Today on the Newcomer Podcast, we’re at MongoDB.Local for a series of conversations on how enterprise AI is actually being built.
MongoDB CEO CJ Desai joins the show 65 days into the role to explain why San Francisco is “back,” how MongoDB is repositioning itself for the AI era, and why unstructured data has made the company’s platform a natural foundation for AI-native applications. He shares his view on the AI hype cycle, the rapid rise of companies like OpenAI and Anthropic, and why MongoDB is staying model-agnostic as AI product cycles accelerate.
We also sit down with Rippling’s Head of AI Ankur Bhatt to discuss how AI is being deployed inside a live enterprise system. The conversation covers building agents across payroll, IT, and finance, why agent identity and accountability matter, and how Rippling is approaching permissions, access control, and AI-driven productivity at scale.
A grounded look at the enterprise AI stack, from the data layer to real-world deployment.
MongoDB #Rippling #AIAgents #VentureCapital
00:00:00
I'm fresh off the Mongo DB dot local event.
00:00:04
They had a ton of developers, partners, startup founders
00:00:07
hanging around and showing how Mongo is trying to respond to
00:00:10
everything that's happening in artificial intelligence.
00:00:13
They are the sponsor of this podcast and they had me on site
00:00:16
at Mongo DB dot local to do a bunch of these interviews, which
00:00:19
I really enjoyed, so we decided to put them in our feed.
00:00:22
I had a great conversation with CJ Desai, the CEO of Mongo, who
00:00:27
dropped by my little studio at the conference.
00:00:30
We have another interview with the head of AI from Ripling
00:00:32
coming after that. This is the Newcomer podcast.
00:00:43
Hi, I'm Eric Newcomer, author of Newcomer.
00:00:46
We're here at Mongo DB's local event, the amazing drop in when
00:00:52
you get the CEO, what, 65 days into the job to show up on
00:00:56
stage. What I mean, this is, this is
00:00:59
your big event. What was What's the message that
00:01:02
you really wanted to carry to the attendees here?
00:01:05
You know, I would say, Eric, first, thank you for having me
00:01:09
and thank you to you for letting me crash the party of.
00:01:13
Course I love it. So we.
00:01:15
Just we just literally we just talked to a three person
00:01:18
company. So now we're going much larger.
00:01:21
Yeah. Sounds good, Eric.
00:01:23
So 65 days in and one of the things that our previous CEO,
00:01:30
Dev and the entire team, we realized, so first of all, from
00:01:35
just being in Silicon Valley for the long time, San Francisco
00:01:40
feels like it's back. San Francisco is back.
00:01:43
During the pandemic, people went a little dark on San Francisco.
00:01:46
But with this AI platform shift, San Francisco is back and as you
00:01:52
know, we had a New York headquarter, New York founded
00:01:55
company. Listen, I'm I'm a New York based
00:01:57
person who also believes in San Francisco.
00:01:59
So we have a sort of shared spirit on that realizing that AI
00:02:03
boom is here, we are also in New York.
00:02:05
So yeah, I feel it. So lot of you know, so San
00:02:10
Francisco is back. Mongo DB about 10 years ago did
00:02:14
a really nice job at San Francisco.
00:02:16
They were in front of software developers, builders saying
00:02:20
build on Mongo DB. Here is why.
00:02:22
Speed, agility, scale out many, many advantages.
00:02:26
And then the team realized, Dave and the team that we kind of as
00:02:32
the company succeeded, we took our eye off the ball for a lack
00:02:35
of better term. And so after four years, we
00:02:40
decided to reintroduce Mongo DB today in San Francisco.
00:02:44
So today's an auspicious day. Thank you for coming and we
00:02:49
wanted to launch and tell people that we are here, we are Mongo
00:02:53
DB. It's a great data platform you
00:02:54
can build on no matter what kind of applications you're building
00:02:57
on, digital native, AI native or you have AI plus plus, whatever
00:03:01
the case might be, it is a great database.
00:03:04
And you know, one of my biggest profound realization was that
00:03:09
when I, when I was doing my own diligence to join Mongo DB, even
00:03:13
though the founders didn't create with AI in mind,
00:03:17
unstructured data, flexible schema, you know, Symantec
00:03:22
Retrieval and all these things that are part of now the
00:03:25
platform, it is like the platform for AI applications.
00:03:29
And so we wanted to tell that to everyone.
00:03:32
And we got a great customer today to validate a great
00:03:36
founder who said he has built three companies on Mocodb.
00:03:39
You're talking about Mike Krieger.
00:03:41
Yeah, you had the nice self video Instagram founder.
00:03:46
I created artifacts, which you talked about briefly, which is
00:03:49
an interesting for me in the news world.
00:03:51
And then obviously he's at Anthropic.
00:03:54
Yeah, that was an exciting endorsement.
00:03:58
What what is your read on? You know, I don't know, AI hype
00:04:01
in 2026. On some level, you must think
00:04:03
this is obviously a phenomenon to say you're doing this big San
00:04:07
Francisco push. Dave, you know, I think I talked
00:04:10
to him last year. You know, he has a dose of
00:04:12
realism to it all. What's your what's your personal
00:04:15
perspective on where we are in this hype cycle?
00:04:17
He's smart and realist and I'm an optimist, so that's how I.
00:04:21
Brought in the guy like, Oh no, this thing is going strong and
00:04:24
we're we're going to get in front of it.
00:04:25
Yeah. I mean, I, I would say since
00:04:28
2022, Christmas ish or October ish, when you look at the
00:04:32
evolution of AI and you look at some of the killer AI companies
00:04:37
where they have scale business in a significant way has been of
00:04:42
course open AI. Everybody understands that that
00:04:44
is a killer app from my perspective.
00:04:48
But you also look at Grog and how fast they have grown X AI
00:04:52
and then Entropic, the entire team at Entropic Lab, that is a
00:04:58
truly killer AI app for coders. I mean, people load that
00:05:04
platform and they sometimes use it with cursor this that.
00:05:07
So you're seeing that these companies, if you look at like
00:05:12
Internet age in 90s, you look at mobile age, they did not scale
00:05:16
this fast. How how fast this companies have
00:05:19
scaled. That's like pretty amazing from
00:05:22
my standpoint. And in 2025, I think the killer
00:05:26
apps were the coding tools. There was a killer app besides
00:05:29
chat GPP of course. And now as we go into 2026,
00:05:34
which ones are going to take off?
00:05:36
I'm optimistic that some very specialized vertical apps
00:05:40
related to maybe a healthcare or insurance insurance assist or
00:05:44
something may take off, but you will see some take.
00:05:48
Off close to, you know, a bridge and open evidence.
00:05:50
Those are super interesting applications.
00:05:52
And health, Yeah. You interviewed Constantine and
00:05:55
Sequoia. Obviously, they're big.
00:05:56
They're bullish on Harvey, you know.
00:05:58
Yeah. So a lot of these vertical
00:06:00
applications are exciting. And Harvey, you know, I met them
00:06:04
a couple of years ago and they were still, you know, truly with
00:06:10
the legal firms they were using them.
00:06:13
And then now you have the in house counsels use them as well.
00:06:17
And so they have gotten a perfect product market fit and
00:06:20
they're expanding now their use cases.
00:06:22
So that's just it. But that's very specialized,
00:06:24
right? It's not just generic.
00:06:25
I come in and help me review this contract from a legality
00:06:30
perspective and so on. This is a drop by, so I don't
00:06:33
want to take too much of your time.
00:06:34
This will be the last question. How do you think about, you
00:06:37
know, there's so many models available.
00:06:40
How do you think about like, oh, where to provide your own versus
00:06:45
to just sort of say you're obviously going to bring models
00:06:48
from other places? We provide sort of the database
00:06:50
layer. For databases, we want to be
00:06:54
model agnostic, Yeah, we want to provide best embeddings.
00:06:57
So your retrieval quality is high, accuracy is high, but we
00:07:00
want to be model agnostic. And when I speak to customers,
00:07:04
including AI companies, like everybody who originally was
00:07:07
telling me open AI, then they shifted to cloud, I think the
00:07:12
innovation cycles are very fast. And then you look at some of the
00:07:15
firms in China and you look at DeepSeek and others that
00:07:18
innovation. Cycles come from anywhere.
00:07:20
Yeah, that could come from and the product cycles are shorter,
00:07:24
right. Product cycles are shorter.
00:07:26
So multi model, not multi modal, but multi model is will be a way
00:07:32
to go and it gives freedom to people to use whichever model is
00:07:37
best. Gemini, maybe today, better
00:07:39
tomorrow, maybe an topic again. And I think that's how it's
00:07:42
going to work. So we will be agnostic
00:07:44
regardless of who you use. Our goal would be always to give
00:07:48
best semantic retrieval capabilities and completely
00:07:54
provide a scalable data plan. Well, I love, you know,
00:07:58
newcomer, we use green so it was easy to share the stage, but
00:08:01
honoured that we get to share branding with Mongo.
00:08:03
Thanks so much for having me here and thanks for dropping by.
00:08:06
Thank you, Our stage. We appreciate it and we'll see
00:08:08
you in Brooklyn soon. Sounds good.
00:08:10
Alright, thank you. Thanks again to Mongo DB for
00:08:12
sponsoring this episode. I feel like I've talked to both
00:08:15
CEOs, the old and the new in the last six months.
00:08:17
So that's been a lot of fun. And now excited to get into it
00:08:20
with Anchor Bot, the head of AI at Ripling.
00:08:23
I know he's big boss Parker Conrad pretty well.
00:08:25
He's been on this podcast before.
00:08:27
Parker has been somewhat slow to embrace AI.
00:08:29
So it's fun to talk to his head of AI about, you know, the
00:08:32
company finding AI Jesus, believing in AIA little bit and
00:08:37
how they're making that cultural and technological
00:08:40
transformation. Give it a listen.
00:08:42
Thrilled to have anchor bot, the head of AI at Rippling here.
00:08:46
I'm just saying, before we got on, I go way back with Parker
00:08:49
Conrad, your Ceoi wrote about Zenefits back in the day.
00:08:52
I I feel like I was, you know, as far as reporters can be
00:08:55
bullish. I was pretty bullish on his
00:08:57
comeback in Rippling and have been following your work.
00:09:01
What's it like to be the head of AI for?
00:09:04
I feel like for a Parker is a technologist for the guy, but he
00:09:08
has not been wrapping himself in AI.
00:09:10
Like how's that How how is it to be the head of AI and CEO sort
00:09:14
of coming to the the AI hype? I think it's been exciting
00:09:19
specifically from the lens of there is so much happening in AI
00:09:22
everyday. So automatically everybody's
00:09:26
curiosity on what's the new model release, what's the new
00:09:30
capability release? And like like these days,
00:09:32
everybody's hyped up on cloud code as an example, what
00:09:36
Entropic is doing. So that automatically starts to
00:09:39
create this pull internally around.
00:09:42
What does this mean for Rippling from a product point of view?
00:09:46
What does it mean for Rippling in terms of day-to-day business
00:09:50
perspective? And what does it mean in terms
00:09:51
of customers expectation on Rippling it, right?
00:09:55
You get to you get to be the guy who knows what's happening
00:09:57
today. I say, oh, how do we respond to
00:09:59
this, this, that and the other we're obviously, you know, at
00:10:02
Mongo DB dot local, you know, obviously sprawling ambition
00:10:07
with Rippling, it's like, oh, we want to stack startup on
00:10:10
startup. So I can imagine strong data
00:10:13
organization capability is part of the business success.
00:10:17
But explain Ripling's relationship with Mongo.
00:10:19
Yeah. So we are being a long time
00:10:21
customer of Mongo from the beginning.
00:10:24
I think as you rightly pointed out, Parker's thesis was
00:10:28
compound startup like product over product.
00:10:32
And the core, heart of it is the employee graph.
00:10:36
And being even though yes people know us that we can run payroll
00:10:41
for you. Core of it is the employee graph
00:10:44
which captures not just your pay related information or benefits
00:10:48
related information. We also capture your ID and
00:10:51
identity related information, your device related information,
00:10:54
because at the end we also have a product portfolio of ID.
00:10:59
Similarly, we also have a product portfolio around
00:11:03
finance, your corporate card, your spend, your travel expense.
00:11:06
We just launched a travel expense product.
00:11:08
So that's certainly the amount of data you have about employees
00:11:12
gets connected in so many different ways.
00:11:16
And having a partnership with Mongo allows us to continue
00:11:20
keeping that graph sanity intact in terms of the relationships
00:11:24
you have in this business data. Rigorous are you guys about
00:11:30
making sure that everything built at Ripling uses the same
00:11:33
tech stack? I mean, if you're acquiring
00:11:35
startups and you're letting people sort of do their thing,
00:11:38
how do you balance the trade off of sort of a consistent stack
00:11:42
versus letting different people build what they want?
00:11:44
Oh, that's a great question. And speed versus consistency is
00:11:48
a constant dialogue we have internally.
00:11:51
I think there is a value in terms of letting people innovate
00:11:56
on the side, but it's a little bit of a trade off because as
00:12:00
soon as they are reusing the common capabilities we already
00:12:04
have in our platform, there is a dimensional speed which comes
00:12:09
with it. And certainly for example, the
00:12:12
employee graph comes with the workflow, it comes with the
00:12:15
notification, it comes with an ability to you know, track the
00:12:19
changes that's happening now. If I am building now our travel
00:12:22
expense product that are workflows for me.
00:12:26
Of course there is a dimension where I need to go into
00:12:30
integrating into flight booking, hotel booking, which is very
00:12:33
unique to my products niche. And there I'm free to choose
00:12:39
what I choose to do that and run faster.
00:12:42
But I think we've been able to guide the different product
00:12:45
teams even though running independently on the core value
00:12:49
of building on that common employee graph and common
00:12:53
capabilities. What has been the value of Mongo
00:12:55
like? Would would Rippling be a
00:12:57
different company who is built on Postgres?
00:13:00
I think, I think it would, no, Rippling would not be a
00:13:04
different company because of Parker's overall hypothesis of a
00:13:08
compound startup and really building it as a connected
00:13:12
ecosystem of products running on a common foundation.
00:13:15
I think it would have been just a little bit harder, I think and
00:13:19
having that partnership with Mongo from day one allowed us to
00:13:23
build this employee graph, build a capability set so that the
00:13:28
next product or the next product we were creating was a lot more
00:13:31
easy. And this is not just about
00:13:33
transactional data which we are maintaining and managing.
00:13:38
It's also about analytics and reporting.
00:13:40
Being able to take that information and have the ability
00:13:43
to report across a cross section of your product portfolio or
00:13:47
your business data, not just from HR or IT or finance allowed
00:13:52
has opened up those possibilities which having a
00:13:56
common data store, having this doc Mongo capabilities opened up
00:14:02
those possibilities for us, which we would have solved
00:14:05
eventually. It would just have taken a lot
00:14:08
more effort. So you're the head of AI.
00:14:10
We touched on some of that role means translating what's
00:14:14
happening in the broader San Francisco ecosystem to to
00:14:17
Rippling. But in terms of, you know,
00:14:19
products within Rippling, what what are the AI products?
00:14:23
Do you have an agent yet? What like what's that look like
00:14:25
inside of Rippling? Great question.
00:14:27
I think originally Rippling also started.
00:14:30
We have almost 70 plus products in the portfolio and originally
00:14:34
each product started innovating around how they can embed AI as
00:14:39
they're building that product, right.
00:14:41
It could be within our recruiting product portfolio.
00:14:43
When you're doing interviews, you are meeting candidates
00:14:46
summarizing those information. It could be within our IIT
00:14:50
product portfolio where you are issuing devices, you are
00:14:54
tracking compliance and security around those things and having a
00:14:58
summary of your policies. So people started embedding AI
00:15:01
as product capabilities very early.
00:15:04
Inadvertently, I think the change happened.
00:15:07
The point in time we actively started investing on for
00:15:11
building AI products, you have to use AI everyday.
00:15:14
Without that, I genuinely believe it's very hard for
00:15:18
people to reimagine what it means for their product.
00:15:21
So 10 months ago, we Albert, our CTO really put a charter out and
00:15:27
that's where I came in is driving the AI transformation of
00:15:31
Rippling internally that we use AI everyday, we embrace latest
00:15:37
and greatest of AI tools every day in our workflow whether it
00:15:40
is product design, engineering, legal, finance, sales, marketing
00:15:45
and that. What is that cursor?
00:15:47
Harvey Sierra? What do?
00:15:49
You use the tech stack is full of all the tools you can think
00:15:53
of, right? So of course Gemini chat GBD
00:15:56
cursor cloud codecs, so does. That engineer sort of pick their
00:16:01
preferred. I think so because Rippling
00:16:04
deals with a lot of enterprise customer data.
00:16:08
We do have AI pilot process and I have built up a checklist with
00:16:14
our security and legal so that we can quickly assess a new AI
00:16:18
solution or somebody will request.
00:16:20
So I get request every week. OK, I want to try granola.
00:16:23
I want to try Bisper flow. Typically what we do is we run
00:16:27
through it through an AI pilot process.
00:16:29
Do you know do you use factory? I'm going to be talking to the
00:16:31
factory. It's on my list of an ask from
00:16:34
somebody. Maybe it'll get accelerated
00:16:35
after yes. That could be.
00:16:37
I'm very keen to learn more, which keeps my job a little
00:16:41
curious because a lot of times I can't stay on top of things.
00:16:44
So then the rest of the company is constantly keeping top of
00:16:48
things and asking me can we try this or can we try that and.
00:16:50
Do you, do you have an overall agent at ripling or?
00:16:53
That's one of the new investments we are making.
00:16:56
I think Mongo's partnership has helped us because if you think
00:17:00
about Rippling, we are not just a payroll system, We are not
00:17:03
just a finance or a travel expense system.
00:17:06
We are a full suite of products. So for us, we can build an agent
00:17:12
which is not a payroll agent or a finance agent or an IT agent.
00:17:17
We can actually build a rippling intelligence agent which can
00:17:20
actually answer questions across your entire day-to-day business.
00:17:24
And that's essentially what we have embarked upon, right?
00:17:26
Parker is extremely excited about business.
00:17:29
Intelligence you think is a key output.
00:17:33
Or I think it is really productivity.
00:17:35
So if you think about AI productivity is the, is the
00:17:39
impact which we we see in engineering or product and
00:17:42
design or marketing, right. The same productivity impact is
00:17:46
what we are hoping to provide to our customers because as our
00:17:50
footprint has increased, our customers growth has also
00:17:53
happened. So they are growing from like
00:17:55
look at Andrew, one of our flagship customers.
00:17:58
As Andrew is growing, they are using more products from we're
00:18:02
playing, they have larger set of ibase.
00:18:05
So essentially their day-to-day operations has gotten complex.
00:18:10
Yeah. So now if you are able to offer
00:18:12
them an agent to operate Android payroll to their IT to their
00:18:18
travel and the agent takes care of, you know, proactively
00:18:22
working with their administrators and making sure
00:18:26
Android stays on top of things which they need to say on is the
00:18:30
value we are aiming for. I hate to bring up a competitor,
00:18:34
but ramp and rippling start in different places.
00:18:37
They overlap on some. You now have, I think credit
00:18:41
cards, yes. And you both want to be sort of
00:18:43
the what, what's the term of art, like the record of all
00:18:47
things business for somebody. They've clearly leaned into the
00:18:51
AI brand more than you have. I don't know, what do you, what
00:18:54
do you you're the AI guy? Like do you want, do you want
00:18:58
that AI brand? Or it's like we have different
00:18:59
sensibilities. Or how do you think about it?
00:19:01
I think there is a, there is obviously an equity in terms of
00:19:05
having a clear AI story around your company and your product
00:19:12
because customers are keen to understand that because
00:19:15
specifically because if I'm a ripping customer today or I'm a
00:19:19
prospect evaluating Rippling today, I come with that
00:19:21
expectation that there is a certain amount of AI automation
00:19:26
productivity I will get. So it's, I, I commend Ramp for
00:19:32
what they have been able to embark on, not just on the
00:19:34
product side, but as well as in terms of positioning themselves
00:19:38
as an AI forward company. I think Rippling is not very far
00:19:42
in terms of our AI transformation internally.
00:19:45
We are very further along than what people may be aware of in
00:19:49
terms of our design partnerships.
00:19:50
We have design partnerships with Cursor, with Open AI, with
00:19:53
Entropic, with AWS where we get early access to the latest and
00:19:59
greatest land chain data breaks early and even Mongo, so early
00:20:03
access to their product capabilities.
00:20:05
We are piloting that at tripling using that to create
00:20:08
productivity and give feedback back.
00:20:10
And when I see those interactions and when I see our.
00:20:14
The ability to influence product road map of so many ecosystem of
00:20:17
AI companies it reflects back on Rippling's AI impact is much
00:20:22
larger than what people may be aware of.
00:20:24
I should know that where is is enterprise search part of
00:20:27
Rippling's vision because that's been an obviously clean is very
00:20:31
promising AI company. I would say part of Rippling's
00:20:37
AI assistant will take care of answering questions you'll need
00:20:40
answers for right across your enterprise, right?
00:20:43
Being able to obviously be being the system of record of a lot of
00:20:47
information, right? You don't need glean in those
00:20:50
cases because we can just answer those questions for you
00:20:52
directly, right? What do you think?
00:20:55
There's been a lot of AI is going to replace software
00:20:58
companies. I mean, so you get it from both
00:21:01
ends, embrace AI, but also like, oh, you know, you know,
00:21:04
individual companies are just going to spin up their own
00:21:07
payroll because, you know, they can build it with cursor on
00:21:11
their own. Like, I don't know, payroll
00:21:12
obviously to me seems sort of absurd because there's all the
00:21:15
regulatory and it just like sensitive, but clearly there are
00:21:19
there must be pieces of your business where people are sort
00:21:21
of hacking together their own apps.
00:21:24
Like, what do you make of this narrative that the rise of
00:21:26
artificial intelligence is going to have all these sort of
00:21:29
homespun software applications? We actually embrace that
00:21:33
wholeheartedly because what we find at at Rippling is what we
00:21:38
find at Rippling is as customers are using our platform, there
00:21:41
are always niche unique use cases for which they like parts
00:21:47
of what we offer to them, but they want to extend, enhance and
00:21:51
add new capabilities on top of. Which is why beginning of last
00:21:57
year, middle of last year, we launched our custom app
00:22:01
capabilities where you can essentially vibe code an
00:22:04
extension app to Rippling ad Rippling's platform itself.
00:22:07
Interesting. And again, going back to since
00:22:10
we are at Mongo's conference, right, being on Mongo in terms
00:22:14
of ability to store generic artifacts and documents actually
00:22:19
opened up that possibility for us essentially then to create
00:22:23
custom objects which then can be used as a container for any
00:22:28
customer to bring in whatever data they want to bring.
00:22:30
In any interesting examples? That you get, oh, tons of very
00:22:34
unique, very interesting examples in in companies,
00:22:39
because we deal with a lot of workers who are dealing with
00:22:42
shifts, for example, and compliance comes up quite often
00:22:45
in terms of having them having gone through certain amount of
00:22:49
rigors in certain industries. Very unique cases in terms of
00:22:53
tracking, you know, ticketing, tracking, sliding sign outs,
00:22:59
tracking. So there are like niche sort of
00:23:02
use cases people are building, which has actually motivated
00:23:05
Rippling to even launch RFD team, which is actually now
00:23:11
going in to our customers and building these niche customer
00:23:14
apps for them. And then seeing how we bring
00:23:18
that into our platform as capabilities in terms of
00:23:20
enhancing a platform to make that easier going forward.
00:23:24
Do you think your code base gets a little worse the more people
00:23:29
use AI tools or what's the risk in terms of over reliance on
00:23:35
cursor? That's a great.
00:23:36
That's a great question. I think we've started to observe
00:23:41
that, yes, there is a dimension of an AI slop seeping in to the
00:23:48
engineering discipline of coding every day, right?
00:23:51
But I think we've been very strict about our AI stance on
00:23:56
couple of things. First, testing is non
00:23:58
negotiable. So any code which is being
00:24:00
checked into the main line and pushed to production has to be
00:24:04
fully tested and vetted. Second, accountability doesn't
00:24:08
just the fact that I used cursor in AI doesn't take away my
00:24:11
accountability of what that code does.
00:24:14
And I think we've been very clear to engineers from day one
00:24:17
that accountability is still there.
00:24:20
So that essentially creates sort of a responsibility
00:24:25
accountability in engineers to be cautious of it.
00:24:28
But that's just on the human side.
00:24:30
From a system side, we're also bringing in additional AI tools,
00:24:34
so AI coding, code review tools, AI tools around looking through
00:24:38
in terms of production outages and bringing that information
00:24:41
back to to engineers to be able to troubleshoot and then
00:24:46
improve. Because at the end of the day,
00:24:49
the AI enablement or AI use is not just about development.
00:24:54
AI can also be used in analysis of what's happening in
00:24:57
production. AI can also be used in reviewing
00:24:59
and verifying. And essentially by bringing
00:25:02
those systems in those places, overall keeps the health of the
00:25:07
code base at a peak and avoids the slops sort of just bleeding
00:25:13
into production. In terms of the rippling
00:25:17
product, maybe have you started to build it at all with the idea
00:25:22
that it's like someday soon people are going to have AI
00:25:25
agents that they treat as a worker?
00:25:28
It's like, yeah, I want to in some ways compare apples to
00:25:31
apple, this human and this AI agent.
00:25:34
And I'm going to like think about it in terms of identity in
00:25:37
the same way I would use rippling is that started to
00:25:40
creep into. Your great.
00:25:42
I don't know, it's like you're just just sort of picking up
00:25:45
things we are discussing internally because one of the
00:25:47
things we've been thinking about is agent identity, right?
00:25:51
Because now if we think about agents we are building for our
00:25:56
customers, at the end of the day, they'll start accessing
00:25:58
payroll information. They'll start accessing, you
00:26:01
know, information which has its own access management of who can
00:26:04
see what and how and what type of actions they can take.
00:26:06
Yeah, So it does need three things.
00:26:11
One, obviously an identity of an agent, which we know who's doing
00:26:14
what. But then second, it should still
00:26:18
inherit from a human, because in some ways that's one of the
00:26:23
conclusions. At least we landed on that.
00:26:26
At the end of the day, humans. Culpable for this agent?
00:26:30
Exactly, exactly, exactly who is who is on the hook for the agent
00:26:35
at the end of the day and whose, whose team This essentially
00:26:39
which translates into whose permissions, whose access
00:26:42
management this agent has to have.
00:26:44
And we found that it's a lot more easier to say that, OK,
00:26:48
there is always a human. There is always a human in the
00:26:50
system. Of course, what we are adding is
00:26:53
a productivity element. So instead of in future thinking
00:26:56
of adding more humans, you're essentially adding more agents.
00:26:59
But they still function as part of a payroll team and then they
00:27:02
get access according to being a payroll admin, right?
00:27:05
They are part of an IT team and they get access as part of an.
00:27:09
I feel like a lot of people are excited about what agents can
00:27:11
do, but not excited about being the one responsible for stopping
00:27:15
the agent from doing what it's not supposed to do.
00:27:17
Is Rippling willing to take that on at all?
00:27:20
Or like, do you, you see a world where you say, yes, we're the
00:27:23
barrier for your agent not getting access to some system
00:27:27
it's not supposed to? Have that already.
00:27:29
That already is something part of Rippling's ethos.
00:27:32
So we because we're not just a payroll company, we also have a
00:27:36
full blown identity system, we have a full blown access
00:27:40
management system. We already control what data who
00:27:43
has access to. We have a very well defined
00:27:46
access management platform essentially.
00:27:49
And if you which is what is allowing us to now bring in
00:27:52
agents and say, oh, this agent inherits from Encore and
00:27:55
whatever Encore can do in rippling this agent can do
00:27:58
right. And that philosophy essentially
00:28:01
allows us to guardrail not just when we are building agents
00:28:05
ourselves, but customers deploying agents in their
00:28:08
landscape, being able to then start performing tasks and
00:28:13
actions and leading to unintended consequences because
00:28:17
we will still keep that principle even for them
00:28:20
accessing ripping data. We guardrail that heavily.
00:28:23
The API access is only enabled through certain controls and
00:28:27
access management principles, so randomly customers can't deploy
00:28:31
an agent and start accessing ripping information.
00:28:33
Well, you have what every AI company wants, which is great
00:28:37
data. And now it sounds like you've
00:28:38
got the AI conviction. So exciting times ahead at
00:28:42
Rippling and thanks for joining us.
00:28:44
No, thanks for the great conversation and really enjoyed
00:28:48
talking to you. And yeah, let's continue this
00:28:50
dialogue. Sounds good.
00:28:51
Thanks so much. Thank you.
00:28:52
Thank you for tuning in to this week's episode of the podcast.
00:28:54
If you're new here, please like and subscribe.
00:28:56
It really helps out the channel. Listen in for new episodes every
00:28:59
week wherever you get your podcast.
00:29:00
Yes.