
all sorts of things with theseagents and they realize, hey,
it's actually this technology,it's so cool because it's so
approachable and all you have todo is you type a prompt and now
you have like a new logicsystem, like, hey, it's going to
do whatever your prompt saysevery time where that would have
taken so long in engineeringand coding and learning a
language before.
So we found it was a lot ofpeople were just weren't feeling
(12:31):
as empowered as they should be,as we felt they should be.
So we decided to start our owncourse to do it and we cover
topics of AI, right, fundamentalfoundational principles, and
then we've still just found it'seasiest to build these agents
through our platform, and sowe're showing people how to do
it and it's the same principlesyou'd use if you were coding
(12:52):
something.
To be honest, if someone youknow was going to engineer the
exact same agent, they wouldmaybe instead put it up doing it
in our platform with the nodebase kind of setup.
They would maybe do it in code,but it's the same principles of
well, I got to call this agentnow and then call these two
systems and then call this agentwith this information, but it's
just so much faster and quickerand easier to do in the
platform which it's designed tobe that way.
Speaker 2 (13:13):
Yeah, it's been great , and this program, the MS in
Digital Media Management.
It's part of Annenberg Schoolfor Communication and Journalism
, so you know we're talking alot about marketing and tech
stacks and ad tech and video anddifferent components of the
digital landscape, but none ofthose exist now without some
(13:36):
element of AI.
Even on the public relationsside, I've used tools for better
pitching, better matchingprofiles for articles or podcast
interviews, things like that.
I've been able to see how youcan create better brand, deeper
brand intentions and they'llhave better personas and even
(13:56):
the research that you can get onyour potential audiences.
But this is actually the nextstep, the next phase, and I
would argue that it doesn'tmatter what business you're in.
You still need to understandthis technology, this side of
the landscape, and we weretalking before we actually
pushed for cord about some ofthe needs and I'm like I just
(14:19):
want bots to do this and that.
So I need a whole team ofagents to be able to be better
at the job that I do, to be thatstrategic, creative brain, the
one who's actually speaking withclients, consulting with them,
coming up with their strategiesand implementing and then using
this technology to help meimplement all the different
(14:41):
components and go out and dothat research or find out if
this is the right person to beon the podcast or not.
Speaker 3 (14:48):
Right, yeah, yeah, entirely so it's.
You know, it's funny.
One of our earlier users isactually a very senior brand
manager inside of Kimberly ClarkWow, and she was not
necessarily tech savvy, eventhough she's brilliant in almost
anything she does.
But we found actually this AIyou know kind of in the in the
platform.
It's though she's brilliant inalmost anything she does, but we
found actually this AI you knowkind of in the in the platform.
(15:08):
It's like she's like oh, thisplatform looks like works a lot
like my brain does, because youcan kind of put like a bunch of
different you know like chatsaround.
You're like, oh, I'm just gonnaexperiment with these, like
different prompts and things,and it's all there, exposed
right in front of you.
And then it's like, okay, thenI'll create this automation over
here.
But she was creating, you knowshe was doing some amazing
things like reverse engineeringa brand profile based on
(15:30):
packaging of competitors, right,so she would load up a
competitor's package and reverseengineer who their brand
persona and profile was and thenfigure out where their brand
persona and profile should fitinside, like next to that one.
And it's like, whoa, like Idon't think you would even do
that, like without AI technologylike it's, it's actually
(15:51):
enabling us to do things that weshould be doing, that we know
we should be doing, that we'renot because we don't have time
to do it Right, and and thoseare like really exciting things.
And but it she just kept goinglike into like writing scripts
that you know, consulting firmswere trying to do but didn't
quite get the.
They wouldn't quite nail theirhead on the scripts for these
commercials.
And then she, we draftedseveral.
(16:13):
She drafted several scriptsinside of the system and like
nailed it.
It was like 90% there when theother firm was, you know, doing
just 50% of where they wanted toget to.
And then she took that and gaveit to them and they were able
to, you know, adjust it andimprove their work.
But it's like the swath ofthings that you can do we found
(16:35):
that there's about 30% of theusage in our platform is not
even automation.
It's just like ideation andcreating these one-off, you know
, like tests and trials and justplaying with all the AI systems
in one place.
Like we didn't even say thatthat our platform is integrated
150 different AI platforms inone place and systems, so it's
easy just to try them alltogether.
It's like, well, I want tochain Anthropic with OpenAI,
(16:55):
with Gemini and then do imagegenerator between all these
different systems and audio andeverything.
So it's kind of a nice as aplayground to start playing
around and actually ideate.
And then we found then there'sanother 70% of the users or the
usage.
That's like building an agentthat does automations and
systems for you that just makeyour whole life easier, because
(17:17):
you know that's the automationyou want and that's the agent
you need.
And so you build it out and itjust repeats it either every day
or waits for your input, orwhatever it might be.
Speaker 2 (17:27):
Yeah, okay, so I realized we should probably
backtrack, because a lot of ouraudience knows what AI is.
They probably don't know.
They think about AI in terms ofgenerative AI, right?
Not in terms of the big AI andML and DL and all that stuff.
So AI agents might not be aconcept people are familiar with
listening because I know a lotof the students in the program.
Some have been told they can'tuse AI in their businesses or in
(17:52):
their undergrad courses and I'mgoing no, you have to use it in
everything you do, almost right.
So can you talk about agents?
Because this is something I'vebeen really excited about,
because I do want to be able toleave a memo to my personal
agent and say I need a book.
I keep forgetting to book dentalappointments for me and my
(18:13):
daughter.
Can you do that?
Can you schedule the dog's nextvisit to the vet, right?
Can you remember to order this?
Can you order this medicine forme?
Go into my profile.
And because I keep forgettingthat one of my dogs needs more
gallop, rant or whatever.
You know those things that takeup time and that if we're so
busy, I didn't iterating andworking and then also trying to
(18:34):
spend time with our families.
Sometimes those things fall bythe wayside.
Speaker 3 (18:38):
Yeah, yeah, they definitely do.
Yeah, I think to answer kind ofthe base, foundational question
and this is this is such afascinating time.
I've seen it probably threetimes in my career now.
Every time a new kind of bigwave happens, all these terms
start getting thrown out thereand they slowly coalesce into
like what is the actual term.
So the term agent has meantdifferent things to different
(19:02):
people all along, but I thinkwe're actually getting closer to
a coalescing of what an AIagent is, and I think it's my
opinion.
I've seen it many other peopleand again, it's still coalescing
.
But is any basic, like the mostbasic agent is any system that
is including an LLM, some kindof large language model, in its
call, in its process.
(19:23):
That's like the most basic formof an agent in my mind.
So it could be.
You know, I have a tech systemthat has to hit an LLM call to
do something particular and thensends back some information to
me in a different structure,like in.
Fundamentally, that could becalled an agent.
Now, some people would be likethat's not smart enough to be an
agent.
I get that argument too, but tome it's like you don't
(19:47):
necessarily know exactly whatthey're calling.
Anyway, that's my initialimpression for the low bar.
Speaker 2 (19:52):
Yeah, like you go onto a website and you have a
customer service bot, right?
Yeah, is that an agent?
Speaker 3 (19:58):
Yeah, it probably is, because it's using an LLM to
reference its memory and itssystem.
And so I think the next levelup there's probably degrees
right, like a really base agent.
And then the more advancedagents, like, well, an agent
that can actually call and useAI and LLM to call multiple
different systems to likeaccomplish an objective and a
goal, like that's probablyanother like the mid tier, like
(20:21):
level of defining an agent and Ithink.
And then the top tier is likean orchestration, where you have
like one agent calling multipleagents and each one of them
have their own kind of tasks,and that's like a larger kind of
agents system that I would say.
So those are probably the three.
I would three ways I define it.
Speaker 2 (20:38):
Yeah, I was listening to another podcast that was
talking about that.
In this form concept is likeyou may have an agent that acts
as a CMO, but it's probably andgoes into a meeting for you, but
that agent probably isn't justone agent, it's a whole series
of agents.
Each one knows there's like anexpert in this one thing.
Yeah Right, so instead ofhaving to have a whole bunch of
(20:59):
human experts, you can have that.
So where does you know what'sthe play between humans and AI?
Because that's a questionpeople get really scared about.
I always try to emphasize no,no, no.
It's not something to be scaredof.
We need to be able to utilizethis technology, but we still
need our brains, because AI justknows what everybody's input
(21:21):
into the internet, and that canbe good, bad, average
everybody's input into theinternet and that can be good,
bad, average, right, yeah, andso I think we're in a.
Speaker 3 (21:30):
Really my opinion we're in a fascinating place.
Maybe I'll go a little too deepon this one, but and you can,
you can pull me back out.
I think we're in a fascinatingplace because a lot of people
expect like, hey, can ai, canyou go do all of these various
things?
For me now, a generative ai islike it's coming.
All it is really doing is it'strained on trillions of patterns
and it's finding a completionof those patterns in text, in
(21:51):
particular, or images, and sowhat it's doing is trying to
complete a pattern based on yourinput.
And so, like I said, it doesn'tknow anything, but it knows.
George Washington was presidentbecause it was told that 10,000
times through the patterns,that's the pattern it knows.
To finish, it actually doesn'tknow it.
So when you think about agents,we start getting into this
(22:14):
multi-orchestration.
Where does the agent know thepattern of how to get this your
dog food Like?
It hasn't.
There's not a million.
It wasn't trained a milliontimes on getting dog food right.
It's because these are pathsthat we're creating, sorry if
I'm getting a little deep here.
It's like when you think aboutan agent and this is actually
why I think it breaks down,still like it's not great yet is
because it hasn't been trainedon all these paths to like, get
(22:36):
dog food.
So you actually have to be veryspecific about how you train
that agent to do exactly whatyou want it to do, because if
you don't, it will break.
And so it because it doesn'tknow, it doesn't have enough
data, it hasn't been trainedenough on how to do it, and so
there's a lot of this kind ofwork on like hey, can this agent
do everything for me?
And it just doesn't.
It's not trained, there's notenough patterns for it to know
(22:57):
how to do these things.
Now they're working on thingslike MCP is a potential you know
solution for that.
That's more of just likeallowing them to talk better
together.
But I think like what we producejust as an end result, like we
produce like 99 or 100% accuratedocumentation through all
through AI systems.
So people put in like fivedifferent documents about a
(23:18):
medical patient.
It reads it all, it breaks itall down and spits out a perfect
report that it looks the exactsame, formatting everything
every time with the new patientinformation.
Now, like no LM can do that.
The way we do that is westructure exactly where the you
have to structure where the datagoes and how the LM is going to
digest the data and spits itout.
(23:39):
So, like you have to createthese, I guess, guardrails to
kind of make it look exactly theway it should look, and I see a
lot of these larger agents andorchestrations of agents.
The same way is that you haveto create these guardrails of
like what it does.
So in that CMO in the meeting,yes, you theoretically and you
probably will have all theselike micro CMO agents that are
doing various things, butthey've been told exactly what
(23:59):
they should be doing.
And like that's my job, that'smy only job, I'm just going to
go do this, always look for the,you know the opportunity in the
room and then find you knowsomething to do with that
opportunity.
Like that's that whole agent'sjob.
So like though that's how youwould break these down.
And that's why they talk aboutswarms of agents, because AI
can't necessarily create allthese and do it very accurately.
But if you're doing theseindividually and you're making
(24:21):
them, you know it can be 99 or100% accurate with the kind of
way that you're building theseagents and then you can use that
forever, right?
That agent is just really goodat what it does.
And then you have another agentthat can call it when it needs
to, and it kind of knows its wayof calling it.
Speaker 2 (24:36):
Now, that was like sorry that was probably way too
broad of a term, Because I thinkpeople need to understand this
because, even if they're notusing it now, these are concepts
that are being talked about andeven if they're not mainstream,
they will be.
Yeah, so you know, we wanteverybody to have as much data
as possible and have theunderstanding of what all these
(24:58):
concepts mean.
Another concern that peoplehave is privacy and security.
That was actually the classthat we had on Wednesday of this
week was about GDPR and CPRAand different like those kind of
consumer privacy restrictionsand laws, but then I had all the
students also add in the AIcomponent which states and
countries have AI regulations,which ones don't?
(25:21):
What do those regulations mean?
Right?
And so I'd love to hear alittle bit, because that's a big
question people have, andthat's why I use certain
products more than others,Because you know, and so if
somebody is building stuffwithin Flux Prompt, you know
it's all secure.
Speaker 3 (25:38):
Yeah, yeah, yeah, it is secure.
Well, actually, we leave itopen to the user.
I mean, it's secure inside theplatform but the API the
platform but the API.
So, as of now, the recording ofthis the most secure legal
documentation is in bothAnthropic and OpenAI, from what
we've reviewed in terms of, like, what they do with your data,
and that's particularly now.
Don't get that confused.
(25:58):
If you put stuff in ChatGPT,it's like everywhere, so don't
put stuff, anything in ChatGPT.
Well, there are some like tiersand then there's some loose
legal documentation around likethe team tiers Enterprise.
It actually gets to much betterlegal documentation.
But if you're using API callsinto these systems, then there's
actually like really strictsecurity measures around, like
(26:21):
what you're sending in andthere's really clear guidelines
of like what is shared andwhat's trained on and what's not
trained on.
And, as of this, you know, asof last time we looked at, which
was, I think, earlier this week, those are the two most secure
systems and all of our calls.
If you're calling in athird-party LLM, like we said,
(26:42):
we have about 150 in our system.
It's all through APIconnections to these other
systems and so we're looking atall the license agreements
between those API between thosecalls and how secure that data
is, and we've been trying toexpose that to people like, hey,
these ones are, you know, a lotmore secure than the other ones
, and so if you want to besecure, that's one way.
(27:03):
There's more ways of going aboutit.
One of the reasons Facebook'smodels have been big is the Lama
models and then DeepSeek isbecause they open source their
models, which means I can takeit and put it on my own server,
which means it will never gettrained on anything because it's
on my server.
I control what's gettingtrained, and so in those
scenarios that's 100% secure ifyou can put one of those models
(27:25):
in your old system, and there'smany companies who've started
building stuff on top of thosemodels.
That's the reason open sourceis good, because someone can
take it and make it securearound their own systems.
So that's like the extra secureway of doing the LLM calling
Calling into a third partysystem like OpenAI and Anthropic
.
Those two are the most securefrom a legal perspective.
(27:45):
Now, yeah, anyway, that's.
Speaker 2 (27:48):
No, but that's, I think, important things for
people to think about, right?
So for students who arelistening, where should they
start?
Maybe they've tried.
They've used Grammarly or Chator Cloud or name your favorite L
, that you know the big ones.
Maybe they've done that.
Maybe they've tried out acouple of other tools.
But what else do they need toknow right now and how do they
(28:11):
move forward in this AI firstworld?
Speaker 3 (28:14):
I would say, first of all, it's amazing that they're
students.
That's really good, because AIwhat it is at its core.
It doesn't know anything, butit's a pattern machine.
It loves, loves, loves patternsthing but it's a pattern
machine.
It loves, loves, loves patterns.
So if you know patterns andframeworks and theories, which
is taught in schools, thenyou're going to have a leg up,
(28:34):
particularly if you know how tocreate patterns and theories.
And now the kind of next levelis using AI to create patterns
and theories, which is actuallyreally fun, to be honest.
That's how part of the thing Ido with my board game that I
created is creating an ecosystemthat works with certain
foundational principles.
But the idea is that, if youknow, really lean heavier into
(28:57):
the idea of creating a framework.
Or when I say framework, I meanlike hey, this marketing letter
should have this kind of atitle and should be this long,
and then this kind of an openingparagraph should be this long.
This, then this kind of anopening paragraph, should be
this long, this mid body, andthen this, this conclusion,
right, like general marketing islike hey, call out the pain and
solution, right, so like that'sa framework pain, solution,
framework, right, or it'susually part of a little broader
(29:18):
one, but knowing thoseframeworks, and then you can
tell AI hey, I want you tocreate this based on this,
follow this framework, and thenyou give it your framework and
you can make it your own.
You can customize someoneelse's a little bit to be, or a
well-known one a little bit tobe your flavor, and then it's
actually going to be able tocreate really good content that
you, that comes almost from youand your guidelines and your
(29:40):
frameworks.
And that's kind of the worldwe're moving into.
Is this world of everyone'skind of the conductor, the
orchestrator?
You're giving the AI the tasksof what to do and if you give it
the right tasks of what to do,it'll do a good job.
If you give it the wrong tasks,you'll still do a poor job.
Right, you're not even givingit the right tasks to do so.
I would actually encouragestudents to think a little bit
more meta not Facebook, butthink a little more meta in the
(30:03):
concept of dealing with AI andhow to create frameworks and
structures in your prompts andhow you're prompting it and then
even using it to help youcreate those Miveta prompts and
those higher level prompts.
But to the actual toolsthemselves.
I mean, I would say, definitelytry Flux Prompt.
Come and try it, hit us up atany point, we can help you.
This is designed for thebusiness person we is.
(30:25):
We almost picture this likeExcel.
I don't know if you rememberthe first time you opened Excel,
but it's got formulas, it's gotlike references, all sorts of
things, but then once you learnit, you're like, oh, this is
actually pretty easy and you getto use it.
It's very similar.
We found very similarexperiences.
Folks come in, don't quiteunderstand it at first and they
start learning like, oh wait,this is super easy.
(30:47):
And they start creating theseincredibly powerful AI agents
and systems, and so I would say,definitely try our system out.
There's several other systemsthat you can try in terms of
agent building, orchestration,but ours is our favorite because
it's easiest to use in our mind.
But it's OK, we're biased.
Speaker 2 (31:04):
I have two questions for you.
One is when you're on a run andyou're speaking to AI, what
folder are you speaking to?
Speaker 3 (31:14):
That's a really good question.
I've used Perplexity, and thereason is I'll give you my two
or three reasons.
Perplexity is internet, first,and the other ones have been
slow to catch up, so it can usethe references of the internet
when I need it to.
Number two is I could switchthe model I use on it.
I can use Claude, I can useOpenAI, I can use Perplexity's
(31:35):
own model, and I still loveClaude the best, and so I use
Perplexity with Claude as themodel that responds to me.
And then, third, is that it hadthe easiest interface for me to
talk and run at the same time.
Okay, but, and I'll tell youthis, it's a weird trick is that
I don't ever press like recordmy audio on the actual app,
(31:56):
because on any of them, becauseit like tries to automatically
stop at the wrong time and likeit waits till I pause and I'm
like no, I'm still running, I'mjust breathing heavily, like I
want you to keep recording me.
And so I actually use the textto speech on my phone, natively
on my phone, and I use that torecord my audio and I may have
like five minutes of talking orwhatever.
(32:16):
And then when I'm finally donewith my concepts.
I press stop and then itconverts it all into text, and
then I say go and then itcreates the whole thing.
And it creates the whole thingand then press the play audio In
perplexity it's been theeasiest for me to use and it
plays the audio back.
So then I listen to theresponse for the next five
minutes.
So within, like you know,within one mile, I'm listening
to, like you know, maybe acouple of responses back and
(32:38):
forth.
Speaker 2 (32:39):
Yeah, between me and Bob, into it Whenever I'm
walking my dogs is when Iremember something that I need
to do, and then, yeah, I caneither text myself or voice memo
, but I'd rather, if it'ssomething more complex,
obviously I'd rather just talkto my AI.
Speaker 3 (32:55):
Yeah, yeah, yeah.
Speaker 2 (32:56):
Fantastic and I understand we also are going to
have a special link discount forlisteners if they want to sign
up for the class that I'm inright now I mean, obviously it
would be different cohort andlearn how to build their own
agents.
Speaker 3 (33:09):
Yes, yeah, we're providing a special discount
just for listeners here.
We love what Anika does and thepodcast, so and again, the
class is designed to be a cohort, so you actually work with
other professionals around youand you get to know them.
There's actually a lot of greatrelationships that people build
, but it's everyone's learninghow to build agents in their own
(33:32):
AI agents and, believe me,we're coming to a world where
there's going to be 1000s ofagents, and that's okay, because
half of them, you know, you'llhave your own several dozen
agents or a couple 100 agents atsome point, and you want to be
able to configure them on yourown, in our view, and be able to
do exactly what you want.
And this is kind of the verybeginning of getting your feet
wet.
We've had people build amazingthings just coming out of the
cohort.
They've created businesses withjust what they've built from
(33:55):
the class of.
They've learned how to buildand we kind of run through
everything.
So we would love yeah, lovepeople to come and check it out.
Speaker 2 (34:02):
Fantastic, brad.
Thank you so much.
It's always fun to speak withyou and we can talk for so long
about all the different ideas wehave and how to implement
different technologies andstrategies.
So I look forward to continuingthe conversation.
And, yeah, thank you forjoining us on Mediascape.
Speaker 3 (34:19):
Yeah, thank you for having me.
It's a pleasure being here,absolutely.
Speaker 1 (34:22):
To learn more about the Master of Science in Digital
Media Management program, visitus on the web at dmmuscedu.
Welcome to Mediascape insights from digital
changemakers, a speaker seriesand podcast brought to you by
USC Annenberg's Digital MediaManagement Program.
Join us as we unlock thesecrets to success in an
increasingly digital world.
Speaker 2 (00:22):
We just had an entire podcast interview before we
jumped on Brad Moss.
You have done so much and youcontinue to do so much, and I'm
so thrilled to have you onMediascape Insights from Digital
Changemakers today.
Speaker 3 (00:36):
Oh, thank you, Annika .
I'm thrilled to be here,thrilled to be here.
Speaker 2 (00:39):
We didn't even realize we had this video game
connection.
I used to work in video gamemagazines and launched the
official Xbox magazine when thatcame out.
Speaker 3 (00:47):
And you have you've created how many Over 60 video
games in my career.
Speaker 2 (00:52):
Okay, amazing, but we're really here to talk about
AI that does tie in, so yeah itdoes.
Speaker 3 (00:59):
I can tell you a little bit of the story of that.
That's actually how I got intoAI.
Speaker 2 (01:03):
Let's do it.
Yeah, please share.
Speaker 3 (01:10):
Yeah, so I mean my background.
After video games I went to bigtech and I worked inside of
Amazon.
I ran the third party sellingplatform the third party
marketplace platform thateveryone uses to sell on Amazon
and then built a new businessfor Amazon.
Then I left after a while aftermaking lots of money for Amazon
, their shareholders, and then Iwent.
I'm a little bit of anentrepreneur and I went and
built some e-commerce companiesand whatnot, but fast forward
like seven years.
(01:31):
I love video games and I'dbuilt a lot of games.
But the game business is reallytough.
It's really brutal and it evenis right now.
There's a lot of layoffs withsome of a lot of my friends
who've been hit by them, butit's.
The process of building games isjust like so cumbersome and
we've seen it from inside theindustry too.
So about three and a half yearsago I got into AI.
Before ChatGPT like exploded,it was GPT-3.
(01:54):
We even started hitting two atthe time from time to time.
But when I discovered that itexisted generative AI I mean
I've been coding regular AI invideo games.
You know, 20 years ago Irealized this generative AI was
like really powerful and that Icould actually accomplish a lot
of things and the whole process.
The whole goal was tostreamline operations of game
(02:15):
development, of like, hey, howdo you burn through iteration
cycles and make sure this is fun?
Because games is all aboutfinding the fun and you got to
go through iteration afteriteration until you finally find
it and you refine it more andmore and more.
And I was trying to automatesome of that process and
discover generative AI, and sothat's how I stepped into.
The whole world is trying tojust refine and make game
(02:37):
industry, the game business,work better, because making
games is super fun.
Speaker 2 (02:42):
Yeah, well, and you just used AI to help you create
a new game.
You want to talk?
Speaker 3 (02:48):
a little about that.
Yeah, this is kind of a randomthing so I could grab it.
I've been working on a boardgame idea I've had for about six
years and it's all about thenot nearly what this whole
podcast is going to be about,but this board game idea I've
had for about six years and Ithink the point of it is that AI
is just empowering and I've hadthis idea for a long time.
(03:08):
I prototyped it with my sonwith old card deck of cards, but
then over time, you know, Istarted using AI and I generated
like these, you know this, thiswhole card game.
You can't really see it with my, with the blur on it
(03:44):
no-transcript, don't go on thatdirection, go in this direction.
But it's kind of this thoughtcompanionship that we do or that
I've been doing both that likewriting executive briefs to my
board, you know, and to myinvestors.
Using AI is almost a companionand a lot of things to get out
of my mind, things I have but Ineed to kind of pull them out a
(04:05):
little bit more.
Speaker 2 (04:07):
You are preaching to the choir.
This isn't AI related.
But a couple of years ago I waswriting a book chapter for a
book about women and businessand business on purpose and
having really purposeful lives,and I was having writer's block,
not because I didn't know whatI wanted to say, but because I
don't have time to sit down andtype it all out.
So I started just voice memosto myself.
(04:30):
Now this is before Gen AI iswhat it is today, so I don't
know if that was an option, butnow that's absolutely something
I would do, and I speak toClaude every single day as my
thought companion my favoriteone, right now at least and so
it's really exciting to see howwe can really utilize this
technology in the best examples.
(04:50):
Right, it's not about writing apaper for us, it is about
taking our thoughts, ouroriginal thoughts, and putting
them in order and helping usthink through what are we
missing.
So I really love that, andthat's where Enhanced AI really
comes in.
I'm in your agent buildingclass right now, just started
this week really exciting and wemet at an AI conference earlier
(05:14):
this year, so I'd love to talkabout that.
So you discovered AI and thenwhy did you decide to create
this brand, this company.
Speaker 3 (05:24):
Yeah.
So when I discovered AI and thepower of it, I realized pretty
quickly so you got to rememberthis is before ChatGPT existed.
This genre of technology wasreally fascinating, but to me it
could do, and I actually bringthis up in the class.
I consider, and I came to thisconclusion after about a year of
thinking through it and workingwith it.
Ai feels like the right brainof technology where we're used
(05:48):
to dealing with the left brainof technology.
Right, like technology, is veryprecise, exact.
It's like a database.
You can get exact things out.
That's very much like the leftbrain.
Ai is like the right brain whereit's really creative and it's
generative and it's inspiring,where it's really creative and
it's generative and it'sinspiring.
And one reason we misunderstandit at first is we're not used
to dealing with a technologythat can do these things, but
(06:10):
we're also expecting the rightbrain to do things that the left
brain can do.
Right, we're expecting it to beexact and precise.
It's like guess what?
It doesn't.
It actually knows nothing, AI,it only knows patterns.
It doesn't know any actualknowledge.
That's what's stored in adatabase and so actually, when
you're communicating with it,you got to be able to connect
the right and the left braintogether the right way and this
(06:32):
is a very long-winded way aboutsaying why I got into this, but
starting to realize that Irealized that generative AI just
needed structure, and if yougive it the right structure and
you break it up enough, then youcan get really, really accurate
responses and results for whatyou want.
And so, but as my thought, myAmazon brain was kicking in of
like, well, how does this scale?
(06:52):
I realized quickly I was goingto have more ideas and more
things that I would want tobuild than any engineering team
could ever build for me, and sowhat I needed was a whole layer
that was like a business, userfriendly layer to be able to
configure AI prompts andresponses talking to my Excel
(07:14):
sheets and sending reports andconnecting to the internet,
various websites, etc.
And I would need a kind of alayer that I could just
configure all this and create asystem that would do accomplish
whatever goal it would be andthat would take an engineer I
could write out the the detailsand take an engineering team
time to do that but I realizedthere's way too many of these
(07:34):
that I would want to do and AIcould do them so quickly.
So I built, I started enhancedAI just all on my own and built
my first prototype myself, andthen it started growing because
it's this whole platform, it'sthis Flux Prompt platform, it's
what we call the platform insideof Enhanced AI, and Flux Prompt
is where you know, Flux pullsthings, pulls it all together
(07:56):
into one place and you'reprompting these various
different components and it putsit in one place and makes you
be able to orchestrate that'sthe best word for it Like you're
the conductor of the orchestra,right?
You're orchestrating these AIagents to do, or these AI calls
to do all sorts of variousthings and tasks and, at the end
of the day, I can now build.
I could build five or 10 thingsin a day, versus, like what I
(08:20):
could have done before with anengineering team would have
taken me, you know, five timesas long to accomplish that, and
so that's the ultimate reason Ijust want to build more and
build more faster, and that'sreally why we started it.
Speaker 2 (08:32):
Yeah, that's amazing because you thought like, if I
have this and I can create thisfor myself, other people
probably have the same thing.
And that's an area, that whereyou and I really connected at AI
Mavericks Conference, because Ihave a million and one things,
I'm doing a million and oneideas, but I'm one person and
finding the right people toscale with you can be difficult.
And, yes, we still need to hirepeople.
Speaker 3 (09:11):
If we can use technology better to do a lot of
the tasks, management and theideation with us and be that
thought partner, then that takes.
I mean, our time to market isso much faster.
Yeah, it also.
It improves the quality.
At the end of the day, when yousay, find the right people, you
find the people.
And now, instead of your timebeing spent on saying, you know,
rewording a paragraph or asentence and now maybe you still
need to do that in some casesbut instead of spending a lot of
time on doing that over andover again, you're spending a
lot of time on just a littlehigher level, thinking of like
well, should we even be talkingabout this paragraph?
Like maybe we should throw thisout, right, and you know
(09:34):
there's people who do that nowin layers, but you can
consolidate to one person.
So I think it's actually thequality of the work actually
improves.
Also with what we're doing, andpeople have said this before
right, we're all.
Ai is kind of up-levelingeverything in society and right
now it's a shake-up and so it'sa little bit nerve-wracking, but
it's the same.
As you know, moving, as peoplehave said, moving from the plow
(09:56):
into the tractor, right, andwhere you're.
Everyone can do more higherlevel work and move on.
And once you learn to use thesesystems, that's actually the
best part is, once you learn todo these, it's not engineer only
, and that's the best part,right, you don't have to be
actually an engineer to bebuilding these things, and you
can build amazing, fascinatingsystems that accomplish things
(10:16):
that no one's ever thought ofbefore.
It's such an exciting time tome.
Speaker 2 (10:20):
Yeah, so what prompted you to have these
classes, these courses thatpeople can take?
And because a lot of peopleoffer a product to the market,
people can buy the subscription,they can maybe look at the
tutorials and they'll use yourproduct and it's fine.
But you actually are investingin people's capacity to utilize
your tools to the fullestadvantage.
Speaker 3 (10:43):
Yeah, yeah, what inspired.
So I think about nine monthsago we would get a lot of
questions like, oh, that's soamazing, I just don't know if I
can do it, or like I just don'tknow if I can, and realizing
that people were just feelingthey weren't feeling empowered
the way they should.
And for me, I'm just kind of adetermined entrepreneur, so I
(11:03):
just like figure it out Right.
But and I'm fortunate that I'vehad been able to have time to
do that A lot of people arereally busy.
They have a lot going, a lotgoing on, a lot of stresses,
pressures from every parts oflife, and so what?
What we thought was like hey,let's try, just as an experiment
, to teach business executivesand business folks how to use or
build the agents on their ownand make it kind of low impact
(11:27):
where you could do it once a.
You know, the current course isset up where it's once a week
for one hour, so it's easy to do, for anyone can spare an hour
and we got some homework, so wemake sure people are learning
and doing things inside theclass and that way it's, it
becomes a less, moreapproachable thing for the very
busy person who actually reallywants to get into this and maybe
(11:47):
a little scared to get in,doesn't know where to start, and
we've had some people who are,who are definitely scared to
start.
They've been consultants or hadother jobs in various spots,
never deep in technology.
There was always a differenttech team.
There was always a differenttech team.
They were never on that.
But they jumped in and nowthey're building chatbots and
they're building sophisticated,automated forms that are doing
(12:09):
all sorts of things with theseagents and they realize, hey,
it's actually this technology,it's so cool because it's so
approachable and all you have todo is you type a prompt and now
you have like a new logicsystem, like, hey, it's going to
do whatever your prompt saysevery time where that would have
taken so long in engineeringand coding and learning a
language before.
So we found it was a lot ofpeople were just weren't feeling
(12:31):
as empowered as they should be,as we felt they should be.
So we decided to start our owncourse to do it and we cover
topics of AI, right, fundamentalfoundational principles, and
then we've still just found it'seasiest to build these agents
through our platform, and sowe're showing people how to do
it and it's the same principlesyou'd use if you were coding
(12:52):
something.
To be honest, if someone youknow was going to engineer the
exact same agent, they wouldmaybe instead put it up doing it
in our platform with the nodebase kind of setup.
They would maybe do it in code,but it's the same principles of
well, I got to call this agentnow and then call these two
systems and then call this agentwith this information, but it's
just so much faster and quickerand easier to do in the
platform which it's designed tobe that way.
Speaker 2 (13:13):
Yeah, it's been great , and this program, the MS in
Digital Media Management.
It's part of Annenberg Schoolfor Communication and Journalism
, so you know we're talking alot about marketing and tech
stacks and ad tech and video anddifferent components of the
digital landscape, but none ofthose exist now without some
(13:36):
element of AI.
Even on the public relationsside, I've used tools for better
pitching, better matchingprofiles for articles or podcast
interviews, things like that.
I've been able to see how youcan create better brand, deeper
brand intentions and they'llhave better personas and even
(13:56):
the research that you can get onyour potential audiences.
But this is actually the nextstep, the next phase, and I
would argue that it doesn'tmatter what business you're in.
You still need to understandthis technology, this side of
the landscape, and we weretalking before we actually
pushed for cord about some ofthe needs and I'm like I just
(14:19):
want bots to do this and that.
So I need a whole team ofagents to be able to be better
at the job that I do, to be thatstrategic, creative brain, the
one who's actually speaking withclients, consulting with them,
coming up with their strategiesand implementing and then using
this technology to help meimplement all the different
(14:41):
components and go out and dothat research or find out if
this is the right person to beon the podcast or not.
Speaker 3 (14:48):
Right, yeah, yeah, entirely so it's.
You know, it's funny.
One of our earlier users isactually a very senior brand
manager inside of Kimberly ClarkWow, and she was not
necessarily tech savvy, eventhough she's brilliant in almost
anything she does.
But we found actually this AIyou know kind of in the in the
platform.
It's though she's brilliant inalmost anything she does, but we
found actually this AI you knowkind of in the in the platform.
(15:08):
It's like she's like oh, thisplatform looks like works a lot
like my brain does, because youcan kind of put like a bunch of
different you know like chatsaround.
You're like, oh, I'm just gonnaexperiment with these, like
different prompts and things,and it's all there, exposed
right in front of you.
And then it's like, okay, thenI'll create this automation over
here.
But she was creating, you knowshe was doing some amazing
things like reverse engineeringa brand profile based on
(15:30):
packaging of competitors, right,so she would load up a
competitor's package and reverseengineer who their brand
persona and profile was and thenfigure out where their brand
persona and profile should fitinside, like next to that one.
And it's like, whoa, like Idon't think you would even do
that, like without AI technologylike it's, it's actually
(15:51):
enabling us to do things that weshould be doing, that we know
we should be doing, that we'renot because we don't have time
to do it Right, and and thoseare like really exciting things.
And but it she just kept goinglike into like writing scripts
that you know, consulting firmswere trying to do but didn't
quite get the.
They wouldn't quite nail theirhead on the scripts for these
commercials.
And then she, we draftedseveral.
(16:13):
She drafted several scriptsinside of the system and like
nailed it.
It was like 90% there when theother firm was, you know, doing
just 50% of where they wanted toget to.
And then she took that and gaveit to them and they were able
to, you know, adjust it andimprove their work.
But it's like the swath ofthings that you can do we found
(16:35):
that there's about 30% of theusage in our platform is not
even automation.
It's just like ideation andcreating these one-off, you know
, like tests and trials and justplaying with all the AI systems
in one place.
Like we didn't even say thatthat our platform is integrated
150 different AI platforms inone place and systems, so it's
easy just to try them alltogether.
It's like, well, I want tochain Anthropic with OpenAI,
(16:55):
with Gemini and then do imagegenerator between all these
different systems and audio andeverything.
So it's kind of a nice as aplayground to start playing
around and actually ideate.
And then we found then there'sanother 70% of the users or the
usage.
That's like building an agentthat does automations and
systems for you that just makeyour whole life easier, because
(17:17):
you know that's the automationyou want and that's the agent
you need.
And so you build it out and itjust repeats it either every day
or waits for your input, orwhatever it might be.
Speaker 2 (17:27):
Yeah, okay, so I realized we should probably
backtrack, because a lot of ouraudience knows what AI is.
They probably don't know.
They think about AI in terms ofgenerative AI, right?
Not in terms of the big AI andML and DL and all that stuff.
So AI agents might not be aconcept people are familiar with
listening because I know a lotof the students in the program.
Some have been told they can'tuse AI in their businesses or in
(17:52):
their undergrad courses and I'mgoing no, you have to use it in
everything you do, almost right.
So can you talk about agents?
Because this is something I'vebeen really excited about,
because I do want to be able toleave a memo to my personal
agent and say I need a book.
I keep forgetting to book dentalappointments for me and my
(18:13):
daughter.
Can you do that?
Can you schedule the dog's nextvisit to the vet, right?
Can you remember to order this?
Can you order this medicine forme?
Go into my profile.
And because I keep forgettingthat one of my dogs needs more
gallop, rant or whatever.
You know those things that takeup time and that if we're so
busy, I didn't iterating andworking and then also trying to
(18:34):
spend time with our families.
Sometimes those things fall bythe wayside.
Speaker 3 (18:38):
Yeah, yeah, they definitely do.
Yeah, I think to answer kind ofthe base, foundational question
and this is this is such afascinating time.
I've seen it probably threetimes in my career now.
Every time a new kind of bigwave happens, all these terms
start getting thrown out thereand they slowly coalesce into
like what is the actual term.
So the term agent has meantdifferent things to different
(19:02):
people all along, but I thinkwe're actually getting closer to
a coalescing of what an AIagent is, and I think it's my
opinion.
I've seen it many other peopleand again, it's still coalescing
.
But is any basic, like the mostbasic agent is any system that
is including an LLM, some kindof large language model, in its
call, in its process.
(19:23):
That's like the most basic formof an agent in my mind.
So it could be.
You know, I have a tech systemthat has to hit an LLM call to
do something particular and thensends back some information to
me in a different structure,like in.
Fundamentally, that could becalled an agent.
Now, some people would be likethat's not smart enough to be an
agent.
I get that argument too, but tome it's like you don't
(19:47):
necessarily know exactly whatthey're calling.
Anyway, that's my initialimpression for the low bar.
Speaker 2 (19:52):
Yeah, like you go onto a website and you have a
customer service bot, right?
Yeah, is that an agent?
Speaker 3 (19:58):
Yeah, it probably is, because it's using an LLM to
reference its memory and itssystem.
And so I think the next levelup there's probably degrees
right, like a really base agent.
And then the more advancedagents, like, well, an agent
that can actually call and useAI and LLM to call multiple
different systems to likeaccomplish an objective and a
goal, like that's probablyanother like the mid tier, like
(20:21):
level of defining an agent and Ithink.
And then the top tier is likean orchestration, where you have
like one agent calling multipleagents and each one of them
have their own kind of tasks,and that's like a larger kind of
agents system that I would say.
So those are probably the three.
I would three ways I define it.
Speaker 2 (20:38):
Yeah, I was listening to another podcast that was
talking about that.
In this form concept is likeyou may have an agent that acts
as a CMO, but it's probably andgoes into a meeting for you, but
that agent probably isn't justone agent, it's a whole series
of agents.
Each one knows there's like anexpert in this one thing.
Yeah Right, so instead ofhaving to have a whole bunch of
(20:59):
human experts, you can have that.
So where does you know what'sthe play between humans and AI?
Because that's a questionpeople get really scared about.
I always try to emphasize no,no, no.
It's not something to be scaredof.
We need to be able to utilizethis technology, but we still
need our brains, because AI justknows what everybody's input
(21:21):
into the internet, and that canbe good, bad, average
everybody's input into theinternet and that can be good,
bad, average, right, yeah, andso I think we're in a.
Speaker 3 (21:30):
Really my opinion we're in a fascinating place.
Maybe I'll go a little too deepon this one, but and you can,
you can pull me back out.
I think we're in a fascinatingplace because a lot of people
expect like, hey, can ai, canyou go do all of these various
things?
For me now, a generative ai islike it's coming.
All it is really doing is it'strained on trillions of patterns
and it's finding a completionof those patterns in text, in
(21:51):
particular, or images, and sowhat it's doing is trying to
complete a pattern based on yourinput.
And so, like I said, it doesn'tknow anything, but it knows.
George Washington was presidentbecause it was told that 10,000
times through the patterns,that's the pattern it knows.
To finish, it actually doesn'tknow it.
So when you think about agents,we start getting into this
(22:14):
multi-orchestration.
Where does the agent know thepattern of how to get this your
dog food Like?
It hasn't.
There's not a million.
It wasn't trained a milliontimes on getting dog food right.
It's because these are pathsthat we're creating, sorry if
I'm getting a little deep here.
It's like when you think aboutan agent and this is actually
why I think it breaks down,still like it's not great yet is
because it hasn't been trainedon all these paths to like, get
(22:36):
dog food.
So you actually have to be veryspecific about how you train
that agent to do exactly whatyou want it to do, because if
you don't, it will break.
And so it because it doesn'tknow, it doesn't have enough
data, it hasn't been trainedenough on how to do it, and so
there's a lot of this kind ofwork on like hey, can this agent
do everything for me?
And it just doesn't.
It's not trained, there's notenough patterns for it to know
(22:57):
how to do these things.
Now they're working on thingslike MCP is a potential you know
solution for that.
That's more of just likeallowing them to talk better
together.
But I think like what we producejust as an end result, like we
produce like 99 or 100% accuratedocumentation through all
through AI systems.
So people put in like fivedifferent documents about a
(23:18):
medical patient.
It reads it all, it breaks itall down and spits out a perfect
report that it looks the exactsame, formatting everything
every time with the new patientinformation.
Now, like no LM can do that.
The way we do that is westructure exactly where the you
have to structure where the datagoes and how the LM is going to
digest the data and spits itout.
(23:39):
So, like you have to createthese, I guess, guardrails to
kind of make it look exactly theway it should look, and I see a
lot of these larger agents andorchestrations of agents.
The same way is that you haveto create these guardrails of
like what it does.
So in that CMO in the meeting,yes, you theoretically and you
probably will have all theselike micro CMO agents that are
doing various things, butthey've been told exactly what
(23:59):
they should be doing.
And like that's my job, that'smy only job, I'm just going to
go do this, always look for the,you know the opportunity in the
room and then find you knowsomething to do with that
opportunity.
Like that's that whole agent'sjob.
So like though that's how youwould break these down.
And that's why they talk aboutswarms of agents, because AI
can't necessarily create allthese and do it very accurately.
But if you're doing theseindividually and you're making
(24:21):
them, you know it can be 99 or100% accurate with the kind of
way that you're building theseagents and then you can use that
forever, right?
That agent is just really goodat what it does.
And then you have another agentthat can call it when it needs
to, and it kind of knows its wayof calling it.
Speaker 2 (24:36):
Now, that was like sorry that was probably way too
broad of a term, Because I thinkpeople need to understand this
because, even if they're notusing it now, these are concepts
that are being talked about andeven if they're not mainstream,
they will be.
Yeah, so you know, we wanteverybody to have as much data
as possible and have theunderstanding of what all these
(24:58):
concepts mean.
Another concern that peoplehave is privacy and security.
That was actually the classthat we had on Wednesday of this
week was about GDPR and CPRAand different like those kind of
consumer privacy restrictionsand laws, but then I had all the
students also add in the AIcomponent which states and
countries have AI regulations,which ones don't?
(25:21):
What do those regulations mean?
Right?
And so I'd love to hear alittle bit, because that's a big
question people have, andthat's why I use certain
products more than others,Because you know, and so if
somebody is building stuffwithin Flux Prompt, you know
it's all secure.
Speaker 3 (25:38):
Yeah, yeah, yeah, it is secure.
Well, actually, we leave itopen to the user.
I mean, it's secure inside theplatform but the API the
platform but the API.
So, as of now, the recording ofthis the most secure legal
documentation is in bothAnthropic and OpenAI, from what
we've reviewed in terms of, like, what they do with your data,
and that's particularly now.
Don't get that confused.
(25:58):
If you put stuff in ChatGPT,it's like everywhere, so don't
put stuff, anything in ChatGPT.
Well, there are some like tiersand then there's some loose
legal documentation around likethe team tiers Enterprise.
It actually gets to much betterlegal documentation.
But if you're using API callsinto these systems, then there's
actually like really strictsecurity measures around, like
(26:21):
what you're sending in andthere's really clear guidelines
of like what is shared andwhat's trained on and what's not
trained on.
And, as of this, you know, asof last time we looked at, which
was, I think, earlier this week, those are the two most secure
systems and all of our calls.
If you're calling in athird-party LLM, like we said,
(26:42):
we have about 150 in our system.
It's all through APIconnections to these other
systems and so we're looking atall the license agreements
between those API between thosecalls and how secure that data
is, and we've been trying toexpose that to people like, hey,
these ones are, you know, a lotmore secure than the other ones
, and so if you want to besecure, that's one way.
(27:03):
There's more ways of going aboutit.
One of the reasons Facebook'smodels have been big is the Lama
models and then DeepSeek isbecause they open source their
models, which means I can takeit and put it on my own server,
which means it will never gettrained on anything because it's
on my server.
I control what's gettingtrained, and so in those
scenarios that's 100% secure ifyou can put one of those models
(27:25):
in your old system, and there'smany companies who've started
building stuff on top of thosemodels.
That's the reason open sourceis good, because someone can
take it and make it securearound their own systems.
So that's like the extra secureway of doing the LLM calling
Calling into a third partysystem like OpenAI and Anthropic
.
Those two are the most securefrom a legal perspective.
(27:45):
Now, yeah, anyway, that's.
Speaker 2 (27:48):
No, but that's, I think, important things for
people to think about, right?
So for students who arelistening, where should they
start?
Maybe they've tried.
They've used Grammarly or Chator Cloud or name your favorite L
, that you know the big ones.
Maybe they've done that.
Maybe they've tried out acouple of other tools.
But what else do they need toknow right now and how do they
(28:11):
move forward in this AI firstworld?
Speaker 3 (28:14):
I would say, first of all, it's amazing that they're
students.
That's really good, because AIwhat it is at its core.
It doesn't know anything, butit's a pattern machine.
It loves, loves, loves patternsthing but it's a pattern
machine.
It loves, loves, loves patterns.
So if you know patterns andframeworks and theories, which
is taught in schools, thenyou're going to have a leg up,
(28:34):
particularly if you know how tocreate patterns and theories.
And now the kind of next levelis using AI to create patterns
and theories, which is actuallyreally fun, to be honest.
That's how part of the thing Ido with my board game that I
created is creating an ecosystemthat works with certain
foundational principles.
But the idea is that, if youknow, really lean heavier into
(28:57):
the idea of creating a framework.
Or when I say framework, I meanlike hey, this marketing letter
should have this kind of atitle and should be this long,
and then this kind of an openingparagraph should be this long.
This, then this kind of anopening paragraph, should be
this long, this mid body, andthen this, this conclusion,
right, like general marketing islike hey, call out the pain and
solution, right, so like that'sa framework pain, solution,
framework, right, or it'susually part of a little broader
(29:18):
one, but knowing thoseframeworks, and then you can
tell AI hey, I want you tocreate this based on this,
follow this framework, and thenyou give it your framework and
you can make it your own.
You can customize someoneelse's a little bit to be, or a
well-known one a little bit tobe your flavor, and then it's
actually going to be able tocreate really good content that
you, that comes almost from youand your guidelines and your
(29:40):
frameworks.
And that's kind of the worldwe're moving into.
Is this world of everyone'skind of the conductor, the
orchestrator?
You're giving the AI the tasksof what to do and if you give it
the right tasks of what to do,it'll do a good job.
If you give it the wrong tasks,you'll still do a poor job.
Right, you're not even givingit the right tasks to do so.
I would actually encouragestudents to think a little bit
more meta not Facebook, butthink a little more meta in the
(30:03):
concept of dealing with AI andhow to create frameworks and
structures in your prompts andhow you're prompting it and then
even using it to help youcreate those Miveta prompts and
those higher level prompts.
But to the actual toolsthemselves.
I mean, I would say, definitelytry Flux Prompt.
Come and try it, hit us up atany point, we can help you.
This is designed for thebusiness person we is.
(30:25):
We almost picture this likeExcel.
I don't know if you rememberthe first time you opened Excel,
but it's got formulas, it's gotlike references, all sorts of
things, but then once you learnit, you're like, oh, this is
actually pretty easy and you getto use it.
It's very similar.
We found very similarexperiences.
Folks come in, don't quiteunderstand it at first and they
start learning like, oh wait,this is super easy.
(30:47):
And they start creating theseincredibly powerful AI agents
and systems, and so I would say,definitely try our system out.
There's several other systemsthat you can try in terms of
agent building, orchestration,but ours is our favorite because
it's easiest to use in our mind.
But it's OK, we're biased.
Speaker 2 (31:04):
I have two questions for you.
One is when you're on a run andyou're speaking to AI, what
folder are you speaking to?
Speaker 3 (31:14):
That's a really good question.
I've used Perplexity, and thereason is I'll give you my two
or three reasons.
Perplexity is internet, first,and the other ones have been
slow to catch up, so it can usethe references of the internet
when I need it to.
Number two is I could switchthe model I use on it.
I can use Claude, I can useOpenAI, I can use Perplexity's
(31:35):
own model, and I still loveClaude the best, and so I use
Perplexity with Claude as themodel that responds to me.
And then, third, is that it hadthe easiest interface for me to
talk and run at the same time.
Okay, but, and I'll tell youthis, it's a weird trick is that
I don't ever press like recordmy audio on the actual app,
(31:56):
because on any of them, becauseit like tries to automatically
stop at the wrong time and likeit waits till I pause and I'm
like no, I'm still running, I'mjust breathing heavily, like I
want you to keep recording me.
And so I actually use the textto speech on my phone, natively
on my phone, and I use that torecord my audio and I may have
like five minutes of talking orwhatever.
(32:16):
And then when I'm finally donewith my concepts.
I press stop and then itconverts it all into text, and
then I say go and then itcreates the whole thing.
And it creates the whole thingand then press the play audio In
perplexity it's been theeasiest for me to use and it
plays the audio back.
So then I listen to theresponse for the next five
minutes.
So within, like you know,within one mile, I'm listening
to, like you know, maybe acouple of responses back and
(32:38):
forth.
Speaker 2 (32:39):
Yeah, between me and Bob, into it Whenever I'm
walking my dogs is when Iremember something that I need
to do, and then, yeah, I caneither text myself or voice memo
, but I'd rather, if it'ssomething more complex,
obviously I'd rather just talkto my AI.
Speaker 3 (32:55):
Yeah, yeah, yeah.
Speaker 2 (32:56):
Fantastic and I understand we also are going to
have a special link discount forlisteners if they want to sign
up for the class that I'm inright now I mean, obviously it
would be different cohort andlearn how to build their own
agents.
Speaker 3 (33:09):
Yes, yeah, we're providing a special discount
just for listeners here.
We love what Anika does and thepodcast, so and again, the
class is designed to be a cohort, so you actually work with
other professionals around youand you get to know them.
There's actually a lot of greatrelationships that people build
, but it's everyone's learninghow to build agents in their own
(33:32):
AI agents and, believe me,we're coming to a world where
there's going to be 1000s ofagents, and that's okay, because
half of them, you know, you'llhave your own several dozen
agents or a couple 100 agents atsome point, and you want to be
able to configure them on yourown, in our view, and be able to
do exactly what you want.
And this is kind of the verybeginning of getting your feet
wet.
We've had people build amazingthings just coming out of the
cohort.
They've created businesses withjust what they've built from
(33:55):
the class of.
They've learned how to buildand we kind of run through
everything.
So we would love yeah, lovepeople to come and check it out.
Speaker 2 (34:02):
Fantastic, brad.
Thank you so much.
It's always fun to speak withyou and we can talk for so long
about all the different ideas wehave and how to implement
different technologies andstrategies.
So I look forward to continuingthe conversation.
And, yeah, thank you forjoining us on Mediascape.
Speaker 3 (34:19):
Yeah, thank you for having me.
It's a pleasure being here,absolutely.
Speaker 1 (34:22):
To learn more about the Master of Science in Digital
Media Management program, visitus on the web at dmmuscedu.
Popular Podcasts
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.
My Favorite Murder with Karen Kilgariff and Georgia Hardstark
My Favorite Murder is a true crime comedy podcast hosted by Karen Kilgariff and Georgia Hardstark. Each week, Karen and Georgia share compelling true crimes and hometown stories from friends and listeners. Since MFM launched in January of 2016, Karen and Georgia have shared their lifelong interest in true crime and have covered stories of infamous serial killers like the Night Stalker, mysterious cold cases, captivating cults, incredible survivor stories and important events from history like the Tulsa race massacre of 1921. My Favorite Murder is part of the Exactly Right podcast network that provides a platform for bold, creative voices to bring to life provocative, entertaining and relatable stories for audiences everywhere. The Exactly Right roster of podcasts covers a variety of topics including historic true crime, comedic interviews and news, science, pop culture and more. Podcasts on the network include Buried Bones with Kate Winkler Dawson and Paul Holes, That's Messed Up: An SVU Podcast, This Podcast Will Kill You, Bananas and more.