Agent Force is
the culmination
of Salesforce innovations
from workforce automation
It comes in the middle
of a technological shift
as more and more
companies employ
AI to help their workforce
create a better customer
To hear how Agentforce
can help you,
I'm very excited to be
joined by Patrick Stokes.
He's Salesforce's EVP of
Product and Industries
I liked your presentation
during the keynote earlier
So let's back up and do
like a Agentforce 101.
We started with
generative AI.
We're now in
autonomous AI.
What exactly is Agentforce
and how does it work?
We're starting with the
hard questions first.
We'll do the all
inclusive one first.
So I mean, look,
we all have--
we were talking about
this before we started.
We all have
jobs that we're
trying to do in our
life or certain tasks.
And within that, there's
probably a lot of stuff
that we don't
really like doing,
or things that we're not
particularly good at,
or that are hard
for us as humans
or just for our
own personality.
And what
Agentforce does, it
gives us an ability
to kind of go do
those jobs in an
autonomous way.
Now that sounds a
little bit crazy.
It sounds a little bit
like science fiction,
The simple way to
think about this
is it's just
kind of like--
imagine a little
piece of software.
But imagine if that
software could think.
And imagine if
it could go out
and get the data
that it needs to help
Imagine if it could even
look at the question
and try to understand does
the question even makes
sense in the first place.
And then it can go out,
and think, and grab
the data, and
then, ultimately,
And maybe, in some
cases, the answer
to the question you
asked for is actually
a plan or a series
of actions or tasks
And then imagine if it
could do those tasks
and then come back and
think about the question
So it's just a
piece of software
And when you put that all
together, what you get
is this idea of something
that is effectively
You give it the
data that you want.
You kind of describe
the environment
that you live in or
that it lives in.
And then you say go and
it goes and does stuff.
How is Agentforce
different from
So the way to think
about Agentforce here
is there's going to be
many other autonomous
kind of agents
that that pop up.
But what's so incredible
about Salesforce
is to get those
autonomous agents to work,
it really all
starts with data.
You need data and
then you need action,
so the workflows
or the business
processes on
the other side.
But without data, you're
not doing anything.
It's similar to bringing
a new human being
You're not just
going to say go.
You're going to sit
them down and say,
we're going to
have an onboarding.
Let me teach you
about the company.
Let me teach you about
the way we do things.
Here's where you find
this information.
Here's where you find
that information.
That's the way to
think about getting
And what makes our agents
or Agentforce so powerful
is Salesforce has
already built that kind
So it's not just
about the AI.
We spent the last year
talking about, oh,
How many parameters
do the models have?
It's really not just
about the models.
It's about
bringing the data,
connecting it
to the models,
and then, ultimately,
connecting that
And what
Salesforce has done
is wired up that
complete system.
So when Marc talks
about don't DIY your AI,
it's because you actually
have a system in place
and you don't have to
kind of do it separately.
Another way to think
about it is, you know,
An automobile is a system.
And some engines are
faster than others.
But at the end of
the day, your car
needs a steering
wheel and some wheels.
And then you're going
to need to optimize
You can have the fastest
engine in the world,
but if your car
weighs 20,000, pounds,
it's not going to
accelerate very fast.
So you want to have that
fully optimized system.
Last year, we were
talking about Copilot.
This year we're talking
about Agentforce and AI.
I guess what's
the difference?
How are they
very different?
Is it the autonomous
part of it?
It's largely the
autonomous part
but it's also connected
to that broader system
So, you know, we and many
others, everybody kind of
rushed to this when we
saw these LLMs come out.
We looked at them as
humans, as people.
And we were like, oh
my god, these things
I don't have to
frame my question
I can just kind of
talk to it naturally
I can brainstorm with it.
It can help me write
doc, things like that.
And so we all
rushed to kind of,
to be plainspoken
for a bit,
just like jam those
LLMs into the right rail
We wanted to put them
there to make it easier
to not have to leave the
application we were in
and open up a new tab
and open up ChatGPT.
We just wanted it right
there in our application.
But what everyone
discovered really quickly
is, well, this
thing doesn't really
know anything
about my business.
So I'm asking it
questions but I
have to like
continuously give it
all of this context
in order to get--
and that's fairly
tedious to type
And then we
learned, well, if we
want it to do something,
it can't really do it.
It might tell
me how to do it,
but I'm still the one
clicking the buttons.
And so what makes
Agent so much different
is it's the combination
of that complete system.
It evaluates all
of that data.
And then it's able to
go out and actually
perform actions in
an autonomous way
but not always
autonomously.
In some cases, it might be
reactively or assistively
to what a user on the
other side is asking.
Well, let's
talk about that,
because I know there's
been a big focus of humans
along with agents that's
been underscored a lot
How will they
work together?
So I mean there's
a lot of things
that humans are
very good at.
There are some things
that humans are not
There are some
things that AI
is very good at
that humans are not
The simple
example here is,
as humans, we're really
good at creative thinking.
We're really
good at coming up
AI is not so good at that.
You may set it
out on a task,
but it's going
to kind of follow
the description
of the task you
It's not very good at kind
of just randomly taking
inspiration from the
world and coming up
with a different
way to do that.
On the other hand,
humans are very, very bad
at kind of looking at
lots and lots of data
and synthesizing
that data.
So you might have a rise
in call center volume,
A lot more customers are
calling in than usual.
And you may ask someone
on your team, hey,
Well, that human,
that's probably
They've got to get
the transcripts
They've got to kind
of comb through them,
read them, figure out what
the patterns are in those.
That's a tedious
conversation
but an AI can do
that in an instant.
So creativity kind
of humans, data
And so if we
can kind of find
the right optimization
of the tasks
that should go to the
humans and the tasks that
should go to the AI, or,
in some cases the tasks
that should
intercept both,
you suddenly have a much
more optimized system.
I only have
about 20 seconds
left for your answer,
but I'm curious.
Marc started his keynote
with core value of trust.
How do you make sure
that people have
Well, so from
the foundation,
when we talk about
bringing data, so
all of your data,
if it's already
sitting in the
Salesforce platform,
it already has all of
the permission models
and sharing
privileges on top.
So that gets
you a long way.
But we also need to
be able to observe.
We need to be able to kind
of supervise these agents
We need to know when
they're hallucinating.
We need those
triggers so that we
can go back and
start to optimize.
So this is what we
mean by that full end
You're not building all
of this from scratch.
Really fascinating,
Patrick Stokes.
Very nice to talk to you.
