Play
Show Navigation
Show Transcript
00:00
Play
Seek 10 seconds backwards
Seek 10 seconds forward
00:00 / 01:23:44
Mute
Disable Captions
Settings
Picture in picture
Fullscreen
Search in video
You can search the video captions for specific words or phrases.
Popout transcript
Hide Transcript
Skip transcript
- So who are the
Agentblazers?
[MUSIC PLAYING]
Well, now that's a story.
Trailblazers,
they've led the way.
People who saw a
mountain and said,
hey, I'll climb it.
The people who saw a
challenge and said,
hey, let's solve it.
They were pioneers,
innovators, visionaries.
But today, today,
something new
is on the horizon.
It's not just about
blazing a trail anymore,
it's about building
a world where
humans and digital
agents work together,
where empathy
meets efficiency,
where intelligence
is autonomous.
We call them
Agentblazers, a new breed
of problem-solvers who
cut through the shiny hype
of AI and answer the
real question of what
AI can actually do.
I mean, what if
workforces had no limits?
What if customer
service could
feel like a conversation
with an old friend?
What if businesses
could make smarter
decisions in real-time?
Agentblazers
see the future
not as something
that happens to them
but rather something
they can create.
Now look, we've
been dreaming
about agents for years.
We've seen them
in the movies--
robots, avatars,
futuristic assistants,
but this, this
is different.
This is real.
Agents who learn
your business.
Agents who know
your customers.
Agents who can
take action right
when you need it most.
Agentforce is about
empowering people,
freeing humans to focus
on what really matters.
Yeah, our stuff.
And the best part?
The admins, the
developers, the dreamers,
and the Salesforce
ecosystem,
you're the ones turning
this vision into reality,
because Salesforce didn't
just build this platform
to automate tasks.
No.
It was built
to empower you
and people like you to
become Agentblazers.
So who are the
Agentblazers?
You are the Agentblazers.
And the really cool
part, the future, it's
already here, right now.
[END PLAYBACK]
[APPLAUSE]
Please welcome EVP Product
and Industries Marketing,
Salesforce.
Patrick Stokes.
[APPLAUSE]
All right.
Good morning, TDX 2025.
Thank you all for being
with us here today.
Man, you all look so good.
This is awesome.
Thank you to the thousands
of you that are here
with us in the room today
and on the event, really
appreciate you being here.
Thank you to the
thousands of us.
Where's the camera?
There's a camera
watching--
tens of thousands
watching online.
Look, we have quite
a bit to do today.
It's going to be a
really, really busy day.
And I'm going
to tell you why.
Six months ago, we
launched Agentforce
just right
across the street
in Moscone South, North?
Where's Jacob?
North, I think.
We launched Agentforce
right across the street.
And we showed the
world what enterprise,
what trusted enterprise
AI agents could look
like in your workforce.
And then, believe
it or not,
just a few months later,
across another street,
at the St. Regis
Hotel in December,
we launched
Agentforce 2.0,
showing what it looks
like unwrapping the Atlas
reasoning engine
and showing you
exactly how
Agentforce can reason
and think and
actually, take action.
And since then, not
a day has gone by,
I swear, not a day
where I don't wake up
and there's some message,
some new news report
of a new AI model or some
sort of new technique
or new application
in how we use them.
It's really
hard to keep up.
The pace of innovation
has been incredible.
So that's what makes
this year's TDX,
I think, more important
than any we've
had in a really, really
long time because we all
are about to go on
an incredible journey
together, an incredible
journey where we use all
of this new innovation.
And our job here is to
show you how to use it,
to show you
what it can do,
and to bring you
along on that journey.
And my job as the host
of TDX of this developer
conference is to get you
to that innovation as fast
as possible.
So let's go ahead
and do that.
Now, of course, let's
start by thanking you.
Thank you to our
customers that have been
with us on this journey.
Thank you to our customers
that are here with us
today, those that
have already started
their Agentforce journey.
Thank you to all
of the builders,
our Trailblazers, our
Agentblazers, developers,
admins, our community,
our architects.
Thank you all.
Thank you to our
partners, our ISVs.
Thank you to all
of our employees
that are waking up
every day dealing
with this pace of
innovation and building
so that all of
us can use it.
We really appreciate it.
And of course, when
we start any journey,
we have to start by
just reminding ourselves
of our values.
This is something we
always do at Salesforce.
Our values of trust,
customer success,
innovation, equality,
and sustainability.
These are what guide
us through everything.
And we're going to
need it more than ever
as we go into this
path, this path where
we're bringing AI
into our workforce.
We're going to need these
values more than ever.
Now, when we started this
company 25 years ago,
I wasn't here, but Mark
and Parker were here,
they started with a
really simple idea.
They called it
the 111 model,
this idea that we
would give back
1% of the
company's equity,
their time, their
product back.
And I'm really
proud of the impact
that this has
had globally.
But the number
on this slide,
I've been at Salesforce
for 13 years that always
makes me so proud is
the 19,000 companies,
the 19,000 organizations
that have adopted this
model as well and
come along with us.
And it's because
of all of you
that we're able to do
well while doing so good.
And it's been an
incredible journey
for Salesforce it's
because of all of you,
because of our customers,
but really because of all
the builders in this
room, the ones that
build and implement
Salesforce and bring it
to life and build these
customer experiences that
will do $40 billion
in revenue this year.
It's just incredible.
OK.
But we have to talk
about something
new here for a second.
And I thought about
this for a long time.
And look, this is a
developer conference.
So we need to get
into the nitty-gritty.
We need to get
into the stuff
that developers
really think about,
that they really
care, that they really
think about
every single day
and wake up
thinking about.
And I thought about
what could that be.
And I realized
it's pretty simple.
It's global declining
population growth.
[LAUGHTER]
That was a joke.
You can laugh.
So look, this is
actually not a joke.
This is really important.
All around the
world, countries,
their population
growth is slowing.
Now, as developers, as
builders, why should we
care about that?
Well, if you
give me a minute,
if you just bear
with me for a minute,
I'm going to try
to tell you why you
should care about that.
It's actually
pretty simple.
Population,
people, us, humans,
we make companies grow.
It's as simple as that.
And if there
are less people,
that's going to create a
plateau where businesses
are going to
struggle to grow.
In fact, I like to think
about three things that
are directly correlated.
If we think back
hundreds of years,
there are three
things that
are always related
to businesses
being able to grow.
One is people, we
just talked about.
The other, this is
a little bit hard
to think about,
is real estate.
And the third is
infrastructure.
Now people we can
understand, let's
think about real estate.
In order to grow,
well, first of all,
if we need to
hire more people,
we need more office space.
But we might also
need more stores.
We need to enter
new markets,
so we need more stores.
And infrastructure,
well, we need compute.
We need storage.
We need to build
more software.
So those three numbers
are always going to go up.
And if we think
about those three,
let's think about--
let's take infrastructure
for a moment.
Our friends at
AWS and Google,
they've built this
incredible capacity
to scale infrastructure.
We can now scale
the software
and the infrastructure
that we need infinitely
and horizontally.
We've done the same
thing with real estate.
It's a little bit
hard to think about,
but you all have really
been responsible for this.
Let's think about it.
If we want to enter
a new market now,
25, 30 years ago, we
needed a new office,
we needed a new store.
Well, today we just
need a new website.
We need a new application.
So we've figured out how
to scale that as well.
But the one thing
that we haven't truly
figured out how to
scale yet is the people.
And that's where
you all come in.
What if we could
use software,
what if we could
use technology
to scale human capacity?
That is the promise
of agentic AI.
And it is creating a
digital labor revolution.
This idea that we
have human workforces
really exclusively-- today
they have fixed capacity.
The thing about humans,
if you're like me,
is you get tired.
You get burnt out.
There's only so many
things you could do.
But what if we
could pair humans
with smart, intelligent AI
agents that can actually
go out and do work
on our behalf, that
could work with us, that
could partner with us?
Well, then we would have
unlimited experts, not
just the agents, but think
about what an AI agent can
do for you.
It can make you an expert.
Our customers,
they don't have
to wait on hold anymore
for another human,
they can just get the
answers that they need.
We could unlock growth.
We would have
hit all three.
We can scale our
infrastructure.
We can scale
our real estate
through our web
applications
and through our
digital experiences,
but now we can also
scale our people.
Now to do this,
we need something.
We need some technology.
And it's really, really
easy to think, well,
the solution must be a
large language model.
That's what they're going
to tell us here, right?
It's like we're going to
just use a large language
model.
And it's going to
work and it's going
to change everything.
And look, this
is a revolution.
This is a revolutionary
piece of technology.
It's this thing that
gives a little piece
of software.
It's not that little.
It's actually pretty big.
But it's a piece
of software
that can understand
human language.
It can converse with you.
It can almost
even fool you.
So we'll just slap a large
language model into it.
Well, it's not
that simple.
You see a large language
model, it's not enough.
It's certainly not
enough for digital labor.
We need something
much bigger than that.
If you think about
a large language
model for a moment,
it's a lot like a CPU.
A CPU was equally
revolutionary.
It could read and
understand computer
language, binary
0's and 1's.
It could run really
complex computations.
It absolutely
changed the world.
And today I
could put a CPU
in the palm of your hand.
In fact, I
tried to get one
and I thought it
would be cool.
I could pull one out and
have a CPU in my hand,
but I'd mess up
all the pins.
But you could put a CPU
in the palm of your hands,
and you know what
it would do for you?
Do you it would do
for your company?
Nothing.
It wouldn't do anything
because the CPU needs
to be part of a
larger system.
It needs to have a
motherboard and storage
and memory and an
operating system
and applications
built on top of it,
and peripherals so human
beings can interact
with it like a keyboard
and a mouse and a computer
screen.
It needs to be
part of a system.
A large language model
is exactly the same.
So what kind
of system do we
need for a large
language model?
Well, there's
three things.
We like to call
this our Trinity.
There's three things that
you're going to need.
The first is data.
You've got to get data
to these large language
models.
You've got to get
your business data.
We've been talking about
this for a little while,
but it's really important.
These large
language models,
they've been trained
on the internet.
They don't know anything
about your business,
so we need a way to
connect the data.
And not just your
structured data,
but all of your
unstructured data,
your knowledge
articles, your metadata,
your semantic data
that describes
what that data is
in the first place
so that the AI
can understand it.
So that's the first
part of the Trinity
is your data.
The second is you
need your apps.
So you need all
of your line
of business applications,
your industry
applications,
your analytics.
You need all of your APIs.
You need to be able
to connect all of that
so that your AI can
actually take action.
And then the
next thing you
need, of course, is your
AI itself, your agents.
We need a way
for the AI agents
to connect to those APIs,
but we need a way for it
to also connect to
all of that data.
We need some technical
techniques like retrieval
augmented generation.
We need things like
vector databases.
We need the ability to
loop and the ability
for the AI to
actually reason
if we want it to come up
with good and accurate
answers.
And these things are not
terribly easy to build.
They're not terribly
easy to bring together
yourself.
And that's why we've
done it for you.
We've brought all
of this together
on one trusted and
deeply unified platform.
And we've added all
of the trust layer,
all of the security and
permissions and governance
that you need
because if someone
in your organization
asks an agent a question,
you want to make sure that
the answer is tailored
to that individual.
You don't want to
give away secrets
from all of your data.
So we really need to
think about how do we
apply all of that trust.
But if we have one unified
platform as developers,
as builders,
look, we're always
a little bit worried
about vendor lock-in.
And do I have
to do everything
with one company?
No.
And Salesforce knows that.
And that's why we've built
this entire platform,
this deeply
unified platform
to be as open as possible.
You can bring any LLM.
An LLM that works
for you may not
work for your
organization.
You can bring any LLM.
You can bring any system.
You can bring any
API and hook it up
to this new platform.
You can bring
all of your data.
And you don't have to
physically move the data.
You can use zero-copy
and connect the data,
leveraging the investments
that you've already
made in data lakes
from our friends
over at Snowflake and
Databricks and others.
And of course, you can
run this whole platform
in any Geo that you need.
Now, we put this
all together.
We put this all together
into one platform,
one Trinity of data
and apps and AI agents.
And we call this
platform Agentforce.
This is our digital
labor platform.
It's trusted.
It's easy to deploy.
It keeps humans
in the loop.
It eliminates
that hold time.
This is one deeply
unified platform,
making it easier
and faster
than ever to build
these agents.
Now, customers that are
using this platform that
are transforming all of
their work, their customer
workflows, their
employee workflows,
and transforming
everything
with agentic capabilities
and reasoning,
we call these companies
Agentforce companies.
RBC is a great example.
This is one of
my favorites.
They have these
wealth managers.
These folks wake up every
morning thinking about,
how do I advance my
customers' portfolios?
How do I help them
manage risk and help them
find new opportunities?
That's a hard job,
especially when
you have thousands of
customers in your book.
Well, now Agentforce is a
partner with those wealth
managers.
It's helping do all
of that work for them
so that they can call and
talk to their customer
about exactly
what they need
to talk about that day.
reMarkable is
an Agentforce
company. reMarkable
is awesome.
I use this every
single day.
It's these new
e-ink paper tablets.
They're awesome because it
strips away all the stuff
that we usually
have on our tablets,
and it just focuses you
on writing and reading.
Those are good
habits, by the way.
Writing and reading.
I encourage that
instead of scrolling.
And reMarkable
does that as well.
It's an incredible
company.
These are amazing devices.
But their problem is
they're growing so fast
and they wanted to keep
their customer service
just as good
without having
to grow all of the
humans that they have.
They wanted to
bring Agentforce in
to be a partner so
they could provide
that white glove customer
service experience
that they need while
they were growing.
Now, how does
this all work?
Well, it sounds
pretty sophisticated.
And frankly, it is
pretty sophisticated,
but it's actually
pretty easy
to think about how
this all works.
It starts with an input.
Now, we've been used
to seeing those inputs
as chat inputs.
You come and you talk.
And you're going
to see that today,
but you're going to see
much more than that.
You're going to see inputs
from APIs, from events,
from things like
change data capture.
We're going to be
able to connect
this thing from anywhere.
Then reasoning.
And there's a little
fancy graphic there.
And it's like,
yeah, trust us.
It reasons.
There's a little
black box.
I'm going to break that
down for you in a moment.
Then there's action.
We want it to be able to
actually go out and do
things.
So we have our action.
We can even enrich that
action with Tableau.
In other words, we
can do the action
and then look
at the outcome
and tell you the
outcome that that's
had on your business.
And then, of
course, there's
the outcome itself, which,
in some cases, might be,
again, another turn in
the chat experience,
but it might also be
some sort of execution
in another system.
Now that reasoning
there, how on Earth
are we thinking?
Well, it's complicated.
But it's not.
It's really just a loop.
It's just a loop.
It just loops.
It thinks the same
way a human does.
When I ask you a question
or when I give you a task,
you don't just immediately
come up with an answer.
You think, do I
understand the question
in the first place?
What data do I need to
understand this question?
Maybe now that
I've got some data,
I realize I need
some other data.
You're just looping.
Atlas is doing the
exact same thing,
and it's because
of that, that we're
able to deliver incredibly
trusted and accurate
answers.
Basically, the
more you think,
the better the answer is.
Funny how that
works, right?
OK.
Now how are customers
getting started?
Well, we've done thousands
of implementations,
not just with
our customers
but with ourselves.
You'll see this
later today
with help.salesforce.com.
And the simple way
to think about this
is the first agent you
want to try to build,
it's really simple,
just an agent that
can answer questions.
But not just any question.
ChatGPT can answer
any question.
What you want is an agent
that's actually hooked up
to your data that
can answer questions
about your business.
And guess what?
That's not magic.
It's not free.
What you're
going to realize
is some of your data, it's
a little bit out of date.
You're going to have
to go back and fix
some of that data.
You're going to have
to update it, clean it
up a little bit.
But once you're
done, you're
going to get this
incredibly intelligent
agent that can actually
answer questions
about your business, not
just for your customers,
but also for
your employees.
The next step
from there is
you're going to
realize really quickly,
it'd be cool if
this agent could
do something, not
just answer questions
but do things.
So you're going to want
to hook up your APIs.
And then the third thing,
which we haven't really
talked about but you're
going to see it today,
is the ability to take
action proactively.
That's that third stage
of your agentic maturity,
being able to build an
agent that you put inside
of your organization
that can just
take action and
go out and start
working on your behalf.
Now, how do you get there?
Well, we need to move from
traditional companies,
companies where all of our
workflows, our employee
and our customer
workflows are hard-coded
with rigid logic.
We need to move to systems
that reason and learn,
really agentic logic,
nondeterministic.
We need to move from
disconnected data systems
to truly unified
data systems.
So how on Earth are we
going to do all of that?
It seems like we
have a lot to learn.
Well, that's why
we're here today.
The good news is, and
we're counting on you
here, the answer
in how we're
going to do all that,
of course, is you,
our Agentblazers.
You are the ones that
are going to join us
on this journey.
You are the ones that
are going to help move us
from these rigid,
old-world, traditional
software companies to
these new agentic software
companies where everything
is run agentically
where we have humans
and customers working
together.
And to do that, we're
excited to launch
our all-new
Agentblazer community.
We are seeking 1 million
Agentblazers globally
by the end of the world--
end of the world.
Jesus.
[LAUGHTER]
Amazing.
See these jokes?
These jokes
write themselves.
[LAUGHTER]
By the end of the year--
or end of the world,
whichever comes first.
[LAUGHTER]
It was good.
I was feeling we needed
a moment of levity there.
And we're going to talk
a lot more about all
of this new content
and community today.
It's going to be crazy.
OK.
Now, we're about to get
into all of the technology
and the innovation.
But I know what you're
already thinking.
I know what some
of you are thinking
because a lot of you have
thought this out loud
and gone to
Twitter and said,
hey, Salesforce,
this is all awesome.
This is great.
You keep showing us all of
this new technology, but
when are you going to let
us take it home with us?
When are you going to give
us this technology so we
can play and experiment?
And so before we show
you all of the technology
today, I wanted to
just close that gap
and announce that
Agentforce and Data
Cloud are now available
in our Developer Edition.
[APPLAUSE]
So you can get it.
You can sign up,
hit the QR code.
You can even
go downstairs.
Do you remember when we
built agents at Dreamforce
and you would build them
and then your scratch
org would expire
nine hours later?
Well, now you can
take it home with you.
Much better, right?
You get 150 LLM
outputs per hour,
one data space,
and 10 gig of data.
And then here's
the important part.
This thing will never
expire, as long as you
use it every 45 days.
So just log in every 45
days and you'll be good.
[APPLAUSE]
And if that's
too hard, just
write an agent to log
in for you every 45 days
and that'll work as well.
I'm not even kidding.
That can work.
OK.
So we have three things
we're going to walk you
through today.
We are going to show you
how you can agentify,
not just through
chat, but we're
going to show you
how you can agentify
every workflow-- customer
workflows, employee
workflows, you name it.
We're going to show
you, importantly,
how you build these
things with confidence.
What does an agentic
development lifecycle
look like?
And I'm really excited.
We announced
this yesterday.
You might have seen it.
We're going to show
you our all-new agent
ecosystem.
OK.
So a lot to do.
Thank you for
bearing with me.
Thank you for laughing
at my jokes and slip-ups.
And I'm excited
to announce
our EVP and GM of
Agentforce, Adam Evans.
Adam, over to you, buddy.
Cool.
Thank you, Patrick.
[APPLAUSE]
All right, TDX.
All right.
This is my first
TDX in the room
and I thought it was
appropriate to open
with a small confession.
You can ask
anyone on my team,
I am a developer at heart.
I'm a builder
just like you.
And so I'm really excited
not about that alone,
but also that we have
a tremendous amount
of new functionality
for everyone
here to build new
agents and do much more.
Well, first, I want
to talk a little bit
about software.
Everything about
software is
changing-- how we build
it, how we experience it.
And you can think
about this idea
that we've seen for
decades where we have
data and logic and UI.
You could call
it a model view
controller or a
three-tier architecture.
We've had different
names for this over time.
This is fundamentally
changing.
Think about data,
with structured data,
relational databases,
CRUD applications.
This is where we've been.
Now we have new
tools finally
to be able to unlock
the tremendous value
in unstructured data.
And not just in
terms of vector
databases and being able
to query and understand
and reason across
it, but applying
that agentic
reasoning technology
to also onboard
product with data
to understand meaning.
It's unbelievable.
And that also
moves into logic.
Traditionally, we've
been writing code.
It's rigid.
It's all the conditionals.
If this, then that,
as you all know.
And now we're moving
into a new world
where we have not just
deterministic but more
dynamic reasoning
across that it that can
be instructed with natural
language, not with code,
and can handle many
more permutations
of different conditions
out of the gate.
And that leads
me next to UI.
And this is probably
one of the most
fundamental changes that's
happening, where we're
thinking about
instead of building
expensive and
specific UIs that
are these boxes that we're
moving in with clicks,
we're moving into a world
where UI can be developed
at incredibly low cost.
As the cost goes to zero,
the agentic reasoning
can build UIs as you
need in the context
that you need them in,
with visualizations
and more.
And it's moving
from things
like boxes and clicks
to conversation
and to voice and to
be more multimodal.
Everything about
software is changing
with agentic reasoning.
And as Patrick
mentioned, in December,
we launched
Agentforce 2.0.
That was massive
leaps forward
in terms of the Atlas
reasoning engine
to be able to reason
across things,
but also lots of work
that we did with data,
with better
understanding with things
like our enriched
indexes that
let the reasoning
of the agents
understand much, much more
for larger sums of data.
It was a true unlock.
And I think about
a lot of data.
I think about
700,000 articles.
You heard the
story earlier from
salesforce.com, we have
700,000 articles about
every product, every
detail from years
of information in
Salesforce that's now
plugged in to Agentforce,
answering questions about
40,000 questions
every week
on help.salesforce.com.
And by the way, it's
answering at an 83%
resolution rate.
And of the time that
the Agentforce actually
engages, only
3% of the time
does it actually
escalate to a live human.
It's an incredible
operational story,
and we see this with
all of our Agentforce
customers.
There's only one thing
that I think about here,
which is, can we do more?
How do we think
about bringing
this awesome technology
and moving beyond the chat
window?
How do we move
outside of the box?
How do we think
about allowing agents
with the reasoning
capacity to create new UI
and think about
applying it
to logic,
nonconversational
experiences?
And how do we give
you, the developers,
more control over
what your agents can
do for every
workflow in a way
that your
enterprises demand?
And we've been
really busy at work.
We've got lots
of new features.
And we're today very
excited to announce
the next installment,
which is Agentforce 2dx.
This is specifically
for you, the Developer
Edition.
And this has got
tons of features
here, the ability
to have more control
over your agent so that
enterprise can count
on it, with features
variables that
allow deterministic
guaranteed logic
to be brought
into your agent
without sacrificing the
dynamic nature of what
agents can do.
It allows you to
build your Agentforce
to move beyond
conversation,
to be proactive as
opposed to reactive,
with things like
triggers and bringing it
in for variables and
do background work
and have deeper reasoning
over longer periods
of asynchronous tasks.
And also it allows
you, with features
like Surfaces, to bring
your agents more connected
to the interfaces in which
we're using them as users,
more awareness of
what's happening
in your mobile
application,
on your website to be able
to take action on those,
not just talk, and be able
to create more custom UI
and components to
bring the experience
to the next level.
And this is a
developer conference,
so instead of just
telling you about it,
I want to show you.
And one of the best
ways we can do that
is through our customers.
And I'd like
to introduce--
by the way, Vivint, thank
you, an Agentforce company
right here that's
paving the way.
And we're going to tell
you the story of Vivint.
And then we're going to
do some live upgrades,
getting into dx,
and what we can do.
First, let's watch
a quick video.
[VIDEO PLAYBACK]
[MUSIC PLAYING]
- There's no
place like home.
It's your family.
It's what you protect.
It's all about
peace of mind.
That feeling of I
know I'm secure,
I know the people I
care about the most
are taken care of,
it is everything,
everything that matters.
- Vivint is a leading
smart home provider.
- On average, we have 14
or 15 individual devices
in our customer's home.
- Our platform brings all
those devices and sensors
together in an
intelligent way
and provides that
magical experience
for our customers.
- Vivint is
growing rapidly.
And as we grow,
we need to make
sure every customer gets
that tailored experience
that uniquely
meets their needs.
With Agentforce, we're
giving our customer
service team a
digital labor force,
helping every customer
know we are there for them
24/7.
- The AI revolution,
this era that we live in,
is really the
most exciting time
I've had in my career.
- It became
very clear to me
that GenAI could transform
our customer experience
in a way that we
have not seen.
We've got to find a
way to leverage this.
Our bias was towards, hey,
let's go and build this.
We had to orchestrate all
of that data and segment
it.
We went and built
our own RAG.
We had to build a platform
before we even started
getting any value.
Agentforce did that
heavy lifting right out
of the gate.
We don't have to
move data around.
It's native.
It's in our platform.
- Agentforce allows you
to piece together very
quickly a set of
technology that
has taken people well
over a year to build out
equivalent platforms.
- Agentforce is
quicker, faster, better.
We got business
value immediately.
Agentforce spoke
for itself.
- Agentforce has
several instructions
that we've told
it about, and it's
going to take some
action based on that.
And so what it's
doing now is
it's actually looking at
our knowledge articles.
And it's also pulling
in specific information
about your system
and combining
that with our CRM data.
And it's coming
up with, OK,
what's the next
step that we should
take to troubleshoot it?
- So it looks like it's
asking me what color
the status light is.
And that just lets
us know what's
going on with the system.
And I can see
that it's white.
- If you can
see here, it's
actually going through
this reasoning part
where it's saying, it
looks like the white light
is indicating that
the camera has power
but it's probably
disconnected
from the internet.
Could you go to your
Wi-Fi connection
and try power
cycling the router?
- What's great is
that Agentforce
isn't data dumping and
overwhelming the customer.
It takes care
of the customer
in a simple, friendly way.
- A chatbot just can't
do this type of stuff.
- We have a tremendous
responsibility
to our customers.
Agentforce allows
us to provide
a very personal customer
experience that's
scalable.
- This is a
technology that
is truly revolutionary.
It allows us to already
do the impossible.
- When you look at the
business and engineering
value that we're
getting out
of just rolling
that solution,
it's like a cheat
code in a video game.
It seems too
good to be true.
- The investment
in Agentforce
is one that pays off
now, and it's going
to pay off in the future.
The sky's the limit.
[END PLAYBACK]
All right.
[APPLAUSE]
So we're going to
dive in to Vivint
and we're going to
make this interactive.
And I'd like to
thank and welcome
our Vivint advisors,
Stephan and Irene,
to help us do that.
Thank you guys.
[APPLAUSE]
And we're going
to start off
with Vivint's
website, right here.
And this is any
customer Vivint
can go experience their
Agentforce right now.
The bottom
right-hand corner,
you can start your
chat conversation here.
And you can ask
questions about this.
For example, I have an
issue with maybe a device
disconnecting.
Can you help me
troubleshoot that?
And this is connected and
grounded in their data.
So they're troubleshooting
articles, their knowledge
about their products.
It understands Vivint,
and it understands
a bit about me
as a customer,
but it doesn't
have all the data.
And if we contrast
this a bit,
if you were to
email in or call in
to the contact center
of Vivint, what
does the human rep, what
does the customer service
rep that didn't have?
Let's flip over to their
view of what they see,
can understand
what's possible.
And this is a
beautiful view
of the service console.
You can see in the
left-hand side, Martinez
Household, bringing
in lots of information
about the customer
they're talking to.
We've got the devices here
that are in that household
that they've purchased,
that they're online,
and their status.
And if we scroll
down a bit here,
you're also going to see
eventing data, and all
those cameras, the
thermostats, all
the stuff, these
eventing level
detailed telemetry data
that's coming through Data
Cloud.
So the human service
reps have access
to lots of data.
They've got their
heads-up display
to help the customer.
They've got all kinds
of buttons and actions
they can take to help
service that customer.
And so we asked
ourselves, what
if we had that same data
that the human reps had we
could connect it
to Agentforce?
What if we took
those same actions
and the same policies that
the humans we connected
to Agentforce?
What was possible?
You got to do this
in the right way
with the right control.
And so let's go
ahead and see what's
possible right now.
So we're going
to move over
to Agent Builder and
that Smart Care Agent.
This is their agent
that's customer-facing.
We can move into it.
And you see the topic
at the very bottom,
Telemetry Data.
This is now connected
to that event level,
the device that's
emitting what's
happening in the health
of those sensors.
And you can see there's an
action there that connects
it through Data Cloud.
So we're allowing
our agent
to actually search the
data that's coming in
from all those devices.
Additionally,
it's a lot of data
and we might want
to analyze it.
And AI is pretty good
at analyzing data.
But having Tableau is
also very useful in terms
of being able to
do deeper analysis
and visualizations
and more.
So we can bring in Tableau
directly integrated here.
Now, before we
try this out,
I want to point
one thing out.
This isn't just
general information.
This is my household's
telemetry, the devices
in my household.
This is private
information.
So we want to make
sure we can control
what the agent can do.
And we all know
the best way
to do this is to go
write an instruction
and just tell it with
really nice words.
Pretty please?
Let's prompt it.
Don't send the wrong
data to the wrong user.
Of course, that's
not what we do.
Don't do that.
That's not great
with prompting.
We need a more guaranteed
way to do this.
And so there's
a new feature
called Variables that
allow us to start bringing
in deterministic
logic directly
into the reasoning engine.
So you have structured
information,
for example, the user
ID or the profile that's
happening in the session.
And now you can take the
structured information
and with a feature
called Filters,
you can create
expressions that
say, hey, look, if
they're logged in,
like we've done
right here,
you can create
a filter based
off those
structured variables
and take that
filter and apply it
to any capacity
of the agent
so that telemetry
topic that
let the agent actually
access that customer's
data will only
be available
if they're logged in.
That makes sense.
There's no way the agent
can hallucinate and get
access to the data.
There's no way that that
user could ask really
nicely and get the agent
to do something that you
don't want it to do.
So new levels of
deterministic control
that don't stop the
awesome capability
of the agent to
reason and be dynamic
across many contexts but
do control it the way
that you want.
All right.
So let's go ahead
and ask a question.
Now, as this is
thinking, you'll
see the selected account.
Notice we don't put
an account ID in.
So how does it know that?
Those variables, you
can also inject them
so you can save a profile,
saving you a lot of time,
by the way, as
you're doing
this to start from state.
It's a great
testing harness.
You can see we've got
the response here.
Now, as we asked about the
Martinez household, what
you can see is it
doesn't come back
with just one event.
It's actually looked at
multiple bits of telemetry
here.
And you can see that
it's experienced
multiple disconnects
recently.
And the agents actually,
this nice visualization
from Tableau, decided when
it received that data.
That's the best way that I
should be able to see it.
So it's creating a
visualization right
on the fly for me.
How do they do this?
Let's jump over and look
at the play-by-play,
the thought-by-thought
reasoning, and the actions
that happened.
And we can
start at the top
and you see that
it's actually
injected who
we are, again,
and it looks
up our account.
So it understands that
we're the Martinez
household and
that now allows
us to go into Data
Cloud and search
across all the
device information,
about what is happening
at the Martinez household
and getting
that data back.
Now the agent can
see those events.
You can see that
output there,
that little bit of
JSON could go into it.
And the agent
understands now
that I need to
be able to show
this trend or this kind
of insight to my users.
So it's choosing to have
Tableau come in here
and it's looking
at this and saying,
can you go ahead and
create this visualization,
this line chart?
And what's happening
here is quite amazing.
You're seeing data with
AI and agentic reasoning,
and UI with the
Tableau all connected
from just one query
that's happening
on the fly with no code.
And this is something
that only a deeply unified
platform like Salesforce
can make this easy.
And it's fantastic.
Now, here's a question.
Do you think that
we can do more?
Can we do more with this?
What do you think?
Yeah.
All right.
So one thing that
I think about
is this is really great,
but I had to ask it.
I had to say, hey, what's
wrong with my house?
Is there anything going on
that I should know about?
And it can connect
to my data.
What would have
been really great
is if my house could
have actually talked
to the agent and then
brought me into the loop,
as a consumer, only
if there's an issue.
And to do that,
we need to have
agents to be able to
work in the background.
We need to be
able to have them
be triggered and work.
And the best way to
do that is in Flow.
Do we have any
Flow users here?
[CHEERING]
All right.
So for all those
that just said yes,
you know this is
what Flow looks like.
It's always and nice
and clean and simple--
I'm actually kidding.
Stephan, can you zoom out?
Actually, let's do this.
Yeah, that one.
All right.
This is what Flow
typically looks like.
Let's just talk about
this for a second.
This is logic.
This is conditions.
This is code.
And by the way, what
are these things doing?
These are saying, if it's
this device or this event
go do these things,
what happens?
Vivint is launching new
products all the time.
New devices, new things.
This logic is
going to grow.
There's maintenance here.
But instead of putting all
this logic and reasoning
in this way, we can
start using agents
now to do this.
So instead of having
conversation in
and conversation out
for an agent that's
talking to us, we can
have data in and inputs in
and action out because we
can put agents directly
into Flows now.
And so this is
now the same Flow,
but with agentic reasoning
as a singular step.
And if we click
on that, you
can see right
there the inputs.
It looks kind of
like a prompt.
Any of your
Flow variables,
you've got who's
the customer, what's
the event, what's
happening in the Flow.
You can actually push
this into the agent
and allow the reasoning
to happen inside
of it, which
means you don't
have to code and describe
every single possible
path.
You can say, hey,
here's an event.
Let's go look at
the telemetry data
and see if
there's a pattern.
And you can say, based
off of a situation, here
are my policies
to handle things.
Here are your
potential actions.
Notify the user or
don't notify the user.
All of those
things are possible
at the agentic
reasoning level.
So why don't we go ahead
and let's go simulate.
Let's run this
flow, we'll simulate
an event going through.
And we've set this
up, so it's probably
at the tipping point for
the Martinez household.
And you'll notice
that when we run it,
it isn't something that
just sends a notification
right away, and that's
because the agent is
looking up the
information.
It's going back.
It's doing research
in the telemetry
and it's making
a decision.
And then it decides--
there we go, for a push
notification that's
now bringing agentic
reasoning and experience
directly into Vivint's
beautiful native mobile
application, as
a new surface.
And we can open
up this message.
And the first
thing that you see
is this is not a
generic notification.
Something disconnected.
This is intelligence.
This is action.
We're seeing
a digest here.
Hi, Emma, and then
you could read down
here our device
shows that Wi-Fi
has been going in and
out over the last week.
It actually looking
at telemetry
because it was
instructed to do that.
That's what the
agent was told to do.
Look at the
bigger picture.
Is there something more
systematic going on?
Look at who the
customer is.
Are they great customers?
What devices they have?
Here are your possible
courses of action.
And you can see
that it's not just
summarizing
what's happening
to me as a customer,
but it's also giving me
a suggestion of what to do
next, which is scheduling
an appointment.
It's looked up
the free, busy,
or the available
time slots,
and it's using our new
UI componentry here
with our time picker to
create an effect that I
can just tap a button to
schedule a field service
appointment as a consumer.
And this is an
awesome experience.
This is a contextual
message with action,
with one click, with UI,
all that coming together,
and it is fantastic.
This is a taste
of what's possible
when you bring
agentic reasoning
into the consumer
experience.
But consumers are
not the only people
having fun here.
We also can apply
all of Agentforce
and all of its
agentic reasoning
to help employees
in every workflow.
And if we want to
break that down,
I just scheduled
an appointment
for a field tech
to come to my house
or the Martinez
house to fix it.
And if we say,
how could we
help that field
service, that technician
with agents?
And the answer is,
when you wake up,
what are you doing?
You're thinking about,
what is my agenda?
What are all my
appointments today?
Do I have all the parts
that I need on my truck
because I don't want to
show up to an appointment
without the right things
and have to reschedule?
Nobody likes that.
And also, once I
get busy with my day
when I'm going from
customer to customer,
what am I busy doing?
I'm fixing their problems.
I'm building
relationships.
There's a lot happening
in the middle of the day
once the day starts.
So how can we make this
more useful and helpful
having an Agentforce
along the way?
Well, let's start with
that first problem--
waking up, making
sure the right parts
are on the truck
for the day.
So we can go into-- now
we're into another agent.
This is our field
service agent,
and understand
parts and inventory
and things like that.
We now need to
connect to our ERP.
So we need to connect to
something outside of that.
And the best
way to do this
is with the MuleSoft
API catalog.
So right here I can
have my agent connected
to be able to order
parts, understand
inventory, and more.
And maybe we want the
technician to also
be able to do
analytic, queries,
and visualizations
with that Tableau.
So let's go ahead and add
in some Tableau actions.
And then instead
of notifying them
in that push
notification, maybe
we want to notify them
proactively in Slack
instead where our
field techs work.
So let's go
ahead and we'll
be able to send
a message here.
And we'll save this.
And let's actually
kind of day of--
actually, if I could ask--
we want to do a
little roleplay.
I'll be the technician.
And Stephan, do you want
to be my next appointment
that I'm coming to?
Absolutely.
So I'm a very happy
Vivint customer.
I know I've got a service
appointment today.
And I've got a
notification right
on my Vivint app.
Hello, we're due to
arrive at your home
within the next one hour.
Is there a gate code
that our technician
needs to get to the home?
Yeah, it's 5321.
Got it.
Would you like me
to send you a link
to track your technician?
Yes, please.
No problem.
I will send you a link
to your technician's
estimated time of arrival.
All right.
[APPLAUSE]
What we're seeing
here, a brand-new
interaction
that's deep within
the mobile application.
So this is part of
what's part of Surfaces.
And also you're
seeing voice,
which is really
incredible,
and the ability to have
custom UIs with Maps.
That's fantastic.
But you know what's
happening right now?
That's the
consumer experience
for my next appointment.
What am I doing
as the tech?
I'm helping a
different customer.
I'm actually busy
with another customer.
And what is happening?
I have a teammate.
Agentforce is literally
one step ahead of me
with my next
appointment, making sure
that they're home,
that they're there,
that I have access to it.
It's working with me as
a team throughout my day.
And now when I'm
leaving my appointment,
I'm ready to go to that
Martinez household, what
am I going to do?
I'm going to open Slack.
With our work
item channels,
I can see my next
appointment as a channel.
All the context is there,
and my field service agent
has beat me to it.
It's already
got my pre-brief
for the appointment.
So I can see that
I'm due to meet
Emma Martinez, a little
bit about the customer,
that the issue
is her Smart Hub.
It keeps displaying
the red light.
It looks like
it needs to--
has an issue.
And then a little bit
about the customer.
She's been a customer
for five years.
It's giving me
kind of a digest
of what I need to know
to go help that customer.
And then ultimately,
it's also-- oh, look,
it's choosing to
use also a Tableau
to show me the
telemetry data.
So it's able to
say, in this case,
I've got some information.
This might be
useful to send you
the trends of what I've
seen from disconnects.
And then additionally,
there's some actions.
And you can read here
that it's based off
of what it's
seeing the issue is
with these disconnects.
It's made sure that I have
a Smart Hub on my truck
because I might
need to replace it.
It might be a
networking issue.
It might be the Smart Hub.
So it's like you have
the parts on your truck
that we likely think
are going to be needed
for this service event.
And also, of course,
the gate code, as you
just saw.
Now, this is interactive.
I could come back
and I could say, hey,
what's a Smart Hub?
I'm new here.
I'm a new technician.
How do I install it?
I have access to
all that information
as a new employee
to help me onboard.
Or maybe I'm going
to say, well, this
is an interesting
trend, but it's only
for a few days
or something.
Can you actually give me
three months or what's
been happening
or the devices,
go back to the telemetry
data and much more--
or take actions for me?
I'm running late.
Can you let the
customer know?
So it is literally
a partner.
Agentforce is your partner
as you're doing work.
And this isn't
just a team where
there's one agent and
working with a human.
And Slack, it's a
multiplayer game.
So when we
check-in, we also
can have things like a
sales agent coming in
with access to
the same data--
the same telemetry, the
same customer information,
the same understanding
of what parts are
on my truck,
what campaigns
are we running right now.
And it can notice,
for example,
that the Martinez
household
has a first-generation
camera.
And I happen to
have the latest
fifth-generation
smart camera here
that I have on my truck.
I'm not going to need it
for appointment today,
and it's authorizing
the ability for me
to do a free install
and give them
a promotion while
I'm out there.
This is the power of
Agentforce coming in,
with more control, better
experiences, background
working across
all the surfaces.
It's really any team
and any workflow.
And we can't wait to see
what you build with it.
With that, back
to you, Patrick.
[APPLAUSE]
All right.
I got to say, nobody
else is doing that.
Nobody else is showing
demos like this.
I'm not seeing it.
And I think it is this
deeply unified platform,
our data, our apps,
and our agents,
all deeply
unified that are
allowing us to do this.
I mean, look at all of
this incredible innovation
that we just saw.
This is why it's
not just Vivint,
but 4,000 other customers
in just a few months that
have signed on and are
now building these agentic
solutions.
It's just really awesome.
But there is a problem,
and that problem
is actually Adam.
You see, the
problem with Adam
is he's very confident.
And everything that
you just saw Adam
built right in production.
And that's kind of stupid.
So--
[LAUGHTER]
--we need someone to
help sort Adam out.
And for that, I'm
excited to welcome up
Alice Steinglass,
who's going
to walk us through how to
do this professionally.
Alice, over to you.
Thank you, Patrick.
[APPLAUSE]
I'm with you, Patrick.
It's incredible, but we
all know it isn't perfect.
I was hanging out with my
17-year-old this weekend.
And we were playing with
the big online models.
We're using all
these cool new tools.
We downloaded LLaMA
and DeepSeek Lite.
We fiddled with
it in Python.
And it turns out, AI
can make a pretty good
new wave song.
I thought it sounded
like The Cure.
My kid had never heard
of The Cure or new wave.
So I said, hey, I'm going
to do something for you.
And I asked a
couple of the models
if they could make a Gen
Z joke that would resonate
in his Discord server.
[LAUGHTER]
Nope.
I'll roughly quote, Mom,
"I wouldn't post a single
one of these."
And that's a challenge.
That's a challenge for
everyone in this room.
Because even if you
don't need a Gen Z
joke in your office,
there are some things
where agents are going
to accelerate us,
and it's going
to be amazing.
And there are tasks where
honestly, the agents
should work
with humans who
actually can tell jokes.
And to figure out
which is which,
you're going
to need tools.
So today we are
announcing the Lifecycle
for Agentforce 2dx.
That's a mouthful.
Basically, we're taking
the entire software
development
lifecycle and we're
going to add
capabilities for you
to build, test,
and deploy agents
with the Agentforce
lifecycle
because it's different.
How do I build with
natural language?
How do I test things that
are nondeterministic?
How do I deploy these,
monitor, and debug them?
We're going to need tools
across the lifecycle.
You're going to want
to see these hands-on.
So we've got Stephan
once more here
at the demo desk.
And everything you're
going to see right now
is live code.
Some of it's in dev
preview, some of it's
in pilot.
We want to test this
before we release it.
But everything
you see here
is going to be
released this spring.
So let's get started.
Should we fire
up a scratch org?
That's a great
place to experiment.
So we're going to start
with the beginning
of the development cycle--
planning and ideation.
And we're going to
start with a hypothesis.
Vivint here has a customer
referral program today.
And they're
wondering, could I
use an agent to make
that customer referral
program better?
Would it help drive more
adoption of the customer
referral program?
I don't know.
Ryan doesn't know.
So we're going to
go test it out here.
And when I start building
with brand-new technology,
with new
capabilities, it's
hard to start
from scratch.
So we've got a
bunch of tools
to help me get
started fast.
Here we have a
bunch of templates.
I can start with
an SDR agent.
I can create a campaign
optimizer agent.
But there's no customer
referral agent here,
so I want to show
you something new.
I want to show
you how we're
going to use AI assistants
to help you build custom
agents fast.
When I create an agent
with generative AI,
it changes how I build.
Instead of specifying
every detail,
I can focus on the what.
What are my business
requirements?
And the AI is
going to help
me figure out the how.
How do I do it?
Here what it's doing is
it's taking those business
requirements and
it's figuring out
what are the topics that
a customer referral agent
should be able to do.
And it's going to
pull those in here.
And, Stephan, can you dig
in on one of those topics?
Great.
Inside each of
these topics,
we have suggested actions
that it's going to take.
And this is one of
my favorite parts.
Do you see that calculate
referral balance action?
That's a flow.
You have a lot of flows in
your organization today.
Some of you have
built dozens of flows,
hundreds of flows.
I know some of
you have built
thousands of
flows and APIs
and Apex classes today.
And when the AI
assistant was helping
me get started
with this agent,
it wasn't starting
from scratch.
It was taking that
business requirement,
and it was searching
through my organization
for actions that
I had registered
as agent actions so that
it could find that flow
and automatically include
it as part of this agent
definition.
So let's click Next again.
And now I can just go
ahead and edit or keep
the suggested description
and instructions for what
I want my agent to do.
And then I've got to
pick the channels.
Where's this agent
going to work?
Stephan, let's have
this one work on
Slack, web, and email.
Now we're going
to add data.
And once again, we're not
starting from scratch.
We have a lot of data
in our organizations.
Vivint has pulled in all
of this IoT and sensor
data.
They're streaming
in all sorts of data
into Data Cloud.
Maybe you've connected to
your Snowflake instance
or Databricks and
you're pulling it
in with zero-copy.
That data is
here, and we're
going to make it easier.
So if you have an
unstructured document
that you want to pull
in as part of creating
your agent, like here I've
got a referral program
description-- basically,
just a PDF that describes
what this program
is, I can just
include it right here.
I can upload it as part
of creating my agent.
This actually takes
a minute to index.
So Stephan and I uploaded
this one this morning.
He's just going to
refer to it right here.
And then we're going
to hit Create Agent.
And that's it.
We already have the first
step of building done.
We already have an agent
that we can get hands-on
with, that we can
try, that we can test,
that we can play with.
So Stephan is going to
ask it a few questions.
And while he does that, I
want you to pay attention
to that middle column
right there because this
is how we understand
what our agent is doing.
And that's going
to change, too.
I don't need to look
through log files
and error codes to
understand why the agent
or why the computer
in the past
was doing what
it was doing.
The agent's going to
explain its reasoning
in natural language.
It's going to tell me
why it picked that topic,
and then it's going
to go make a plan.
And it's going to
have some actions,
those same actions that
you saw before that it's
going to execute.
And when it finishes
executing those actions,
how is it going to
display the result
to the customer?
Again, I don't
have to hard-code
exactly what
the padding is
and what it
should look like.
It's going to be able
to use reasoning to say,
this answer probably best
displayed as a movie.
So let's just put the
movie right there.
And you can turn
that on today
with adaptive responses.
These adaptive responses,
not just pick a movie,
but some of the things
that Adam was showing you
earlier today, it can pick
a form or a catalog, cards
that it can
rotate through.
And it does it
across surfaces.
So those will show
up in WhatsApp,
but they'll also
show up on the web.
And if you want to
be more precise,
I know some of you
do, in dev preview,
you can build your
own Lightning Web
component that displays
the answer exactly how you
want it displayed.
So you're not just saying
what you want your action
to do, you're saying
how you want it to look.
So you can make
that pizza tracker
that Adam showed
you, and all
of this, that
Lightning Web
component, those flows,
those topics, the actions,
the instructions,
all of it
is metadata in
your org, which
means we can do
all the things we
do with metadata.
We can version it.
We can play with it.
We can edit it.
And we're doubling
down on our tools
for metadata in scratch
orgs and sandbox orgs.
I'd like to introduce you
to the new dx inspector.
Here, I can see all of
the metadata for my agent.
And I can jump in
to the builders
and the specific metadata
for my flow, for my topic,
for my prompt
right from here.
And some of you use
source-driven development
today.
And you've got
a good handle
on exactly what's going on
with all of the metadata
and all of the changes.
And some of you are
working in your org
and it can be hard to
keep track of what are
all the changes we made.
So here, we also
have a change tracker
so we can see exactly.
I got some applause here.
Yeah, I'm excited
about that one.
Thank you.
[APPLAUSE]
So we can see exactly what
Stephan and I have been
changing for this agent.
And we can also
see the changes
that AI assistants
is helping us make.
So we saw a little
bit about how
we can build as
low-code developers,
but I said that this was
going to work for low-code
and pro-code.
So, Stephan,
today, we are also
announcing the
Agentforce module
for the Salesforce CLI.
That's a lot of words.
Can you show us how
pro-code developers
can build agents?
Absolutely.
And of course,
we're going to be
doing this in VS
Code or Code Builder.
So we now have an agent
as part of our extensions
in the Salesforce CLI.
And it starts
out as simply
as typing sf agent create.
And then in the same way,
we just created an agent
using AI, or
from a template,
we have access to
create agents right here
from the command line.
Now, we've already
created ours,
so we actually want to
pull down that metadata.
So we're going to go
ahead and retrieve
that agent down from the
org that we just created.
But you've got
to keep in mind,
we're not just pulling
down the agent.
We're pulling down
the references
for the actions.
We're pulling down the
topics, the instructions,
everything right
here into our CLI.
And so what you'll
see here is we
have all of that
great information--
the plugins, the functions
right here for us.
Now there's one more
thing I want to show you,
and it's that we can
actually attach and build
actions here.
We have an Apex
class that's
already in this project.
It's not yet in
our scratch org.
And we can go ahead
and attach that here
and run that deploy
right at the same time.
And so we're going to
go ahead and create
that action.
It's going to map the
inputs and outputs
from the Apex
class, and it's
going to save that right
here in our org for us.
Great.
So we built an
agent with low code,
we built an agent with
pro code, let's move
on to the next step of
the software development
lifecycle.
We need to test
this agent.
And it's going
to be different
because we need
to test something
that's nondeterministic.
So, Stephan,
can you show us
how we would do
this with pro-code?
Absolutely.
So it's going to start
out with a test suite
that we've
pregenerated here
for the demo today that
just takes in some inputs,
takes in some outputs.
And we're actually
going to attach that
to our test runner and
generate that test.
So I'm going to go ahead
and get it started here,
and we're going
to run that test.
And it's going to be
a little different
than we're used to seeing.
Yeah, it is.
So while it's
running, I want
to talk about what you're
looking at here, because
do you see that test
case right there?
It doesn't look like a
standard Apex unit test.
We're not taking one
input and mapping it
to one output.
I'm not checking to
see if a variable is
true or larger
than some value.
The input is
human language
and human
language is wild.
People can say anything.
So we have an
utterance here that's
just a natural language.
And the output, again,
it's not a variable.
I'm specifying an outcome.
I'm talking about what do
I want the result to be.
And we can't just
test it with one.
We're going to have to
test this with thousands
of different
utterances, and we're
going to have to do
stochastic testing
and have probabilistic
thresholds for what
success looks like.
And I want all of
these new testing tools
to work as part of our
CLI, part of the test
tools, the
DevOps processes
that I use today.
And so here,
they're coming back
with just passes or fails.
If it was fail, we'd
see a fail here.
But it looks like--
Stephan, we passed?
Yep.
Excellent.
It's not just
pro-code developers
who need these tools,
low-code developers
need exactly
the same tools.
So this is testing center.
And here, because our
pro-code and low-code
environments
are connected,
I can see those same
tests that Stephan
was running in pro-code
and I can run them here.
But low-code developers
in the room you
want to be able to create
your own tests, too.
So here I can
upload a CSV file.
Maybe I want to scrape
a bunch of comments
that we've gotten from
our bots on the website
and use those as
a starting point,
or I can write
my own tests.
Or, again, we have
AI assistants.
So, Stephan, let's get
the AI to write some tests
for us.
It's the AI, so we
don't have to write 20.
We're going to
do 2,000 tests.
I'm going to get it to
write 2,000 tests for us.
And let's generate
those test cases
so that we can use
them here in low-code.
We've built our agent,
we've tested our agent,
now I want to move
on to the next step
in the Agentforce
lifecycle.
Let's deploy this agent.
And to do that, we're
going to head over
to DevOps Center.
And here our low-code
and pro-code developers
can work together
connected
to our processes, our
release processes,
our source code.
We can use it to
promote things
through the
pipeline and make
sure we have
the right checks
and balances in place.
So, Stephan, why don't you
go ahead and promote this
to the UAT environment?
And we just got
to the point
where we could get it
into a UAT environment,
we could do this
deployment step
in 13 minutes
and 32 seconds.
That was fast.
That was fast because
we had AI assistance,
but it was also fast
because we weren't
starting from scratch.
We were using all the
things that were already
in the environment.
We were taking advantage
of the data and the flows
that we'd already
built so that we could
get going quickly because
we need to bring it out
to more users.
It's great that Stephan
and I can play with this.
But, Ryan, I want
you to play with it.
I want to get it out
to our business users.
I want more people to
get hands-on with it,
and then we want to
deploy it to more people.
Maybe everyone in this
room should check it out.
And then after
everyone in this room
has given us some
feedback and we
can see what's
going on, we
should deploy it out to
everybody online watching
us.
And as we do this,
as we roll it out
to more and more
users, we need
to know what's going on.
Is it working?
What's working well?
What's not?
So let's move on
to the next phase
of our development
process.
We're going to
go into observe.
And here, we
have new tools
so that you
have visibility
into what's going
on with your agents.
We're going to look
at some agents that
are already in
production here.
So we've got some data
that we're collecting.
And you can see the
number of users.
We can see the
number of sessions.
We can see the latency.
We can see the
quality scores.
And once again, there's
something new going on.
We've got millions
of people saying
millions of things.
How do we even make
sense of this data?
So we're using AI
to automatically
tag and categorize the
conversation topics that
are coming in.
So I can dig into a
conversation topic,
like in this case,
connectivity.
And then within the
conversation topic,
I can see what worked
well and what didn't.
And when it
doesn't, you need
to be able to dig
in even deeper.
And so this is the
conversation explorer.
And here I can see
exactly what happened
during this conversation.
I can see exactly
what was said turn
by turn to the agent.
I can look at the
quality score.
I can look at the latency.
And here-- oh,
yeah, this is great.
He actually clicked
on an action here.
And when you click
on an action,
we can see the trace
of that action--
what actually happened
in the system,
where was it
going, where was it
spending time
turn by turn.
These are the
tools that we
need to be able
to debug to get
real visibility into
what's happening
with our agents.
I can see almost
everything here,
but there's one
thing I can't see.
There's one thing nobody
in this room can see.
That's right.
The email address
of the customer.
Because I shouldn't
be able to see it,
because I'm a
developer here.
Stephan and I should
not have access to that.
So let's talk for a moment
about data governance.
You have seen a lot of
data across this demo.
You saw sensor data and
home data and customer
data.
You need control
of that data,
and you've got
control today.
You've got your
data set up.
You've identified
the sensitive data.
You've identified the PII.
You've set up your
sharing rules.
Maybe you're
running Shield.
Maybe you're using Data
Mask to mask that data
in preproduction
environments,
but we just added
a lot more data.
We added a lot of
unstructured data,
we added all of this
conversational data,
so we need to
add new tools
to this step of the
development process
as well.
And that's why
today we are
announcing new
data governance
tools for Data Cloud.
Here, we can use AI to set
up rules and automatically
tag and detect
sensitive information
across structured and
unstructured data.
Once we've
tagged the data,
we have fine-grained
access controls
that allow me to control
exactly where, when,
and how this data is used.
So I can set up policies
like masking data in--
I don't know, where you
get some policies here?
We're going to redact
smart lock pin codes
or obfuscate device
serial numbers.
And as I set up these
policies, I'm in control.
I can decide how do I
want to mask this data.
I can decide
when and where.
So maybe we
shouldn't have access
when we're building.
Maybe we shouldn't have
access when we're testing.
Maybe when it
reaches production,
only some people
should have access.
My HR department
maybe has access,
my finance
department maybe
doesn't, I don't know.
It's up to you.
So when we go back and
look at that chat log,
that data masking
that you saw there,
you're in control of that.
So we just saw a
quick run-through.
You're going to see a lot
more over the next couple
of days of how every
part of the development
lifecycle--
building, testing,
deploying, observing,
all of it is changing.
And we are giving
you the tools
to enable you to do
agentic development so you
can build your
own agents and you
can build your own agentic
apps on the platform.
And with that, I'm going
to turn it back over
to Patrick.
[APPLAUSE]
All right.
Thank you, Alice.
Again, I just cannot
believe how fast this is
all moving and how
many tools we have
at our disposal.
Now, when we first started
talking about this,
there's so many customers
were like, I don't know
if I can get started yet.
I don't know if
I can trust it.
I don't know how to
really control it.
I mean, here it is.
If you're in a
regulated industry,
if you're in a bank
or health care,
these are the
tools that you
need to bring back
to your leadership
and show them that you
can do this really, really
safely across any
type of workflow.
So listen, as we see
now, we can not just
build agents but
we can explore
all of the metadata, we
can take that metadata
and promote it.
Well, if we can promote
it through orgs,
what if we
could promote it
into some sort
of marketplace?
We'd all be able to share
all of our work together.
And so for that, I'm
excited to welcome,
as soon as he
fixes his mic,
there he is,
Christophe, to walk us
through how this
all comes together
in our new AgentExchange.
[APPLAUSE]
Thank you so
much, Patrick.
It's really an
amazing time
to be developers,
admins, and architects.
I don't know
about you, but I
think the pace
of innovation
is just incredible.
And to be at the forefront
of that innovation,
I think that all of us
in the room and everyone
online, we have an
unfair advantage,
and that is a strong
partner ecosystem
that is built on an open
and extensible platform.
For example, you can
integrate any data
with our zero-copy
data partners.
You can choose the
right model for the job
with partners like OpenAI,
Anthropic, and Google.
And of course, you can
extend the platform
with pre-built
solutions from ISVs
because as a
developer, I don't want
to reinvent the wheel.
I really want to focus
on high-value features.
And together with
these incredible ISVs
and the community,
all of you,
we built the leading
enterprise app ecosystem
in the world.
Look at these numbers.
Over 7,000 partners,
9,000 listings.
And did you know that 91%
of Salesforce customers
actually use
an application
from the AppExchange?
That is incredible.
And every single
time there
was a shift in
the industry,
these ISVs built
new solutions
so that you can
innovate faster.
And now, as we are
facing maybe the biggest
shift in the
industry ever,
we are doing it
again for AI agents.
That's why today we are
announcing AgentExchange,
the trusted Agentforce
marketplace built
into Salesforce.
With AgentExchange,
you can deploy
Agentforce faster
using hundreds
of pre-built actions,
topics, and templates.
You can extend
your own agents
with partner
actions that you
can discover in
Agent Builder
as you are building
your agents.
And as a developer, you
can package your agents
using the agent for CLI
so that you can then
sell it or share it
on that marketplace.
That is AgentExchange.
It's incredible,
it's available today,
and I want to show
it to you in a demo.
So let's see the demo.
All right.
So here we are in
the Vivint service
application.
And their agent just
escalated that request
from a customer.
They want to upgrade
their cameras.
And look, next
best action already
identified an offer that
just perfect for them.
So the last thing to do
is to create an order
and then send an updated
version of their contract
to the customer so
that they can sign it.
So let's ask
Agentforce to do that.
So we're going to write a
prompt, create an order,
and send contract
to customer to sign.
So the agent is
reasoning over
the available actions.
OK.
So it was able to
create the order,
but it looks
like it was not
able to send the contract.
Now, why is that?
Well, that is simply
because that agent
doesn't know how to handle
an eSignature workflow.
So what are we
going to do?
Well, option 1, we
could reinvent the wheel
and build a brand-new
eSignature action
from scratch
for that agent.
Option 2, we could
use a pre-built action
from a trusted agent.
So what do you think?
I need help.
Option 2.
Option 2.
OK, I think we should
go with option 2.
Let's go with option 2.
And where are we going
to find that trusted
solution?
AgentExchange.
AgentExchange.
Of course.
So let's check it out.
And there it is--
AgentExchange for
the first time
ever on stage at an event.
Look at everything
that you have.
[APPLAUSE]
It's awesome.
You have hundreds of
pre-built actions, topics,
and templates.
It's just incredible.
And they're all
ready to go.
I could install
them from here.
But guess what?
It gets even better
because coming soon,
you will be able to access
AgentExchange directly
in Agent Builder as you
are building your agent.
So, Stephan, let's
go to Agent Builder
and check it out.
So here I want to add
an action to my agent.
So we'll click
New, and look,
there is a new
AgentExchange option here.
So let's click that,
and boom, there you go.
This is AgentExchange
in the flow of work,
in the middle of
Agent Builder,
really incredible.
Now, remember, we want
to find an action that
can generate a contract
and send it around
to sign.
So let's search for that.
I will search and--
great.
DocuSign came back as
a recommended solution.
So let's check it out.
And here I can learn
more about that solution.
I can even look at the
topics and the actions
that are available.
This looks perfect
for what we need.
So let's go ahead
and install it.
And easy-peasy.
It's just going to take
a few seconds here.
And I will be able to
see the list of topics
that are available and
even the list of actions.
So let's select
all the actions
and then add them
to the topic.
And now we can try again.
So we'll ask
the agent again
to generate a contract
for that order
and send it to the
customer to sign.
And again, the
agent is reasoning
over the
available actions.
Remember, there are
a few more actions
because now we
install DocuSign
and now it was
able to do it.
How did it do it?
Well, let's take a quick
look at the tracer.
Here, you can see
that first, it
selected the
Create Contract
action from DocuSign.
Great.
And then the Send for
eSignature action.
That makes sense.
And that is really the
power of AgentExchange--
extending the capabilities
of your agents
with powerful
actions from partners
in the flow of work.
That's incredible.
But there is
one more thing
that I want to show you.
What if you wanted to
create your own agent
and then sell it
on AgentExchange?
So let's go to VS
Code and take a look.
So here I'm
building an agent.
And on the left,
you see a ton
of metadata for that
agent, the metadata
for the agent itself.
For the actions flow
Apex, there is a ton here.
And now, today, I'm
excited to share
that using the
agent for CLI,
you can package all
the metadata related
to an agent with just
a few simple steps.
So let's do it.
Let's open the
command line.
And here I'm going to
use a first command
to generate a template
for the agent.
So that should be quick.
There you go.
And now I'm going to
use a second command
to actually generate the
second-generation package.
Remember there is a
lot to package here.
And so it's doing it.
It's going to package
everything, and boom,
package created.
That is all you need to
do to package and share
your agents inside
your company
or on AgentExchange.
We can't wait to see
what you will build.
Back to you, Patrick.
[APPLAUSE]
OK.
All right.
We're just about done.
We're almost there.
Stick with us for
a few more minutes
because we have our most
important part coming up.
Now you may remember,
about 50 minutes ago,
I said that we want a
million Agentblazers
before the world ends.
I have great news.
[LAUGHTER]
The world has not ended.
And I have even
greater news.
My favorite
three-word phrase
in the English
language here
to talk us through our
own new community, Leah
McGowen-Hare.
[APPLAUSE]
Agentblazers.
Shout-out to all the
Agentblazers in the house.
My gosh, Josh and I were
watching these demos,
the innovation
is innovating.
I mean, you all are seeing
some amazing things.
We're talking
about agentic apps.
We're talking about
managing digital labor,
like, what?
Technology
continues to evolve.
The way we work is
continuing to evolve.
But the question
is, are you?
Are you evolving?
Are your skills evolving?
Now, I don't say that
to throw shade or make
anybody feel shame.
These are questions I have
to ask of myself regularly
throughout my career when
I started as a mainframe
developer and
had to evolve
into an on-prem
developer and evolving
into Cloud computing
multi-tenant developer.
And now we're talking
about natural language
processing development,
like, what?
This is amazing.
And I am so excited
at the opportunity
to learn new skills.
And here's the thing.
I know you all are as
well because you email me,
you DM me.
You tell me when you've
learned a new trail
or you've gotten
a new cert.
And I love it because it
inspires me to continue
to learn, too.
And as we evolve,
we have to learn.
Because here's the thing--
you all know that
the landscape of work
is changing.
And the reality
is tomorrow's jobs
belong to
today's learners.
Let me say it again
for the folks online.
Can you all see me?
OK.
[LAUGHTER]
Tomorrow's jobs belong
to today's learners.
So it's time to get
our learning on.
And here's the thing--
we want to help
everybody come along.
We don't want
anybody behind.
We're going to bring
you all along with us.
And we're doing
this intentionally.
And we've made it easier
for you to jump-start
your AI learning
journey today
with Agentblazer
statuses on Trailhead,
the first of its kind,
I'll have you know.
Yes, you can give
it up for that.
[APPLAUSE]
And just like AI these
statuses have levels.
So let's look at
the first level.
The first level
is Champion.
This is where you're going
to build your fundamentals
and you're building
your building blocks
all around data governance
with Data Cloud.
And you're going to
build your first agent
and you're going to get
all of your building
blocks.
Now, once you've done
that, you then unlock--
that's right, we don't
just give it to you,
you got to earn it, you
unlock the next level,
which is Innovator.
Now an innovator,
you're going
to take all of those--
cheese, I see you.
You're going to take all
of those building blocks
that you've
learned in Champion
and bring them
into Innovator
and actually build.
You're going to start
looking at use cases.
And you're going to start
setting up and customizing
Agentforce or
service and sales.
It's amazing.
But now, are you
ready for this?
I don't know.
The way you were answering
some of Christophe's
questions, I don't know.
I don't know if
you're ready here.
Because then once
you unlock Innovator,
you then go into Legend.
You got to say
it like that.
Everybody say it with me.
Legend.
This is heavy.
This is serious.
You really got to put your
elbow grease into this one
to earn this.
And, now,
listen, all of us
are rich enough
to pay attention,
so pay attention.
Here's the thing.
You must take the
Agentforce Specialist
certification
to earn level 3.
That's part of it.
So this is some serious
level, leveling up.
And I really want to make
sure you all understand
this is so important that
all this technology is
going to require us
to continue to learn.
And we're giving you a
chance to not just learn
but earn.
Now, learning is
great, but you
know what's even better?
Learning together.
And we are creating
so many opportunities
for folks to
come together,
whether it's one of
our Agentforce world
tours or our awesome
community conferences,
you can get together and
get hands-on and really
learn together.
We are doubling down
our investment in this.
But what I'm really
excited about,
I don't know why
I'm whispering
but I'm really excited--
[LAUGHTER]
--is that we are
taking TDX global.
Yes, global.
[APPLAUSE]
We're going to Tokyo,
Bangalore, London,
we're going
around the world.
We're going to bring
the learning to you.
You are going to get
your learning whether you
like it or not.
We're bringing it to you.
Now so many ways to
learn is building, too.
And we've had some amazing
builders here at TDX take
place at a hackathon.
And I want to hear more
about this hackathon.
So, Chris, if you don't
mind coming back up
and sharing with the
folks about the hackathon,
that'd be great.
All right.
Well, the hackathon
was incredible.
We had hundreds of
participants, developers,
admins, architects.
The energy was
just incredible.
And I know you want
to get to the winners,
but before we
do, before we do,
I want to make a
quick announcement.
All right.
Go ahead.
Because I know that all of
you, online and globally,
you are not able
to participate,
so that's why today we
are announcing a Virtual
Agentforce Hackathon.
You can sign up today.
Here is a short URL.
You can all take that
URL and sign up today.
All right.
And now let's get
to the winners.
And I need some friends.
I need some friends.
And maybe a
little drumroll.
Yeah?
A little drumroll.
All right, so
the winners--
awesome, the winners of
the TDX '25 Agentforce
Hackathon is
Team Agent Halo.
Please come on up.
[MUSIC PLAYING]
[APPLAUSE]
We're going to
take a picture.
Oh, move on.
Move it.
Sorry.
Congratulations.
Congratulations, guys.
All right.
Well, Joseph, just a
few quick questions.
Tell us a little
bit about your team,
where you come from, and
how you came together
as a team.
We actually crisscrossed
the country.
We came together through
the Trailblazer community.
All of us overlap in the--
[APPLAUSE]
We're all community group
leaders or MVPs, Golden
Hoodie winners and been
involved in the ecosystem
that way.
Awesome.
And what people
really want
to know, what
did you build?
Our concept was
an agent network.
Eventually, everyone's
individual agents
at all of these
organizations
are going to need to
talk to one another
across the whole
value chain.
So we took ideas that
applied to not just
things like on-road
assistance, with IoT
devices in your car
looking to contact EMS,
your insurance company,
your tow truck, a body
shop all at one time with
one point of reference,
but even things that are
fun, like a book podcast,
and you connect
with the author
and you buy tickets to
their local performance.
Awesome.
So there's all kinds
of applications
across different
industries.
That sounds
really amazing.
And one quick
last question.
What were you the
most surprised
by as you were building
with Agentforce?
I think the most
surprising thing
we found about building
with Agentforce
is just how fast
things change.
I mean, honestly,
every time we
build something new,
there's something new
to discover and
something new to find
about how we can
make it do something
exciting for us.
Awesome.
Well, congratulations
again.
And, Patrick, why
don't you take us home?
All right.
Congratulations, guys.
Enjoy the $100,000,
really appreciate it.
OK.
We are all done here
today in the keynote room.
Download the app,
build your agenda.
We have a ton of stuff.
Go to True to the Core.
Join us at the
celebration tonight.
And one last thing,
first of all, thank you.
Second of all, please
fill out the survey
and tell us how we did.
We would appreciate it.
All right.
Good morning, everybody.
[APPLAUSE]
[MUSIC PLAYING]
Resume AutoScroll
Link copied to clipboard
TDX Main Keynote
Become an Agentblazer with Salesforce and discover how to bring humans and agents together to drive customer success.
Starring
- Leah McGowen-Hare, MSEdSVP, Trailblazer Community and WFD, Salesforce
- Patrick StokesEVP, Product & Industries Marketing, Salesforce
- Christophe CoenraetsSVP, Technical Audience Relations, Salesforce
- Alice SteinglassEVP & GM, Platform, Salesforce
- Adam EvansEVP & GM, Salesforce AI Platform, Salesforce
Up Next
Questions about Salesforce+? We have answers.
What is Salesforce+?
Salesforce+ is the free streaming service for business. It’s a place to watch live events that help you grow your career and company, on your own time. Salesforce+ is produced by Salesforce, so you’ll find our company’s big events like Dreamforce here — as well as virtual learning made for your role and industry. Everything is free and unlimited.
What can I watch on Salesforce+?
You can watch all kinds of helpful, entertaining, and inspiring videos on Salesforce+. One thing our members love is our events. Whether you catch them live or on demand, Salesforce+ gives you a free front-row seat to our biggest global events, like Dreamforce, TDX, Tableau Conference, and Connections. These entertaining business shows star people in roles and industries just like yours, giving you fresh ideas for how to grow your company and do better work. The best part is, it’s all free.
Do I have to pay for Salesforce+?
Salesforce+ is free. Watch any Salesforce+ events or series at no cost. To enjoy unlimited free access to our entire library of videos, become a member by signing up for Salesforce+ with a Trailblazer account. To join us for virtual events like Dreamforce or Connections, you need to create your own Trailblazer account (it’s free!).
What is a Trailblazer account?
Your Trailblazer account gives you access to a unified profile across the Salesforce ecosystem and allows you to manage your settings and access across multiple Salesforce sites.
How do I sign up to register for live and on-demand events, like Dreamforce?
Sign up or log in to your Trailblazer account and complete a few additional details to watch Salesforce+ events live and on demand. We recommend preregistering for all events on Salesforce+ to get your own personalized journey, reminders, recommendations, and more.
Do I need to be a Salesforce customer to watch Salesforce+?
You don’t need to be a Salesforce customer to watch Salesforce+. Our virtual learning is open to everyone.
