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.
I want to walk through something we
have been kicking around at the desk.
You know, it has to do with
software, as you probably expect.
It also has to do with AI as you a-- you
know, probably expect and really how AI
might be changing what actually wins.
You know, here's the, the, the
setup, at least the framework.
You know, really over the last twenty
years, software companies have been
trying to do one main thing, and
it's really a combination of things.
It's own the workflow, it's own
the user, it's own the data.
It worked.
You
know, you look at Microsoft, you look
at, you know, Adobe and HubSpot and
Salesforce and ServiceNow and Workday, um,
and others, you know, that come to mind.
You know, all of them
had that same playbook.
But I keep coming back to one question
is, you know, if humans are not the
ones picking the software anymore,
only, if they're only not the ones,
then does that game still work?
And because, you know, as time
goes on, AI becomes a buyer.
Not a user only, but a buyer.
And AI does not buy the same way humans do
Welcome back to Avery Around the Desk.
I am Shawn Emery, founder
and chief investment officer
here at Avery and Company.
Each episode, we take one question
and try to think through it
from those first principles.
Today's question is simple: What happens
to software when AI stops buying platforms
and starts choosing capabilities?
Before we go further, though, this
is for educational purposes only.
Nothing here is personalized
investment, legal, or tax advice.
Views are the opinions of myself.
Use your own judgment.
Use your own analysis.
Do your own research.
And with that, let's get into
the episode So let me back up.
You know, software has really
gone through three different eras.
You know, era one I think was,
you know, individual software.
Uh, you know, you buy Word, you
buy Excel, you buy Photoshop.
Everything stood alone.
Uh, you kinda bought a tool.
You used a tool, uh, and, you
know, eventually that tool was,
uh, something that really drove
productivity in your life.
You know, era two is the one
we've been living, really.
It's really around platforms.
And again, you've gone
from tools to platforms.
So think of, again, Microsoft 365, you
know, Salesforce and Adobe Creative Cloud.
You know, clouds, everyone kept talking
about, you know, we are the platform,
we are the cloud, we are, you know, the,
uh, the ecosystem you wanna live in.
And again, I think that's gonna
hold for a lot of, uh, you know,
companies still in how they operate.
But, you know, the pitch there was,
you know, do not buy single point,
you know, tools, point solutions.
You know, buy the workflow.
And it worked because ultimately
those integrations, um, they're hard.
They were hard.
Um, they're still pretty hard,
but they were much harder before.
So, you know, because switching
was expensive and because your data
lived, uh, inside the platform, it was
painful to get out of that platform.
And so suites and platforms,
one, they would bundle price.
There were so many reasons
why you would go that route.
And for a long time, uh, it was
the same thing over and over
again, but something is changing.
It's shifting.
I'm seeing it in some of the
companies we own, and I'm happy about
it, and that is era number three.
And this is kind of our view here.
And again, it's not to say that this
view is widespread, uh, but I think,
you know, I think we're, uh, we're,
we're on the precipice of something
different as it relates to software.
And I'm going with composable software.
That's the terminology, where AI doesn't,
um, or, or, or really the AI does, you
know, the orchestration and the platform
is no longer as much the point, right?
And, and you may have, like, these
suites, these legacy suites that
are, again, the orchestration layer.
So what does that actually mean?
And, like, what actually
changes in that framework?
And this is what I keep coming back to.
You know, historically, humans,
you know, they navigated software.
You logged in, you clicked around, you,
you kind of, like, learned the UI, which
is the uner-inter-- user interface.
You did the work.
You know, the company that made the, the
UI that was easiest to learn You know,
really won that game over time, and
then they would build on top of that.
You know, but now imagine an agent,
you know, doing the work instead.
You know, the agent just-- it really
doesn't care about, you know, UI.
It doesn't care about the
interface of the product.
The agent, you know, really does not
care which vendor's login scr-screen,
uh, it's, it's truly looking at.
Um, it only really cares about, you
know, one thing, you know, did the
task get done and done, you know,
proficient to the goal at hand?
And that is a completely
different customer.
Um, you know, if you think about
it, humans optimize for convenience.
Agents, I think, optimize for performance.
And so there's just channels that they
go through, and, you know, they're
just ultimately trying to perform,
uh, a task and get it done in, in
a, you know, in a proficient, easy,
cost-effective manner, assuming that's
all, you know, part of the goal.
And when the customer changes,
the product changes, right?
That's how evolution tends to
happen, whether it's technology
or, you know, other parts of life.
And I think I'm coming back to, you know,
the orchestrator ultimately becomes the,
you know, the OS, the operating system.
And here's kind of the mental model or
mental picture, um, you know, I think
about, which is think of AI orchestration
as that new operating system.
The software you used to open is
now the application The application
exposes capabilities, the
capabilities, uh, compete really,
uh, and then the orchestrator
decides which one it gets called.
And again, that may sound a little
abstract, but it's very simple.
It's again, application exposes
capabilities, capabilities,
uh, compete against each other.
An orchestrator, i.e.
AI, uh, chooses which one gets
called based on, uh, you know,
probably numerous factors.
And we've seen this movie before,
you know, we just haven't seen
it in, in software specifically.
You know, you think about the App
Store, you know, before the App
Store, you had software, and after
the App Store you had apps, right?
So before the App Store you had software,
after the App Store you had apps.
So same technology, different
units, and I think we are about
to see the same shift again.
And we're seeing it incrementally
if you're paying attention to what's
happening at different, uh, you know,
software companies and companies
that are, you know, new startups.
Um, and 'cause before, look, AI
orchestration, again, you had platforms.
After AI orchestration,
you have capabilities, and
capabilities are apps, right?
Um, and that's the way to think about it.
And every big software company, uh,
is going to have to figure this out,
uh, which of those capabilities are
worth exposing on a standalone basis.
And again, I don't think
platforms go away by any stretch.
I think humans are creatures of habit,
creatures of workflow, creatures
of governance and, and guardrails.
Uh, not only humans, but also obviously
enterprises that are run by humans.
But in general, I, I don't
think that's going to change.
But I think we will start to see a
landscape, uh, that is fracturing
on the, on the edges, and then as
time goes on, it works its way up.
Um, and so, you know, I, I, I
obviously want to be careful
there when you think about it.
You know, I'm not saying
platforms disappear.
They do not.
Um, you know, they become
command centers, right?
And you've heard that from like
ServiceNow, uh, you've heard that
from Salesforce and some others,
and I think the smarter companies
are, are thinking this way.
And ultimately, you know, we've had our
software piece where we wrote our article
and, you know, the conclusion there was,
you know, the companies that do this
become a command center ultimately, uh,
have the opportunity, again, keyword
is opportunity, to become multiples
the size of where they are today.
Um, and I think that's important, you
know, to have as, as fa- fabric here.
And look, there is still a lot of
stuff that has to live somewhere.
You know, e- everything from permissions
to, um, you know, security data, uh,
governance, audit trails, uh, compliance.
I was just on a call with somebody that
is, you know, the head of compliance
at a large, large, you know, public
company, and that was the whole
focal point was permission, security,
governance, audit trails, compliance.
Uh, I don't think that goes away,
and ultimately, that enhances and,
and helps the incumbents for sure.
Um, and enterprises still kind of
need a place where all of that lives.
So the platform stays But what happens on
the top of the platform, I think changes.
The features stop being, you
know, locked inside, let's say.
The features start being called from,
uh, you know, the outside, let's say.
Uh, and I think if you go back in history
and you look at payments as an example,
that was ultimately the sign of what
I think is, you know, taking place.
Take Shopify, started
as e-commerce software.
You bought Shopify to run a store.
That's it.
Then Shop Pay showed up.
Shop Pay, uh, you know, we all
use it in some form or fashion,
whether, you know, we don't even
know how the app was downloaded.
Um, but it is an easier means of,
um, you know, specifically you're
here in the US, you know, an easier
means of tracking, you know, some
of the stuff you're buying online.
And Shop Pay really, again, started
to show up everywhere on sites that
had nothing to do with Shopify.
You know, the capability, um, you know,
or their payment capability, you know,
escaped really the platform, right?
So again, you had a platform that then
you had this, um A composable element
to it, which was its payment solution
that started to g- move away and,
uh, drive value back to the platform.
Uh, same thing with Stripe.
You know, Stripe was not
sold as, as software.
Stripe was sold as just an API.
Uh, developers do not buy
Stripe, they call Stripe, right?
Uh, Twilio, same thing.
Communications became, you know,
infrastructure, um, and really
not, you know, about software.
Uh, and I think that is the model.
You know, the feature ultimately becomes
valuable enough that, uh, it no longer
needs to sit inside of anyone's, uh,
suite anymore, and that's the power.
That's the power where ultimately,
you know, the use case goes from,
"Hey, this is a nice feature to have
on my platform," to, "Hey, this is
a sole product that can drive a lot
of value incrementally to others."
And that's where you, you start
to get the multiplier effect.
Um, you know, Zoom has meeting summaries,
uh, phone AI, uh, a virtual agent.
Uh, those are capabilities.
Those are capabilities.
Right now, uh, they mostly live inside,
uh, you know, a Zoom single subscription.
You know, but there's-- is nothing
physically stopping them from being
called by any orchestration layer
that wants to summarize a call.
And, you know, that's important, right?
So we're starting to
see some elements there.
You know, they have Zoomtopia coming
up, and obviously we're investors in
Zoom, so we follow it closely, uh,
and I'll talk maybe more about that.
You know, Box has document
intelligence and enterprise retrieval.
You know, that could live outside,
uh, you know, Box's workflows.
You know, we know Salesforce, you
know, has a large customer graph.
We know Adobe has, you know,
incredible creative models.
You know, HubSpot has
its marketing engine.
You know, every one of these companies
has capabilities, and again, maybe
not specifically to the ones I just
mentioned, but they do have capabilities
that sit inside their platforms
that could escape the platform.
And I think a lot of them will,
uh, because it's going to be more
valuable to be the tool every agent
calls than to be the platform, uh,
that, you know, every user, you know,
is sitting there avoiding, right?
So you want to be the thing that is being
used more than trying to be rigid and
assume everyone's gonna buy your platform.
Very few win the platform game, and
that's why it's so, uh, powerful to
be a platform in era two point oh.
And in era three point oh, you know,
again, you're going from Apple to apps.
Uh, yes, you wanna be Apple, but you also
wanna be the top apps in the App Store
because that puts you front in line.
'Cause eventually, like everything
else things start to get static
again, and, you know, whatever you've
built, you're comfortable with.
And that doesn't mean an agent
can't go find a solution that is
better, uh, different, cheaper.
And yes, you may, you know, think
about how some of the stuff gets
commoditized if this is the way we go.
But ultimately, again, it go-- comes
back to governance and some of the
other security elements that, um,
will have to live, and you will have
a trust factor between you and the,
the company that has the feature that
is now, uh, what an agent is calling.
And so, like, what does
determine these winners?
You know, so if we are moving into a
world where AI is picking the tool for
the job, then what actually matters?
You know, because, you know, the,
again, the old, like, scorecard here
was, uh, it was much different, right?
It, it was software winners were built
on distribution, uh, you know, big
sales forces, you know, big brands,
you know, slick UI to some degree,
especially early on, uh, you know, deep
integrations, whether it's with, you
know, their consulting firms and/or,
um, in, in just generally in the
ecosystem and, you know, uh, bridging
the gap for integrations with, with
the other large software vendors.
And, you know, contracts that were
long in length, which allowed you, you
know, three, five years to, uh, you
know, hold a customer, create, um, you
know, new solutions, and then try to,
um, renew and upcharge, uh, thereafter.
And that mattered because, again, a
human was making all the decisions.
And again, I say all very, uh, pointed
because, again, I don't think this is
gonna be, you know, one or the other.
It's gonna be the combination, but I
think it's, you know, what we're trying
to do is open the eyes to the idea here.
Um, and again, humans
respond more to distribution.
But an agent, you know, doesn't
really d- respond to distribution.
You know, an agent responds
to whether the tool works.
Um, so the new scorecard will
be, you know, much different.
It'll be everything around, you know,
actual reliability, actual latency, actual
accuracy, actual, you know, security and
governance tools that are sitting there.
You know, it'll measure and, you
know, the, the companies that are
managing these agents will, will,
will focus on cost per call, you know,
the quality of the API, you know,
downtime versus, you know, uptime.
And, you know, we all-- If you, if
you play enough with these agents, you
know, you're constantly having to, you
know, reorient your credentials on APIs.
Um, and ultimately, it's, you know,
gonna be cost and then success,
uh, rate, uh, on each task.
And ultimately, that is the,
uh That is the path here.
That's the scorecard.
Developer experience will matter too,
obviously, because there'll be humans
involved in, you know, different ways.
Uh, and over time, again, feedback
from agents themselves, which is,
you know, sounds kind of strange,
but, you know, I do think, you know,
there could be a feedback, you know,
marketplace review system where agents
are giving feedback, you know, generally.
Again, it doesn't have to be like,
you know, how we think of feedback.
It's just more, you know, um, you
know, almost like, uh, you know,
the data that, you know, our, our
devices provide to Apple is, you
know, feedback to them that then they
use to then ship better products.
So, um, again, I think exposing
the capabilities drives more
usage, drives more feedback.
That drives a better model for,
you know, the end model companies.
And then, you know, that ul- uh,
ultimately then drives, you know, higher
rankings inside of an orchestrator,
which then means more calls to it.
And it, and it becomes kind of like this
feedback loop, again, of more usage.
And that is less so, uh, what we learned,
uh, you know, in past, uh, iterations
of, you know, the platform effect.
This is truly like a platform
product network effect, uh, at play.
And it, it's definitely something
new, just given the fact that
it's, you know, not directly only
tied to a human-to-human, uh,
judgment and, and, and buying.
And, you know, the two questions I think
I keep thinking about though is, you
know, question one is, you know, if AI
disappeared tomorrow, uh, would, uh, you
know, customers still buy this platform?
And again, I'm, I'm looking at this
from the standpoint that, um, you
know, if you're looking at a software
company right now, again, if AI
disappeared tomorrow, would customers
still buy, you know, your platform?
And if the answer is yes,
obviously that platform is durable.
It has a real customer value that, you
know, really has nothing to do with AI.
Uh, and if the answer is no, I think, you
know, there is a problem because you are
not really, you know, selling software.
You're selling kind of a wrapper.
Um, question two is, you know, if
AI accelerates, you know, which
capabilities inside the company,
you know, really deserves to become,
you know, a standalone business?
Um, 'cause there's probably, uh, you know,
the various, you know, growth engines.
Um, and in this case, you know, those
ones that I'm referring to, you know,
those company capabilities become, uh,
you know, growth engines, you know, not
the suite, you know, the specific feature
that is good enough to escape it, uh,
again, back to, um, you know, kind of
like the Shopify, uh, example on Shoppe.
Uh, but every management team really
should be, uh, able to, you know, answer
one of those two questions, uh, and, and,
you know, what is our strongest feature?
You know, what is our strongest data set?
Um, you know, what models have we
created in-internally, you know,
that are strong and, and can, can
be, you know, potentially used?
And, you know, do we have strong
APIs, uh, whether it's, you know,
platform level or feature level?
You know, and if you can't answer any of
those, I think, um, you may be in trouble.
Uh, that's how we're looking at companies
today, which is, again, across that chain,
you know, if, um, any of those are nos
or yes, you know, h-how does that, uh,
you know, suggest where you should look?
Um, where we could be wrong, I guess,
on this is, you know, if we pressure
test this, um, because again, I
don't wanna, like, overplay the idea
that we're going only to an agent,
you know, uh, uh, picking world.
You know, first, you know, enterprise
buyers might just, you know, keep
buying, you know, platforms and suites.
Um, you know, it's convenience.
Convenience does matter.
You know, compliance again matters.
You know, nobody got fired, you know,
generally speaking, for buying Microsoft.
Um, we know that's real.
Um, second is, you know, if I'm thinking
out loud, you know, Microsoft specifically
could, you know, just absorb, you
know, the entire orchestration layer.
You know, we know Copilot, you know, is
a dud here for the first three years.
You know, but if it, if it, if it
does become the orchestrator, then
Microsoft is not losing, uh, you know,
to, you know, composable software.
You know, Microsoft then becomes the
composable software to some degree.
Um, I'd say third, you know, security
and governance may just ultimately
limit, um, you know, how much third-party
orchestration, uh, is actually allowed,
especially in, like, regulated industries.
Um, you know, there's obviously real
reasons capabilities might have to
stay inside a platform, so the platform
still exists just given the fact that,
uh, it, it's gonna be hard for, uh,
a legacy incumbent built, uh, around
governance and compliance, uh, with
thousands of customers and thousands
of employees would be able to do that.
Uh, specifically, uh, you know,
with any timeframe that is, you
know, within one to a couple years.
You know, fourth, I think, um A
lot of features are, you know,
not actually differentiated.
Um, and you would argue, like if you go
into most software companies, you know,
most of their software is pretty average.
You know, most, most of those features
are, you know, somewhat average.
That's why, you know, the term
best in class or best in breed,
you know, has come about.
And I think actually like those companies
that have built those best in breed
platforms have a opportunity here, um,
i-as long as there's a runway to continue
to develop, uh, you know, that feature.
Uh, again, if you're not best in
class on a feature level, then I
think you're in big, big trouble.
Um, and, you know, again, I don't
think it's a clean story where
platforms lose and capabilities win.
I just do think it's more of a hybrid
over time, that platforms do sit today,
but they matter for different reasons,
and capabilities become more valuable,
you know, really than they used to.
Uh, and you have that
composable software element.
And the bigger picture, you know, if
you kinda like step back, um, this is
kinda where I land on it all, which is,
you know, I think AI is going to spend
the next 20 years making, you know, many
platforms, um, modular, uh, composable.
You know, the winners are not necessarily
going to be, you know, the companies
with the biggest suites and platforms
as much as they were in the past.
You know, they'll probably be the
companies with individual capabilities
that become indispensable, uh, no matter
where the user starts its journey, and
that's a, a much different game overall.
And most companies have
not adjusted to that yet.
I brought up the Zoom example because
now they have Zoom Notes, Zoom AI Mate,
which are now sol- single products
that you can buy on a standalone basis.
And for me, looking in, like, you
know, from the outside looking
in at that, it's, you know what?
An actual somebody that is building
a agentic, you know, AI-first
company is likely going to be tool
calling and calling on features
and best-in-class features.
And those companies that have all and
many of their features that are best
in class exposed to the agentic world
will have the ability to likely, you
know, position themselves at the top
of, you know, the App Store as that,
you know, idea continues to evolve.
You know, the first apps in the App Store,
if you think about it as a history, you
know, not all of them are obviously at
the top, but, you know, if you go back
ten, 15 years, you'd argue, you know,
the top 50 haven't changed all that much.
Uh, and if they have, they're coming
generally from the same companies.
And so again, it's all about composable
software, where if you can make
best-in-class features like, you know,
AI Zoom, uh, summary has been really
good, and now you can buy that on a
standalone basis as opposed to forcing
your way for people to buy the platform.
So again, having multiple ways
to win is super, super important.
Uh, a few frameworks to, you know,
remember here, you know, if you,
if, if you take away one thing from
this episode, take one of these.
You know, humans buy
platforms, AI buy capabilities.
You know, platforms become,
you know, command centers.
Features are ultimately, you
know, the products, let's say.
Um, I've talked about this over and
over again, but the orchestrator,
the orchestration layer is arguably
the most important thing, and that
becomes the operating system, um,
really the OS to the whole ecosystem.
And that's why we're so, um, um,
focused on that part of the stack.
You know, everyone's focused
on infrastructure, models.
You go above that, orchestration.
You go above that, you go
applications and, and some other
things, uh, that you could, uh,
you know, you could assume there.
But really, every great software
company really should be asking,
you know, what capability of ours
really deserves to escape the system?
And again, if that answer is yes and you
have many or some, you're in a good spot.
If it's not, then you're selling
a platform where none of the
features or products inside
that platform are any good.
I think you're in trouble.
Um, and again, the future is
not fewer software vendors.
It's probably more, but it's more
specialized capabilities connected
to AI that are best in class.
So again, the question is not
whether software platforms disappear.
Um, it's more about, you know,
everything moving away from what has
been the past , the eras one and two
and moving to era three, where we're
talking about a structural shift of
buying capabilities versus platforms.
And that is what we wanted to
talk about, bring that to you.
If you have any thoughts,
I mean, on it, let us know.
Make sure to subscribe so we have
more kind of open discussions like
this around the desk and give you
more insights to what we're thinking
about here at Avery and Company.
Make sure to check our
website, uh, averyfunds.com.
Make sure to check our YouTube.
Subscribe to us on Apple or Spotify.
Thanks for listening.
See you next time