Subscribe
Share
Share
Embed
Around the Desk: This is where we at Avory think out loud, challenge narratives, and look for signal through the noise. Each episode, the Avory & Co. team dives into what’s moving markets, how companies are performing, and where opportunities may be forming. We break down earnings, macro trends, and investor sentiment — all from the lens of a concentrated, high-conviction portfolio.
*** The views expressed on Avory Podcast: Around the Desk are those of the hosts and guests and do not constitute investment advice. This podcast is for informational purposes only and should not be relied upon to make investment decisions. All investments involve risk, including potential loss of capital. Avory & Co. may hold positions in the companies discussed.
Speaker: All right, we just
finished up a conversation.
We're all playing with Meta Muse,
trying to figure out what this
product is, how it's working.
I'm using it a lot.
I'm using it a lot for
many different things.
I'll talk about that in a
second of how I'm using it.
Also share with you some of the use
cases I'm seeing on the internet.
A lot of it is saving people
money, saving people time.
It's number one in the App
Store for a reason as of today.
It's been a week now, so being at
the top of the App Store means it's
growing in virality, not contracting.
A lot going on there.
Zuckerberg has been calling it
personal super intelligence.
He's been echoing and, and s- stating
that type of vision for quite some time.
I think that vision is here.
Alex Wang is on X smashing all the
different use cases and doing a
good job and making sure that Meta
Muse is viral and the adoption
is centered around real value.
With that, let's get started today.
I am Sean Emery, founder and chief
investment officer at Avery and Co.
Today, we're pulling on the thread
of Meta Muse, the why, the who, the
when, and what's happening there.
We'll be quick, but before we
begin, this discussion is for
educational purposes only.
It should not be considered
investment advice.
We are investors in Meta,
so take that for what it is.
Opinions reflect our views as the time of
, this recording.
Those may change.
Do your own research.
So let's kick it off.
What Muse actually is.
Let's step back for a second here.
Uh, what's it good for?
Uh, it's powerful.
You know, I've downloaded it again this
week, and it-- again, it's sitting at
the number one app in the App Store for
a reason, and usually, when you see these
viral moments, they dissipate over time.
If anything, uh, some of the
virality again has been building.
We've gone from, , top ten to top
three to top one, surpassing ChatGPT.
What we wanna do is obviously find
and, uh, track retention over the
next thirty, sixty, ninety days.
If you follow our investingwithdata.com,
our Investing With Data newsletter You'll
be able to follow along with us over
the, the course probably over the next
month or two as we, uh, share some stats
each week on retention levels there.
But, you know, I think the Muse app
has done a great job at building in
retention mechanisms inside the app.
Uh, starting off with kind of what it is.
It's a personal agentic app.
Right when you open it, it already knows
a little bit about you, given that it's
connected to things like Instagram and
Facebook and WhatsApp, mostly Instagram,
I believe, and they're highlighting some
of those use cases that are, are popular.
So right when you open it, you see
those use cases, and it's already
prompting you, "Hey, do this.
Hey, do that."
I think some of the lowest-hanging
fruit is where people are
seeing real value immediately.
And when I say value, again, I don't
mean Instagram value, where you
go there and you're entertained.
Uh, it's, it's far from that.
Some of the use cases going viral are
people lowering their auto insurance,
so people finding, uh, what their
current, uh, auto insurance is and
having their muse, their agent, go out
and, uh, get quotes and garner quotes.
And I've seen screenshots,
uh, of hundreds of dollars to
thousands of dollars of savings.
I did it myself And, you know, I'm
happy to say that I have the lowest,
uh, quote, uh, i- you know, relative
to the quotes that I receive.
But my agent went out, my muse went
out, uh, and sourced those quotes
for me without me lifting a finger,
which is pretty astonishing where
we are, uh, in the personal super
intelligence, uh, agentic, uh, ecosystem.
Another one is subscriptions.
You know, they didn't know they
were running in the background,
and so there's some apps that
have historically done that.
"Hey, you have these, uh,
reoccurring payments, you know,
do you wanna cancel them?"
Now, canceling them is the hard part.
The difference here is that your
muse, your agent, will go out and
do that for you on your behalf.
So people are saving time, people
are saving money, and setting up
some of these scheduled tasks to do
reoccurring things, which I think
again, will keep retention elevated.
Now, why the product actually works,
really the next segment I wanted
to frame here, like, why is this so
different from anything else we've
seen if you haven't played with it?
I think the most important
thing here is why it's so good.
And look, in the early days of AI, we
did see some browser use, computer use.
Um, and there has been really
two lanes to try to tackle an
agent doing things for you.
One was through the connector lane,
so connecting your agent tool to
MCPs, which is Model Context Protocol,
which is really a newer way in the
last couple years for agents to
talk to agents to get things done.
Communicates in a certain way
that is, uh, clear to the agents.
Then there's APIs as well, which is the
traditional sense of, you know, connecting
two databases or two, um, ecosystems.
So we've seen those, and those work
pretty well The next stage here, and
the one that a lot of companies have
been trying to develop, is really
that computer use of which your agent
will spin up a virtual browser, a
browser like Chrome or something, and
actually go out and go on that webpage
and try to figure out the login and
get to the right spots on that page,
depending on what you're trying to do.
Now, that was hard, and
it was also, um, costly.
Hard being, you know, it would take a
long time to figure it out, and that has
a feedback loop of cost because it would
take quite a bit of tokens for a agent
to go on a browser, iterate, uh, and try
to figure out all the different pixels
on that browser and find the login page.
Then eventually you would have the
login page change, and at that point
in time, your connection would break,
and things were just messy there.
So retention on that type of product
would essentially break because
you're either paying too much money
and not getting enough value, or
secondarily, it would just break too
much and, you know, consumers don't
have, um, the ability to, you know,
try to figure it out on their own.
And so again, we went from this MCP
API world, which is -- does a good
job if there's connectors there.
But if you think of something like auto
insurance, the use cases I gave earlier,
auto insurance companies, like they
have no interest of giving connectors
to Claude Code or, you know, um, Claude
Coworker or a ChatGPT or Perplexity
Computer or some of these agent harnesses.
And so what computer use does is spins
up a personal, uh, secure browser
and goes out and does it for you.
So now those, those platforms,
those ecosystems, those interfaces
that were once, uh, prevented
from, you know, being able to do
anything with it are now available.
And so those auto insurance examples
are, you know, some of the lowest
hanging fruit that I'm talking about.
And so Meta figured this out.
They've done a really good job at this.
Uh, I'm actually pretty shocked that
they were the first ones to figure
it out, uh, in a way that is snappy,
fun, uh, playful, and reliable.
And it's just a fast, really
well put together product.
So we've been hearing from
Zuckerberg, he's focused on super
personal, super intelligence, and
I think we're seeing that now.
Now why Meta?
Like, why do we think potentially they
have been able to get to this place?
Let's start with first principles.
You know, an agent needs, I think, three
things to be, you know, really useful.
It needs to know you Uh, it needs a way
to reach you and communicate with you,
and it needs to act on your behalf.
If you have those three things,
you have a really good setup.
And that first one is
context, knowing you.
You know, in work, we all have context
where it's files and folders and email.
It kinda knows you.
The output of these agents are pretty
powerful the more context you give it.
In the consumer world, i.e.
what Meta's developing with
Muse, is the context is sparse.
It's not in your friend conversations.
You don't really email your friends.
How you communicate w-with your friends
is through text, WhatsApp, uh, group
chats, Facebook, Instagram, and some
of the other platforms out there.
So clearly, Meta stands to benefit as they
have more context than everybody else.
They have all the personalized
information that you need, all the
Instagram channels that you like, all the
brands that you follow, all the people
that you interact with, and so that
social friend graph is very powerful.
On reaching you and communicating
with you, if you want your agent
to talk to you, or you wanna talk
to it, historically it's text, a
little bit of audio, depending onâ¦
That's more in the enterprise
space, but mostly text.
And so the two text interfaces,
or the interfaces that most people
interact with are, uh, WhatsApp,
your iMessage, depending on, you
know, if you're on Samsung or Android
or iOS, and then, you know, some
of the social platforms out there.
So Muse is its own app.
You can communicate
directly there through text.
Uh, you can connect your Muse inside
of your WhatsApp, so then it's, uh,
you can communicate through it there.
So two channels on your device.
Um, I imagine that this will
come to Instagram and some of
the other platforms where you can
communicate with your Muse as well.
Uh, the fun thing is you can name
your Muse, you can change the avatar
of your Muse, has a little bit of a,
um, motion graphics as it's working,
and it just makes it very playful.
So they did a really good job with the
interaction experience of the device.
But eventually we're gonna get to a
place where you're communicating with
it, uh, through your voice, and I think
obviously Apple with their AirPods
is well set and positioned for that.
But Meta has their Ray-Bans and that whole
line with Oakley as well that they've
done a really good job at creating nice
products with Ray-Ban that people like.
Now, they're taking off the camera,
which is interesting here, uh, in
one of their models, and I think that
is a play on having glasses that are
these Meta, uh, glasses, but you're
gonna communicate it with it not to
film the world Which I think, uh, some
people have viewed as, uh, s- you know,
intruding on people's privacy, which I,
you know, somewhat agree with for sure.
Uh, but if you, if you have these glasses,
but you can, you can talk to them and
have really good audio and sound, so if
you're listening to music, I think it
positions Meta really well to communicate
with your Muse and your, your AI agents.
And then oc- a-acting on your behalf,
the third thing, which has really
been the missing layer, is, you know,
Muse has these secure VMs, virtual
machines, uh, where, you know, they
have a dedicated cloud computer,
uh, per user on its own browser.
Plus then you have payments
through something like Link,
which is owned by Stripe.
Uh, we know Stripe just acquired, uh, Open
Router, which is a model, uh, platform.
And so there's a clear connection between
models and payments and, you know,
uh, Meta's ability to have computer
use in a reliable, seamless way with
connectors for payments and agentic
use, I think is so, so powerful, and
it really starts to blow out the water
when you think of what ChatGPT has.
ChatGPT today has a lot of information
about us, but not as much on a consistent
basis if you're not using it every day
as, you know, someone like Meta has.
You have Anthropic, which isn't really
going after the consumer, so they have
less information about your personal life.
You have Google has some of your search
history, but again, if that's decaying
and people are moving to ChatGPT for
some of their search queries and all
of these other platforms for search
queries, they start to probably
lose a little bit of signal there.
And so, look, there's gonna be the
dominant players that we all think of
in this space, but ultimately, I think
it comes down to those three things.
Do they know you?
Can you reach it and reach you?
And, you know, is there an action
layer that can, can help assist you?
Give you a couple of use cases that I use.
I, uh, um, I go out and buy seltzer
water, uh, every month in bulk,
and I try to, um, you know, search
for the lowest cost per ounce.
Uh, that's just the way I'm built.
So, you know, historically,
it would take me, you know, a
little bit of time, not too much.
You know, you'd go on Instacart, you'd
go on, you know, Costco directly, um,
you'd go on Amazon, Walmart, and you
would just, you know, quickly try to see
what's the, the lowest cost per ounce.
Now, I have an agent doing it.
So I already have that scheduled,
and every month it's doing it.
Uh, so we'll see as time
goes on how reliable that is.
I'll, I'll pressure test it,
and I'll come back to you guys.
Another one which is pretty
interesting is theâ¦
You know, my, my, uh, little boy
has a baseball team, and they don't
provide the schedule very frequently.
It's kind of like random and sudden,
and checking every day isn't very, uh,
efficient, and the idea-- They don't have
an iCalendar to connect to your calendar,
uh, to, to notify you of anything.
I've tried to create computer use
for this, and it breaks every now
and then, and so far, so good.
It's going on there,
checking it every single day.
Don't tell me if anything's changed,
but tell me if something's changed,
add it to my calendar, create me
a message so that I can forward to
the, the people that need to know.
And so, one, I'm getting money and
value, like value through money and
time, meaning the seltzer water cost per.
And then one, I'm just getting time
back, and as I continue to stack
some of these things, clearly there's
value being extracted by this Now,
the next thing I wanted to mention
was the brand inflection here.
I think it's super important.
You know, Meta has had a bad rap.
Really, Facebook has had a bad rap.
There's real brand damage
at the Facebook layer.
I think over the last several years,
when they went from Facebook to Meta as
a corporate company, I think I posted
this on the internet, I said, "Outside of
just the metaverse, I think the biggest
thing here is that, uh, new products will
be developed by Meta, not by Facebook."
And I bring that up because they had
the Facebook Portal, and the Portal
was a really good in-home camera.
You could communicate with friends,
uh, very interesting product,
well-designed, well-crafted.
The biggest issue there was not the
product itself, it was the brand.
And I don't think anyone felt
comfortable, including myself,
putting a Facebook camera in my home.
And I think there's a couple
companies that have earned that
right and trust, but very few.
And so I think brand matters as we get
more and more personalized, and having
Meta as kind of that parent, uh, as
opposed to Facebook is so important.
So the timing of that at the, at,
you know, uh, several years ago was
perfect for this bigger story here.
And, you know, ultimately,
I think that's paying off.
Um, people are, you know, hooking up
their bank accounts to Meta Muse, so
there's a lot of, uh, um, you know, a
part of that thesis of the brand and
people trusting it and people oversh--
you know, overlooking the idea that
this i- is owned by Mark Zuckerberg,
uh, is very, very, uh, powerful here.
But ultimately, it's because the
p- the, the product itself is
well-built, the interface is playful
and fun, it's snappy, uh, powerful,
well-designed, and very little friction
as it relates to getting in there.
Now, what does this mean
for ChatGPT and Anthropic?
I think this should put a lot of fear in
ChatGPT and OpenAI, obviously the same
com-- uh, you know, company and product.
But I believe less so for Anthropic.
But if we think of ChatGPT, they've
been trying to build Many of these type
of tools over the last several years,
they came out, they had computer use
for a little bit, wasn't very good.
Uh, I'm sure it'll get better over time.
And they have had the
ability to reimagine search.
So they do have a lot of personal, uh,
hist-- or information about you in there.
If you ask it to tell you about you,
it's probably gonna do a pretty good job.
But it doesn't have the Instagram, it
doesn't have those private conversations
that you have, uh, through WhatsApp
that you can open it up to, uh, to,
um, you know, learn more about you.
So I think OpenAI is trying to get
into that agentic, personalized
kind of super intelligence layer,
uh, but they've failed to do so.
I think part of it is user interface,
part of it is everyone started
there as almost a search browser.
Yes, they're doing really good with Codex
on the enterprise side and model layer.
And I ultimately think Meta has
done a good job at skipping in
front of OpenAI and ChatGPT as an
agentic tool for personalized use.
We'll see what happens again over the
next month, two months, three months.
I fully expect, uh, new experiences
inside of ChatGPT to try to mimic
the success that Muse is having.
Anthropic's a little different.
You know, I think ultimately
they're focused on the
developer, on the enterprise.
Great.
Uh, the consumer is really not the
fight they're trying to get into.
I think that's fine on paper, but I do
think ultimately over time, as people
trust more Muse the app and then Muse
the models, it does open up the door,
uh, for enterprises at least to pipe
into the APIs and, uh, AI infrastructure
that Meta has developed and built out.
And they're trying to be the
low-cost provider there on the
API cost level for model delivery.
And so that does open up a little
bit of that because they're
clearly seeing success for, uh,
computer use and other forms of AI.
And so I would be still concerned, uh, if
I was, uh, on the Anthropic side as well.
Now, how do they plan to make money here?
I think this is, uh, important You
know, at this point, if you use
Muse enough, it costs something, but
right now it, it's pretty much free.
Has a really high ceiling.
I've used it a lot.
I haven't hit any limits.
Eventually you hit some level, and
it's around twenty bucks a month,
so that's one m- economic model
I think they're gonna work with.
Uh, right now I think it's five hundred
million tokens that you get up to.
I'm not sure exactly when that resets.
I think it's either weekly or monthly.
But either way, that's a high ceiling.
I use it a lot.
It's doing a lot of browser use,
and I'm not hitting any ceilings.
So, so far so good.
I think the second monetization
lane is the transactional element.
You know, historically, you know,
we viewed, uh, Meta, Instagram,
Facebook as the new form of the
mall, where you go there to entertain
yourself, you go there to window shop.
The difference is conversion.
How do you convert people from
window shopping to purchasing?
And then how do you, uh, h-
how can you be part of that?
And so that is, I think, what's
happening here, uh, a little bit.
Historically, they had the ability to
pay through Instagram, but it wasâ¦
You know, if you've used it,
it's somewhat friction-filled.
Uh, there's a little bit less trust
inherent in kind of m- the, the, the flow
of that product and going from clicking
a product Shop Now to purchasing it,
putting in your cart, and, and so on.
And, and reality is with agentic
use and computer use and an agent
and a payment link, like link from
Stripe And I'm sure there'll be more.
I'm pretty confident that that friction
is going to drop significantly, you
know, by fifty to a hundred percent,
and everything will be consumable.
And therefore, Meta will be part
of that transactional layer of
which they can experience quite a
bit of transaction volume on their
platform and monetize it that way.
And then lastly is advertising.
Clearly, you know, uh, either
sponsored, you know, things on the
Muse platform, new offerings on the
platform that are, you know, upcharges,
um, and other ways that, you know,
Meta could, could, uh, monetize this.
Now, where could we be wrong
and kinda pressure test this?
Obviously, Muse could stall.
They could have a security event.
All these things could create some sort
of, uh, headwind for additional usage
if the computer use side of the business
is ultimately not, uh, as reliable as
it feels like in the first week, uh,
and eventually it starts breaking.
Or the ecosystem that these
agents are going on, i.e.,
going on platforms, and they're
getting blocked, um, that could
slow things down and make things
a little bit more difficult.
Uh, but ultimately, I think it's
Meta's job to incentivize external
platforms to embrace agentic commerce
as opposed to, uh, push back against it.
Ultimately, I think the winners that
embrace it will win, and those that push
back against it will lose long term.
But in the short term, you could
go-- you could get some pushback,
and I think that's, uh, important.
But we'll be watching that again,
uh, on our investingwithdata.com
newsletter, where we'll be posting again,
uh, the The data on retention and use
case and engagement on these platforms.
So again, go on investingwithdata.com,
subscribe to that and
you'll see it every Friday.
But what we're watching going forward is,
you know, how Muse adoption ramps here.
Again, ninety days, sixty days,
thirty days, those are pretty good
retention curves that you wanna track.
Uh, it's, it has a, had a great
launch, number one in the App Store.
If it starts to fade and engagement
starts to drop, then we'd be
curious just to understand why.
Again, I'm a user, so I'll, I'll have
a, a pretty good, um, insight into
what's going on, uh, there as well.
So with that, you know, stay tuned.
But, you know, just to close out, I
think the question we started with
was, what does this all mean for Meta?
I think the answer, at least for one
week, is that Muse and Meta are really
joining, leading the agentic race.
It's a big deal.
Uh, I think they're showing that
all of this infrastructure that
they've built, all the talent that
they've brought in, if anything,
this will further supercharge that.
People wanna work where the action's
happening, and clearly Muse and
Meta have just taken a step forward,
really over the last month with the
models that they launched, plus these
agentic tools that they've launched.
So all of that is incredible.
I think, uh, we also wanna see how
the competitive response is from
OpenAI, Anthropic, Google, and some
of the other platforms out there.
I'm sure we will see them here shortly.
But with that, we'll end there.
Make sure you subscribe to the
podcast, Avery Around the Desk.
Make sure you subscribe
to investingwithdata.com.
Go check us out, avery.xyz.
And with that, we'll see you next week.
Thanks for joining us Around the Desk.