Latest / AI for Founders with Ryan Estes / Your company Is sitting on a data gold mine

Transcript

Your company Is sitting on a data gold mine · AI for Founders with Ryan Estes

0:00
Welcome back to AI for Founders. I'm Ryan Estes.
0:02
There's a mountain every company is quietly climbing, not of code
0:06
or customers, but of unstructured data.
0:09
Billions of pages, decades of contracts, PDFs and reports
0:12
trapped in digital silos no human has time to read.
0:16
It's where trillions of dollars in untapped insights are hiding
0:19
and where most AI systems still fail.
0:22
Today we meet the man building the rope line up that mountain.
0:25
Mehul Shah, a formal AWS engineer behind Open Search and
0:30
Glue, now leads Aaron dot, AIA company on a mission to help
0:34
enterprises turn the 90% of their data they can't analyze
0:38
into an engine of intelligence. If you're a founder, data
0:41
leader, or investor wondering where real AI gold rush begins,
0:45
not in a flashy demo, but in hard, unseen work of structuring
0:49
the world's chaos, this episode is for you.
0:52
You'll learn why enterprise AI deployments fail.
0:54
What human in the really means and how the next wave of
0:57
productivity won't come from chat bots, but from invisible AI
1:01
that simply makes work work. So grab your coffee, open up
1:05
your mind, and get ready to rethink what data really is.
1:08
Here we go Now do you want AI that actually ships revenue, not
1:12
hype? Join the founders building it at
1:15
AI for founders.co. This is a newsletter featuring
1:18
the next great startup that won't sell prompts, it sells
1:21
outcomes. Learn about it at AI for
1:23
founders.co. Every week, we deconstruct the
1:26
humans behind the code. This is the newsletter at AI for
1:30
Founders Co, and if this episode made you rethink what's really
1:33
possible with AI, leave us a review before an agent write
1:36
ones for you. This episode is brought to you
1:39
by Kitcaster, the agency that gets founders booked on the
1:41
world's top podcasts. Your story deserves a stage.
1:45
Kickcaster Teams finds the perfect podcast where your
1:48
customers are already listening and handles everything from
1:51
pitch to prep to placement. No cold emails, no guessing
1:54
games, just great conversations that build your brand, drive
1:57
leads and land deals. Founders from companies backed
2:00
by Sequoia, Andreessen, and NYC trust Kitcaster to get their
2:04
voice heard because it's a noisy world and conversations cover.
2:08
So if you're ready to stop chasing attention and start
2:11
earning it, head to kitcaster.com/application and
2:15
let them book your next big interview.
2:17
That's Kitcaster by MO Burst. Here we go.
2:34
Go. Welcome to the show.
2:36
Thank you for coming on. Ryan, thanks.
2:39
Thanks for having me here. Cool.
2:41
I am so excited to talk to you about data.
2:44
This is data is the the component I think of technology
2:47
that I know the least. So if I stumble and stagger
2:50
around this conversation, I just want to apologize right now, but
2:54
we're going to be talking about your company, Aaron, which is a
2:57
RYN. Give us a brief overview of of
3:01
what your company does and who they serve.
3:04
Absolutely. So Aaron, Aaron, the name
3:08
actually it's a, it's a, it's a Gaelic word.
3:11
Also a Hebrew word in Hebrew means tall mountain.
3:15
And the connotation there is that we want to help enterprises
3:20
scale the tall mountain of unstructured data and summit
3:24
that peak. And so that's where the the, the
3:26
name comes from. And what we do is we really help
3:32
people, help enterprises pull out insights, pull out key
3:39
information from all the unstructured data that they
3:42
have, in particular, you know, documents and, and the current
3:48
product that Aaron has is really helping you structure your
3:53
unstructured data. OK, so imagine you have, let's
3:57
say you have, you know, insurance submission packets, or
4:02
you have illegal filings, or you have real estate offering memos.
4:08
In all of these cases, enterprises have tons and tons
4:11
of knowledge workers. We're literally going through
4:15
these documents by hand today and, you know, painstakingly
4:19
pulling out key pieces and for information like, you know what,
4:23
what am I actually insuring? If it's an insurance document or
4:27
if it's an M and a document, you know, what are the key
4:31
provisions in the M&A when the company or asset that's being
4:35
acquired is being considered? Or if it's a real estate
4:39
offering memo, you know, what's the asking price, you know, is
4:43
it something that I can rent? And so this kind of information
4:46
is actually usually in all of the unstructured data that you
4:49
have. And you and these enterprises
4:52
and the knowledge workers in these enterprises need that
4:54
information to do to do their daily work.
4:56
And today it's incredibly, incredibly, painstakingly,
5:02
painstakingly hard for them to go do this.
5:05
They spend hours and hours doing this.
5:08
And it's critical for that they get this stuff right.
5:10
Otherwise their businesses start losing money.
5:12
And So what we do is we use modern AI technologies to be
5:15
able to pull this unstructured dis structured information out,
5:19
out of their unstructured data. And then they can take that
5:22
structured information and do their downstream tasks.
5:26
They might do analysis, they might quote a premium for a, for
5:31
a policy, you know, they might decide, you know, what their
5:35
risk profile looks like. If they're going to do a
5:37
particular M&A, they might find, you know, legal precedent that
5:44
they need to be able to go and prosecute a case.
5:46
And so you can do a lot of different things if you can pull
5:49
key information out of this data.
5:50
And that's what we'd help help. Them do amazing.
5:53
And I think it's an important differentiation that when you
5:55
say documents, you're meeting paper and you said by hand, it's
6:00
like the it's in their hand and they like literally will have
6:03
like a highlighter or something writing in the margins, you
6:06
know, So it's actual paper, which means that like, you know,
6:09
and if you, if you see like, remember like a like a business
6:13
movie from the 80s and you'll have just stacks of papers and
6:17
boxes in in rooms. And I think a lot of people just
6:21
put doors on those rooms and they still exist, right.
6:24
And so that this becomes some kind of like not only like
6:28
information and knowledge that's trapped, but it's also like tons
6:32
of money is locked up with like the the ways that they could be
6:34
like deploying that information. Absolutely, absolutely.
6:38
So there's a goldmine of information in the sun,
6:41
structured data that they have. And right now it's locked up.
6:45
Now the the good news is that we've moved a lot of enterprises
6:49
at least in, in, in the Western world, in the sort of the larger
6:52
economies. I should say though, in some of
6:55
the sort of emerging economies, you know, the picture that you
6:58
drew is actually still still correct.
7:01
But in sort of the, the, the more advanced larger economies
7:05
that are out there, we've moved past the, the rooms that, you
7:10
know, it's stacked full of papers, though I bet, I bet you
7:12
can still find them. You find them, I bet.
7:14
Yeah. We've moved past that but but
7:17
imagine a virtual world of a stack room full of papers, but
7:20
sitting like inside of these silos in your enterprises.
7:24
And the kind of silos I'm talking about is, is electronic
7:26
silos from databases from the 60s or 70s, Cobalt databases,
7:31
for example, or, you know, sitting in literally just
7:36
spreadsheets, Word documents, PD, FS, you know, just text
7:44
files. In fact, in some cases in filers
7:48
in the cloud on in your, you know, online cloud stores that
7:53
you have and so on. And it's really just a
7:56
disorganized mess, if you will. OK.
8:00
And, and if, if companies can really, you know, harness that
8:04
information, they're not only going to be able to be much more
8:07
productive, meaning they're going to be able to save their,
8:11
their, you know, their, their employees a lot of time hassle,
8:15
but they're also going to be able to make a lot more money.
8:18
And to be able to make that money.
8:20
What we have found, you know, as we're kind of working with these
8:23
companies is it's not just enough to be able to find the
8:27
right piece of information. It's important to be able to
8:30
accurately pull out the key pieces of information so they
8:34
can act on that information. OK.
8:36
And and that's where I think we differentiate and it's not just
8:40
finding the things, it's that we can pull out key pieces and
8:44
information, we call them fields or properties with extremely
8:48
high accuracy. And we can do that at scale.
8:51
And so not just simple demos where I show, you know, A10 page
8:55
document and look, I can, you know, I can tell you what the
8:57
addresses look like, but you can give us documents with thousands
9:00
and thousands of pages in them, all kinds of graphs, all kinds
9:04
of crazy, you know, looking tasks.
9:06
Sometimes it's going to be a scan of a document or a fax.
9:12
And, you know, the AI systems that we build and that we use
9:15
can accurately pull that information out.
9:19
And use it in novel ways. I'd love to know, you know,
9:22
maybe some unexpected ways that you're using AI or, and I'll put
9:29
it in a frame for you because, you know, I realize I'm kind of
9:34
giving AI broadly this kind of like personification in a
9:38
similar way that when I was a child, I thought about bears
9:42
because I love blue from Jungle Book.
9:45
And so, you know, you're out hunting in the woods.
9:47
You're like, oh, a bear. And like, I think of it like
9:50
Disney, a Disney version of bears.
9:53
I'm finding myself more and more largely because AI has gotten so
9:57
broad and so such a a large swath of like machine learning
10:02
and just basic automation. Of course you got agents and
10:05
this and this. All the terminology is kind of
10:07
like dignifying AI to me that I almost, I'm seeing it like some
10:11
kind of like invisible person that lives in the Internet.
10:14
Yeah, exactly right. And it's like this elf that you
10:18
just kind of, you know, you, you feed you, you feed in something
10:21
and outcomes something magical. And, you know, and many of our
10:25
customers actually have this kind of expectation.
10:27
Ryan, we, we come in and we'll be like, Hey, we use AI to do
10:30
document processing. And, you know, do you have a use
10:33
case for us that we can help you solve?
10:35
And they come to us and they're like, sure.
10:37
And I'm like, guy, what's the use case?
10:39
Can you do my job for me? I'm like, maybe it's like, you
10:43
know, this expectation that, you know, AI is going to come in and
10:47
just, you know, completely replace your job.
10:50
Well, and I think. Even like literally we'll punch
10:52
the clock like like if that your agent that you built it like
10:55
waits till 9 or it's or even better yet, it's like it works
10:59
when you're sleeping, you know, it never stops working.
11:01
So you like imagine like a diligent employee that's, you
11:05
know, an elf on the Internet working where it's like, dude,
11:09
agents aren't real. Thanks, you know.
11:11
They're not, No, they're not. Yeah, they're, they're not real
11:14
things. And I think, you know, I, I
11:16
think this character caricature that of, of AI that you have is
11:21
certainly how people think about it.
11:23
And I, you know, I, I tend to think that it's overblown.
11:27
OK, so I first let me be extremely clear.
11:31
AI and you know, the progress that we've made over the last, I
11:35
would say, 5 to 6 years with LLMS, large language models that
11:39
process not just text, but now video and multimodal data and so
11:43
on, images and sound and audio has been just like tremendous.
11:48
No, I couldn't have predicted it.
11:50
You know, we in technology often think about like, hey, what's
11:52
going to happen next? This just blew my mind.
11:55
OK, so the advances that we've made are just enormous and so
11:59
much so that people think that like just magic is going to
12:02
happen. But I think it's important to
12:04
understand that AI isn't this sort of magical creature in the
12:09
woods that's just going to go, you know, do stuff for you.
12:13
AI is, you know, exactly, you know what the, the inventors of,
12:20
you know, these models intended it to be a general tool, OK?
12:24
A general tool that has a bunch of capabilities that that, you
12:30
know, reflect human type capabilities.
12:33
So it can read, it can respond to you, it can, you know,
12:39
summarize things for you. I'm, I'm talking in the terms
12:42
of, you know, documents, but you can think about all kinds of
12:44
other use cases and so on. And it is incredibly fallible,
12:50
just like humans are, right? And, and you know, you can see
12:54
that AI is getting better and better.
12:56
It's getting more and more intelligent in some cases.
13:00
You know, the places where it's getting more and more
13:02
intelligent, it's actually getting more intelligent than
13:05
humans. For example, They can identify
13:07
as plant species better than humans can.
13:10
Now, if you take a picture of it or it can, you know, identify
13:14
breast cancer in a picture sometimes.
13:17
And again, I don't quote me on any of these things because I
13:20
think doctors will, will, will come back and say no, no, not
13:23
really. But they can really do a certain
13:25
set of things extremely well. The thing that it can't do is
13:29
just be all knowing and everything for everybody.
13:35
OK. And so, So what the what's
13:37
happened now is that if you take, you know, you sort of old
13:42
AIMLI think is what you called it, but let's just call it
13:47
traditional AI. What you have to do is you have
13:50
to have a bunch of experts, AI experts, ML engineers and so on.
13:54
And they had to really focus on the one very specific use case
13:58
that you have. And then they had to train
14:02
models to make sure that they had lots and lots of data.
14:05
I've seen lots and lots of examples of the kind of use case
14:09
that you had. Maybe it's, you know, invoice
14:11
processing and they've just seen lots of invoices.
14:13
Or maybe it's, you know, maybe it's a classification of your
14:19
documents, whether it's a patent or whether it's a manual 1.
14:22
So, and they've seen lots of examples of these things.
14:24
And it took a lot lot of effort and a lot of energy and a lot of
14:27
money to get these things to work well.
14:30
And what I think fundamentally has happened in the last five
14:33
years is that barrier to entry has dropped a lot because these
14:38
huge models that people are building are trained on
14:41
everything in the world. OK.
14:44
And so now, you know, your starting point for the model,
14:47
your starting point for intelligence is just much, much
14:49
higher. That doesn't mean it solves out
14:51
all your problems. It just means that your starting
14:54
point is is much higher. And so that's where we come in,
14:58
you know, we take that starting point and we say, OK, how can we
15:01
get this to work for enterprise use cases?
15:05
And you know, just like a, you know, a human, right, you can
15:10
have real experts in a particular space and then you
15:15
can also have generalists, OK. And so, so these models, these
15:20
frontier models, they're generalists when you, when you
15:23
go into an enterprise, when you hire an underwriter, for
15:27
example, at an insurance company or you hire, you know, attorney
15:31
in a, in a law firm, you're not hiring a generalist, you're
15:35
hiring a specialist. And so this is where, you know,
15:39
the, the gap between a generalist and a specialist is
15:42
where companies like us come in and focus the AI so they can
15:46
reach, you know, reach the, the, the level of the specialist and
15:52
assist the specialist in what they do.
15:54
Now, we haven't completely replaced the specialists and nor
15:58
have we gotten to a point where you can just, you know, automate
16:01
everything where these agents are going to go off on their own
16:04
and you know, you know, hit buttons and, you know, you know,
16:07
run processes for you. But what we have been able to do
16:11
is get them so well, working so well that just a human in the
16:16
loop looking at the results of this AI.
16:20
OK, can, can very easily solve the, you know, do their old job
16:25
much, much faster, much, much more efficiently.
16:29
And then they can be freed up to do the things that they're
16:32
intended to do, which is to use their intuition and decide if
16:36
there is a business decision that needs to be made.
16:39
Use their intuition to decide if you know you should, you should
16:45
take a different direction and so on and so forth.
16:48
So humans have gone are going to go away from just doing menial
16:51
knowledge work to much more high valued intuition based knowledge
16:58
work. And I think that's the
16:59
difference from ML 5-6 years ago and you know, the knowledge work
17:05
that's happening to what's going to happen in the next 10 years.
17:09
Yeah, you know, and it's it's definitely, we're starting to
17:13
get that picture, I think more clearly now.
17:15
The last two years, every time ACEO wanted to make headlines
17:19
for two days, they'd just be like, oh, we're not hiring
17:22
anybody. We're just going to use AI to
17:23
replace everybody. And people panic.
17:26
Oh, no, we're all losing our job now, as we've seen from GPT 3
17:30
all the way to wherever it is now, 40, that it is getting
17:34
smarter. So you can see like, OK, just
17:36
it's going to compound and it's going to be much more smart.
17:38
So one thing I've been thinking about is what happens when, you
17:42
know, AI is 1010 times smarter than humans.
17:45
And so I started looking at like intelligence, as far as like
17:49
output generative intelligence as witnessed by, let's say John
17:53
Coltrane versus Kenny G. Right.
17:56
You know so. Now, no disrespect to Kenny G at
17:59
all. I'm just going to go ahead and
18:01
say that like, you know, if you're looking at circles of of
18:04
fourths and fifths, like what, what John Coltrane is doing is
18:08
much more technically brilliant and and difficult.
18:13
But you know, if you're having a dinner party and you put on
18:17
giant steps, you're not going to make any friends.
18:20
However, if you put on a little Kenny G and turn the lights down
18:25
low, it might be the perfect appropriate thing.
18:27
So, you know, as we're getting to like points where a guy is
18:29
going to be like vastly more intelligence than than humans,
18:33
the will be actually sculpting away.
18:35
So maybe some of the complexity and intelligence and scaling it
18:39
back, you know, which is an interesting kind of, it's just
18:42
an interesting thought experiment because, you know,
18:45
one thing that I'm finding more and more is that in conversation
18:49
with humans, I, I'm much more it kind of like aware that I, I
18:55
want to befriend them. And so a lot of what I say or
18:59
don't say are the questions I ask or the tone or anything else
19:02
is largely because I want to win their favour.
19:04
Whereas when I'm talking to AI, particularly if it's very
19:06
intelligent, that I'm going to be more honest and I can
19:08
actually go for what I want and need much faster, you know, So
19:13
to your point, you know, if we're using like Aaron to fine
19:17
tune this massive database that's building on top of, you
19:21
know, a, a great generalist of these like legacy LLMS, you
19:25
know, to really get like a character that we're building
19:28
that has the intelligence of our organization, it's going to be a
19:31
lot easier to actually go and access that information because
19:34
you don't have to worry about its feelings potentially.
19:38
OK, I've got a confession. As a podcaster and founder, I'm
19:42
constantly meeting incredible people, guest listeners, or
19:45
other entrepreneurs. But here's my little secret.
19:48
I'm terrible at following up. My inbox is basically a
19:51
graveyard of missed opportunities.
19:53
Does that sound like you? You meet someone amazing,
19:55
exchange emails, have great intentions, and then life
19:58
happens. Six months later, you're kicking
20:00
yourself thinking I should have stayed in touch with that
20:02
person. This is exactly why I started
20:04
using Warm Star dot AI and honestly, it's transformed how I
20:08
think about my network. Instead of letting relationships
20:11
go cold, Warm Start analyzes my existing contacts and helps me
20:15
identify who. I should be reaching out to and
20:18
why it even writes the outreach messages for me.
20:21
But here's the key. I review everything before it
20:24
goes out. I'm not just talking about this
20:26
because they're a sponsor. I'm talking about it because
20:28
watching a healthcare executive use this to build his entire
20:31
Advisory Board in one weekend with four CEO sign ups made me
20:35
realize I was leaving money on the table every single day.
20:38
Your network is your net worth, but only if you actually nurture
20:42
it. Check out warm start dot AI.
20:45
Use code AI IFF to get 50 bucks off your first month.
20:49
Stop letting your relationships turn into miss revenue.
20:54
OK, everyone loves to talk about vision, the pitch, the funding
20:57
round, but you know what gets 0 love?
20:59
The actual product Build, that magical, terrifying phase where
21:03
nothing works, the docs are out of date, and you've asked your
21:06
dev team how's it going? For the 47th time today on
21:09
Slack, That's where Code Story comes in.
21:12
It's a podcast for start up founders who know that building
21:15
The thing is the thing. Hosted by NOAA Labheart, Code
21:18
Story goes deep with tech leaders who've been through it,
21:21
shipping real products, surviving messy launches, and
21:25
figuring out what actually works.
21:26
And when you're on the edge of chaos.
21:29
If you're building now or ever plan to, you'll want these
21:32
stories in your ear balls. Search code story wherever you
21:36
get your podcasts. Or visit codestory.co.
21:39
Because founders talk vision. But builders?
21:42
Builders tell code stories. I, I certainly, I, I actually,
21:47
you know, it, it's, it's odd. You know what, what?
21:51
Look at, look at Aaron. What we do is actually really
21:55
basic to the point where you know, yeah, the our users don't
22:00
even know that they're actually interacting with AI.
22:04
So, so for example, you know, you ask, you know, who do we
22:06
serve? Some of our key customers are,
22:09
are insurance underwriters in in Mgas.
22:14
These are specialty brokers that basically, sorry, they're
22:17
brokers of brokers effectively that underwrite specialty
22:20
insurance policies. OK.
22:22
So for example, reinsurance or, you know, addiction treatment
22:26
programs or you know, property and casualty for like trucks in,
22:31
in, you know, with various parts of the country, some very unique
22:34
types of insurance that they're doing.
22:36
And what they do every day is literally cut and paste, OK,
22:40
from, from their documents. And I'm going to get to your
22:42
point in a second, but they're literally cut and paste from
22:44
their documents into spreadsheets or into policy
22:47
admin systems. And what our AI does is instead
22:51
of them doing the cutting them and pasting, the AI is doing
22:54
cutting and pasting. Now you're going to ask me, why
22:57
do I need AI? Can't you, Couldn't you have
22:58
automated this stuff? And the short answer is no.
23:02
And, and the reason they can't automate it is what's coming in
23:04
is so varied and so broad. There was no way to automate it
23:08
in the past. You needed a human looking at
23:10
it. And so now all we've done is
23:13
we've taken something and that a human does OK and automate it.
23:17
The human doesn't actually talk to the AI with Aaron.
23:21
The AI is actually, you know, folding into the background,
23:25
meeting the humans where they are and the workflow workflow
23:28
that they have. And what happens is instead of
23:31
like, you know, reading emails, cutting and pasting and putting
23:34
them into spreadsheets, these emails, the submission forms and
23:38
all these things basically drop in, get processed and outcomes a
23:43
perfect spreadsheet that they can look at, you know, in their
23:46
SharePoint. And if it's not in their
23:48
SharePoint, it could be in their, you know, in their policy
23:51
admin system or can be in their sales force, whatever, you know,
23:55
database or downstream system that they want to use.
23:57
And it can come in, if it's not from e-mail, it can come in
24:01
from, let's say you know, from their, from from their data
24:06
stores that they have on the cloud, or it can come in from,
24:10
you know, some API through a SAS vendor that they've been, you
24:13
know, working with and so on. And so, So what we try to do is
24:18
just take businesses, existing workflows that they have and
24:22
just make them better. OK.
24:25
And so, so, so users don't even know that AI is actually
24:28
happening in the background, but they know they're doing, they're
24:31
doing, you know, much less tedious work.
24:34
And then they're able to then the underwriters are able to
24:36
spend their time working with brokers and working with
24:39
carriers and maintaining those relationships that they have and
24:43
getting more business for the company.
24:45
And that's the win when when Aaron is actually, you know,
24:48
inside your organization. Now that said, what we have seen
24:53
is in many cases, especially in sort of the, the sort of
24:58
customer support scenario, OK, where you actually do want to
25:01
interact with somebody, say, hey, this is my problem, can you
25:04
tell me how to fix it and so on. If you're willing to actually
25:08
talk to a human, you're going to you're going to you're going to
25:12
have a certain persona when you talk to that human, you're going
25:16
to have a certain set of things that you're willing to have have
25:18
a conversation about, Hey, my bike is broken.
25:22
You know, I need a new part. Can you get me this part?
25:26
And if that part doesn't exist, you're going to be polite.
25:29
You're going to say thank you very much.
25:31
But if you're talking to a chat bot and you know it's a chat
25:33
bot, you're going to say all kinds of things.
25:36
You know, you're, you're going to say, you're going to say
25:38
things where you're going to ask questions that might seem like
25:41
dumb questions you wouldn't ask a human.
25:43
You don't want to appear dumb, but you're going to ask that
25:46
question to the chat bot to basically, you know, verify your
25:51
knowledge or, you know, learn something new.
25:54
You're going to go in all kinds of other directions that you
25:57
wouldn't have, you know, actually happen, you know, with
26:00
a, with a real human. So I, I actually agree with you.
26:04
I think, you know, when it looks like a human and talks like a
26:08
human, if you, if you know it's act an actual human, you're
26:10
going to, you know, you're going to act, you know, you're going
26:13
to, you're going to have a different behavior if it's a
26:14
real person versus a chat bot. And so that's part of the reason
26:19
we've kind of gone away. I'm not saying entirely, but
26:22
we've kind of gone away from, you know, focusing on just the
26:26
the interaction parts and really trying to bring the AI and
26:30
meeting humans where they are in the current workflows that they
26:33
have. Cool.
26:36
And I this this kind of like etiquette, AI etiquette, human
26:39
etiquette. I I've really been like kind of
26:42
racking my brain on this one because take for example, I was
26:46
talking to the founder, Loman, and they basically replaced the
26:49
receptionist at a restaurant. Oh, they did, yeah.
26:53
Like you call the restaurant to order a pizza and it's AI and
26:58
just kind of going through that. I was like, you know, when I
27:00
call a restaurant, first of all, I feel guilty, right?
27:04
Because they're busy and they're like, what do you want?
27:07
You know, And they're like whatever.
27:09
And you then you put the order together.
27:11
Now if it's AI, then I'm going to be a lot more honest because
27:14
let's say I got stuck giving the guys a ride home from football
27:17
practice and I got 5 football players in my car, teenage boys
27:21
and I and I was like, all right guys, we're getting dinner now.
27:24
I'm not going to let them go ham with the menu because I'm not
27:26
trying to spend $200 on on burgers.
27:29
So, but so if you call the AI, you're like, look, I got 60
27:32
bucks. Give me the best order you can
27:34
for five people. If we're taking into
27:35
consideration price per calorie. And then they're like, cool,
27:41
they'll be just whipped together your order, like, boom, you pick
27:43
it up and it's good. You know, so like I find myself
27:46
more and more and particularly the more I use AI, particularly
27:49
with kind of voice. And I know we're not, we're not
27:51
kind of getting into like kind of the, the interface really.
27:55
But the more I do talk to it, the more I can actually be
27:58
really honest of like what I'm trying to do and where I'm
28:02
trying to, what kind of information I, I wanted to lead
28:05
to and things like that. So it there's like, and I don't
28:10
even really know what this kind of like, is it etiquette?
28:13
Like what? What is this kind of?
28:15
It's, it's a different persona that you have is what I think it
28:18
is. You know, it's, it's, it's
28:21
unfortunately, I think a human characteristic more so than a, a
28:25
characteristic of the AI. It's that, you know, we come in
28:29
with different personas, whether we're, you know, just just
28:31
imagine like right now, for example, right, you and I have a
28:35
persona, you know, we're have, we're doing a podcast here,
28:39
right? The kind of persona that I have
28:40
when I'm at work and working with my employees is very
28:44
different. We're just much more direct with
28:46
one another because that's how we get things done.
28:48
The persona that I have, you know, with my children is also
28:52
different. I cannot be direct with them.
28:54
They, you know, and sometimes when I am, you know, they, they
28:56
fall into tears. So I don't, I don't want that
28:58
reaction out of them. And the beautiful thing about AI
29:01
is, at least for now, AI doesn't have any emotion.
29:04
It just, it just does what you ask it to do.
29:08
And so just like as you said, just, you know, quickly get me
29:11
the, get me the order that's going to maximize the ROI.
29:16
But these boys that I have in my, in my van, the same thing
29:20
happens in enterprises. The same thing happens in
29:22
enterprises. So when we work and we build
29:25
these sort of data pipelines, these what I call extract,
29:27
transform and and load pipelines, ETL pipelines for,
29:32
you know, underwriters or estimators and logistics
29:36
companies or legal analysts, the way they work with our AI, you
29:41
know, they're happy to throw lots and lots of workload at the
29:45
AI. No problem.
29:47
Yeah, if it was a human, you'd be like, hey, this is my high
29:49
priority stuff. This other stuff, you know, can
29:52
wait. But, you know, they're just
29:54
like, hey, this AI should be able to be able to handle it.
29:56
And so they're happy to do it again and again.
29:58
They're happy to send the same thing over and over again until
30:01
it gets tweaked just right the way they want because they know
30:06
that they're actually not taxing a human.
30:09
You know, there isn't an emotional component to it.
30:11
They're just basically, you know, working with an automation
30:13
that has an incredible ability that we didn't have before.
30:17
But I think it just comes down to task and persona.
30:21
We do it and then, you know, with AI different task and
30:26
really no persona and so that that's why you act differently.
30:29
Yeah, yeah, it's interesting. You're totally right.
30:32
Because you, you, if you, you got a new hire and they're like,
30:35
boy, I really actually do need the last 10 years of line row
30:39
14, you know, column B. Exactly.
30:43
But I want them to come back on Tuesday, so I'm not going to
30:46
give them that. That's exactly right.
30:49
Yeah, right. And you know, the other way
30:52
around as well. A great, you know, so I come
30:56
from the database community, OK. And, you know, actually come
30:59
from the structured data community, you know, this is
31:01
this what's what the LLMS and and and, you know, unstructured
31:04
data is going through right now. Structured data went through in
31:07
this in the late 60s and early 70s with relational databases.
31:11
OK. And there was a, there was a
31:13
professor and you know, maybe I'm quoting the wrong person,
31:18
but there was a professor at Stanford at the time.
31:21
It was Geo Wiederholt. I never met him, but he, there
31:24
was a, there's a quote that he had, which is, you know, humans,
31:31
humans are fallible. And in enterprises, you know,
31:34
you things are so critical that sometimes you just want the
31:37
humans out of the way. And so that's why people and
31:40
that's why people are been, you know, been resorting to database
31:42
systems and why they've been more so successful because data
31:46
database systems, you know, they do what you tell them to do when
31:49
they do it accurately and they're predictable and they're
31:53
trustable. And humans, you know, while we
31:56
love them, we know that they have a wonderful, you know,
31:59
capabilities, creative capabilities and so on.
32:03
In some cases, we just want to, you know, we just want the
32:05
database to do its job. OK.
32:08
And I think, I think with, you know, unstructured data and AI,
32:12
that's exactly what's happening is that the AI is now getting
32:17
good enough where it's going to for those sort of mundane,
32:20
tedious, error prone tasks, right?
32:23
Imagine if you're spending all day and all you do is data
32:26
entry, right? It's just, it's, it's just going
32:29
to get, you know, it's going to, it's going to wear you down,
32:32
right? Or, you know, I'm going to give
32:34
you an example. Imagine, you know, your job is
32:36
to look for, you know, distressed sailboats or
32:40
distressed boats right in the harbor.
32:43
And what you see is images all day long.
32:45
And you're supposed to, you know, I try to identify one of
32:47
them. You're going to get tired of
32:49
doing it. There's going to be fatigue.
32:52
And this, and this is where I think, you know, the AI can come
32:55
in and, and, you know, really help, you know, supplement what,
32:59
what humans do. So like, I, I think AI is going
33:04
to have a completely different, you know, much more very
33:08
specific use case in enterprises where things are super critical
33:14
and where you kind of want to get the human out of the loop.
33:17
Human out of the the the critical pass.
33:20
OK, yeah, I'm not going to say out of the loop.
33:22
You want the human in the loop to make the decision.
33:24
Yeah. Hey, I found a distressed boat.
33:26
Is this a distressed boat? Should we send out a signal?
33:29
Yeah, looks like a distressed boat.
33:31
But get him out of the critical pass of looking at all these
33:33
tiles of, you know, blue images or images with, you know, you
33:37
know, small things in them that then they all of a sudden, you
33:40
know, just get tired of looking at it.
33:42
Similarly with data entry and so on and so forth.
33:44
So you get humans out of the critical path because they're
33:47
fallible. They, they're, they fatigue.
33:49
Yeah. And but you still leave them in
33:51
the loot to make the critical decisions.
33:53
And I think that's what's going to happen long term.
33:56
Yeah, it's so cool. It's like the the observation,
34:01
they become the observer in a sense.
34:03
So like they're the the the Taria of consciousness at of the
34:09
AI. That's exactly right.
34:11
You add the consciousness, you add the intuition, you add the,
34:15
you know you, you add the thing that makes you human, but you
34:18
get rid of you know you. You get rid of the mundaneness.
34:21
You get rid of the differentiated heavy lifting.
34:24
And maybe why it's, it's just surprising to me is because to
34:27
your point, like it happened really quick.
34:29
You know, we've had like 3 or 4 years where it's like, oh wow,
34:31
what's the pace is like really coming on now?
34:34
It's like every quarter, it's just like really coming on.
34:37
Whereas for like the last 10 or 15 years, we were stuck with
34:39
Siri, which was so bad. You know, where you like you,
34:43
you know, you, you do the opposite.
34:45
You talk very slow. You're like, take me to
34:48
Starbucks near my house and it's like taking you to my house,
34:51
Switzerland. You'll be there in 18 hours.
34:54
You're like, no Dang, you know, But now it's gotten so good that
34:58
like it, it really kind of invites you to play with it.
35:00
And that which ultimately invites you, your, your
35:03
creativity of like, OK, we're parsing this data in a certain
35:06
way, but like, actually, what are the capabilities?
35:09
What can you do with this data? Am I even, am I, you know, this
35:12
is another kind of fun thing that I've been doing to follow
35:17
up with AI. If, if I'm using like ALLM is
35:20
say like ask me 3 questions to make my question better, you
35:25
know, so it's, you start to invite it to be like, hey, as
35:28
opposed to being like, oh, you're so smart, Ryan, good
35:30
question. It's like, no, actually, I think
35:32
I know where you're going with that.
35:33
Let me lead you to like maybe a more sophisticated kind of
35:37
question or answer, you know, So this has just become more and
35:41
more relevant. I know.
35:43
Should I be thinking of Aaron as basically you guys are creating
35:47
like an MCP server for kind of companies that they can then
35:52
access all their data and they really are fine tuning their
35:56
knowledge base with like a whole new knowledge base for that has
36:01
been kind of locked away? That's exactly right.
36:03
So I think that MCP server is a is a sort of particular
36:07
instantiation of of what we do. You got a lot of knowledge
36:11
sitting in that unstructured data.
36:13
Yeah. You know, we, we basically, you
36:16
know, structure that knowledge for you.
36:19
Yeah. And then?
36:20
That that makes it searchable, that makes it analyzable.
36:25
And you know, you can do a lot of other sort of downstream
36:28
tasks like, you know, you know, make estimations, make
36:33
predictions, you know, you know, present, you know, present
36:39
quotes, you know, make proposals, decide on strategy,
36:45
all kinds of other things. So we basically unlock your
36:48
unstructured data for you. I guess what we would talk about
36:54
like interface a little bit because I really do like talking
36:57
to it. Unfortunately I have to be
36:59
totally alone because I'm paranoid that people would think
37:02
I'm weird when I'm talking to AI about.
37:04
I think, I think that in terms of interface, that's how people
37:07
think about AI as and, and you know, rightfully so, because the
37:11
media has kind of portrayed it that way, that AI is this thing
37:15
that you talk to. It's AI only if you know it acts
37:19
like a human and you know when you work with a human, what does
37:21
a human do? It interacts with you and
37:23
therefore it's it's only AI if I can talk to it like a human.
37:28
I actually did think that there are like.
37:30
Vast, vast. Amount of use cases where AI
37:35
blends into the background. Yeah, whenever you don't ever
37:38
talk to it. It doesn't hit you over the
37:40
head. It doesn't hit you over the
37:42
head, but what it's doing is it's making your environment
37:46
much more streamlined, OK. And, and, and you kind of see
37:51
this everyday, you know, you, you, you.
37:56
What you don't know is that this is happening already.
38:00
So I, yeah, the companies that we work with, you know, what we
38:04
do is, you know, we're basically taking away the mundane work
38:09
that a lot of the underwriters are doing or a lot of the legal
38:12
work that people are doing and what, you know, suddenly comes
38:16
into their SharePoint or their inbox.
38:19
They don't ever have to interact with an AI system.
38:22
It just shows up. OK.
38:24
Just there, yeah. There, you know, other things
38:28
that people have been doing, for example, is, you know, they'll,
38:32
they'll create documentation, OK, and they'll update that
38:36
documentation on, on the software that you've built
38:40
automatically. And, you know, you as a
38:42
developer, as a user of the documentation, don't even
38:45
realize that most of the documentation that you're
38:47
looking at was generated by AI. You're not actually interacting
38:51
with AI, but that documentation is all AI.
38:55
Humans aren't actually building a lot of the documentation
38:58
anymore because the AI has gotten so good that it can
39:01
create the documentation. And so, so documentation is 1
39:07
dashboards. You know, imagine, for example,
39:10
you know, you're, you're, you're in marketing and what you need
39:13
to know, you know, across all your headlines are the most
39:17
important things that have happened for the product that
39:19
you're trying to sell. Is there a new competitor?
39:22
Is there a, you know, a news event that's going to affect,
39:26
you know, the sale of a particular drug or the sale of
39:29
a, you know, a particular product that you have on the
39:32
shelf. All of a sudden it's going to
39:34
run out. How are you going to do that?
39:36
Well, you're going to have AI sweeping the headlines, looking
39:39
through and then highlighting the things that you should be
39:42
paying attention to, right? Guess what?
39:45
That's AI. You're not interacting with the
39:48
AI in those same way that I'm interacting with you right now.
39:51
AI is doing work on your behalf. So the right thing shows up for
39:55
you everyday. So you can then decide what
39:57
you're going to do with your product, what you're going to do
39:59
with your marketing campaign, how you're going to react.
40:02
Actually has been permeating, by the way, what we've been doing
40:05
for a long time, your news feed. Right in your social networks.
40:10
Totally. Or, you know, I like to look at
40:13
Hacker News, you know what, what shows up in your Hacker News,
40:16
you know, front page, what shows up on your, you know, on your
40:20
Instagram or you know, on your Facebook or on your LinkedIn.
40:24
This is all AI. And, you know, now, now we just
40:28
have a different type of AI that's that's doing a lot of
40:30
this work and, and it's everywhere, OK.
40:34
And, and that's the larger part of the transformation that we're
40:37
that enterprises are going to go through.
40:39
It's not just the chat box, which I think we see in a
40:42
colloquially, all kinds of enterprise, enterprise processes
40:48
are going to get replaced. And this is where the whole, you
40:51
know, agentic work flows or agentic systems people are
40:55
talking about is that instead of, you know, doing these things
40:57
manually or doing them sort of, you know, with old work flows
41:02
that were much more programmatic, you're not going
41:04
to be doing them with AI systems.
41:06
So they're going to be much, much more reactive.
41:09
You know, they're just going to be much more pliable.
41:12
They're going to be much more forgiving.
41:14
And so, so, so I think that's a transformation that people are
41:17
going undergoing in the enterprise and make that work.
41:21
And the last point I think is important here, I think the, the
41:26
thing, the place where I, I like in the enterprise, if this is
41:29
going to be successful. And This is why you, you hear
41:31
about these kinds of things like, hey, AI hasn't been
41:34
successful. And 95% of the enterprise, you
41:37
know, engagement so far is that we, we, we really need to
41:41
understand what workflows we can automate with AI.
41:45
We need to really understand, you know, how to make sure that
41:51
the results that come out of AI are high quality because in
41:56
enterprises, the workflows are much more critical.
41:59
Your business depends on it. If I tell you, you know, and I
42:03
go to online, I go to ChatGPT and I tell you, hey, man, you
42:06
know, go figure out a vacation for me.
42:08
And it tells you to go to, you know, I don't know, tells you to
42:12
go to Casablanca, right, in Morocco or something and
42:14
recommends a couple of places for you to go.
42:17
And maybe they're not the best places in Casablanca.
42:20
It's not the end of the world for you, right?
42:24
But the penalty penalty for failure is much higher.
42:27
If you are an analyst, a legal analyst and your job is to go
42:31
figure out precedent for a particular case.
42:34
If you don't find the right precedent, then all of a sudden
42:37
you know you don't. You don't have a, have a
42:39
strategy that you can, you know, you can rest your, your case on
42:43
or, you know, if you are a, an estimator and you're estimating
42:49
for construction like complex, you know, compact complex
42:54
construction for, for commercial buildings.
42:57
If you get that wrong, you got to eat the cost.
43:00
Yeah, yeah, $10 million job and you quote, you know, $7,000,000
43:06
or whatever it is, you got to eat the $3,000,000.
43:09
And so things are much, much more critical.
43:12
And that's part of the reason that we see a lot of these
43:14
failures. And it's also part of the reason
43:16
why, you know, I'm coming to you and saying, hey, there's a
43:19
different type of AI that needs to happen in the enterprise.
43:22
You've really got to focus on the use cases, figure out how
43:24
you bring the quality up. It's easy to get to 80% quality,
43:28
OK. But for a lot of enterprise use
43:30
cases, you really need to get to the 95, the 9899% level.
43:36
And that's what we do in Aaron, right?
43:37
Like we, we figure out those use cases and we figure out how to
43:40
get the last 15%, which is all of the work, by the way.
43:44
That's where all of the energy goes in.
43:47
And that's what also distinguishes us from all of the
43:49
other sort of, you know, AI startups out there and AI
43:52
projects people have in enterprises.
43:54
Because if you can't get to that last 15%, you know, I think it's
43:59
OK if you're going to be in the consumer world, but if you're in
44:02
the enterprise world, I, I, I think, you know, it's not going
44:05
to go beyond just demos. Yeah.
44:08
Well, and that's that's the hard part with AI traditionally, I
44:11
guess if there is a traditionally it's like how many
44:13
people have a a vibe coded project 85% of the way there.
44:17
That's exactly right. That's exactly right.
44:20
You know, Yeah. And I guess there's that, that's
44:23
other, that's another thing that I think we should talk about,
44:26
which is, you know, where has AI been successful?
44:30
Yeah. Where, where where do you think
44:32
AI has been successful in your in your mind?
44:35
Well, I really like it. The world around I'm you're
44:38
supposed to be asking the question.
44:42
I, I just, I'm probably my favorite application is just
44:45
making my writing more coherent and easier to read.
44:48
You know, like just the the absolute competency of like
44:51
writing minus the M-IS is obnoxious, but it's it's
44:56
fantastic. You know, I think it's really
44:58
good at that. The the new generative stuff I'm
45:00
really excited about. But I think also to your point
45:03
of staying invisible is like is really cool too.
45:06
Like it's it predictive and it knows what you need.
45:08
So if you have all this untapped context and information, you
45:14
know, and you're going to underwrite or you're going to
45:16
like check legal precedents, right?
45:17
It knows you need to find legal precedents.
45:21
So you you don't necessarily have to ask it, you know, just
45:24
say like, hey, fine, It's like, no, no, I know what you're
45:26
doing. I can read between the lines
45:28
here. And then it just like fills it
45:29
in. So you and you're seeing this a
45:30
lot with like warm start does it with like old emails that you
45:34
haven't contacted in a while. So it's like providing you with
45:37
like context to reconnect with people to all over the place,
45:41
which is really cool. Or you have ambient that's
45:43
taking your calendar as the point of truth.
45:45
And it's like, OK, you're meeting with Mehul today.
45:48
We know that this is what's important to Mehul recently.
45:51
Here's some events that have been important to him.
45:53
It just fills out the context naturally and it it's just there
45:57
and it works magically and it's just giving you context now.
46:01
I'll tell you what, I when you all those things that you did
46:04
that you described have a common thread in them, OK, The common
46:08
thread is their output is easy for you as a human to verify.
46:14
Yeah, when when you like when you when you get your your
46:19
documents rewritten and it's just clearer.
46:22
You can read it and you're like, yeah, that's what I meant.
46:24
Yeah. Yeah, Exactly.
46:26
Can you take this essay and, you know, bring it down to like a
46:29
page? Oh, yeah, Yeah, this is good.
46:31
Those are the break points. Exactly.
46:32
You got it right. It's easy for you to verify.
46:34
Or, you know, you're looking at ambient and you're looking at
46:36
your calendar and you look at the background and if it's all
46:39
gibberish, you're like, OK, got it wrong.
46:41
But if it comes in, it's like, you know, Mahole's a guy that's
46:43
worked on data for an AI for decades.
46:46
Now he's working in the unstructured data world.
46:48
I should ask him questions about unstructured data.
46:51
OK, great. It's like you can verify it very
46:53
quickly. It's done the right thing and
46:54
it's just done a good job of, of getting the key information out.
46:58
Yeah, in my opinion, that's, that's exactly where AI has been
47:02
the most successful in the last and and will be the most
47:05
successful in the last 10 years and will be in the next 10 years
47:09
is where the output of AI can be independently verified quickly.
47:14
OK, it doesn't have to be quickly.
47:16
I should take quickly back but it can be verified
47:19
independently. So just to you know some some
47:21
news stories, right? So AI has basically made it
47:25
impossible to discover protein structures.
47:28
Yeah. Right.
47:29
Well, why? Well, it, you know, turns out
47:33
that AI, you know, predicts a protein structure and the
47:36
scientists go back and then they verify that the structure is
47:38
correct using some other method. AI has been great at generating
47:44
code. Well, if actually when you look
47:47
at the code that it's generated, it actually generates a lot of
47:49
mistakes. But the people that are using
47:51
the code, it's generated enough code, but it and it is enough
47:55
correct that they can look at it and say, wow, this is pretty
47:59
good. I just need you to, you know,
48:00
make this tweak, make this tweak and you're done.
48:03
But the people that are using that code are typically
48:05
programmers that have been around for a while and they can
48:07
verify it very quickly. They're like, yeah, this is
48:09
correct or no, no, no, this is not.
48:11
Please restructure it this way. And they keep going.
48:14
Where AI is continuing to be successful is where in a humans,
48:19
when they're in the loop can verify it quickly and say, yeah,
48:22
this is what I want. Where AI still is hard OK is
48:29
where you, you know, you have to have it it it takes a long time
48:37
and takes a lot of energy to verify its output.
48:40
And so that's actually the place where I think AI still needs to
48:44
make a lot more progress. And it might take another 10
48:47
years, 15 years until we get there.
48:50
Very similar to, I would say, you know, self driving cars, you
48:54
know, five years. AI can do all these other
48:56
things. Why did it take us 15 years from
48:58
the first waymos that came out to now where they're actually
49:02
finally driving around San Francisco, coming as far South
49:06
as like San Mateo. And if you're, you know, if
49:10
you're used to the San Francisco Bay Area, to the the the airport
49:14
SFO and you know, in certain cities in Mountain View, why is
49:17
it only now that took 15 years to get there?
49:21
Well, two things. One, how do you verify that The
49:24
thing is driving correctly? OK, well, you got to have humans
49:28
in there or you got to make sure it's not killing anybody.
49:30
And two, like the cost of failure is high.
49:33
If it does it wrong, people are going to die.
49:36
And so it just took us, took us a long time to get that AI to
49:38
work. And the way they did it is they
49:40
did just tons and tons of simulations where the simulation
49:44
would verify and tell the AI, no, you did it wrong.
49:47
Try again. OK, this is what is known as
49:50
reinforcement learning. But that's, that's really what
49:51
it is, is you simulate the situation over and over again
49:56
and you know, you tell the AI, hey, you know, you got it wrong.
50:01
Try again. And so the, the, the places
50:03
where AI in all those examples that you you described are
50:08
places today where AI is just relatively easy to verify and
50:12
the places where AI is hard to verify still not out there yet.
50:16
OK. And I think that frontier of,
50:19
OK, let's get the AI to verify more and more stuff.
50:22
All right, let's get more and more things out there that that
50:25
are going to kind of make it easier for us to verify the
50:28
output of AI as that frontier increases.
50:32
So will you know the AI that looks smart and you you know is
50:36
useful in enterprises and all other contexts?
50:40
Amazing Mehul, this has been so much fun.
50:44
I I'm starting to think now that like everything is unstructured
50:48
data, like literally everything like I want AI to be reading my
50:53
wife's voice. So I'm like, hey babe, that was
50:54
the day. And she was like, good.
50:56
That it, I'm like, OK, cool, no, it wasn't good.
50:59
And I could have like AI whisper in my, in my ear, be like, hey,
51:02
man, you should probably pick her up some flowers.
51:04
She'd appreciate that today. You know what I mean?
51:06
It's like, like the tone and, and of her voice is
51:10
unstructured. The, the gestures of people's
51:12
face is unstructured. Like if you have AI structuring
51:16
all this data, it's going to be able to give recommendations to
51:18
live a happier, healthier life. I, I, I think so.
51:21
I think AI, yeah. And part part of the reason why
51:25
we've come from unstructured data is because that's where our
51:28
roots are. You know, I, my Co founder and
51:32
I, we've been working together for a dozen years and and you
51:35
know, we worked on, you know, back 15 years ago, unstructured
51:39
data was all the sort of machine data, telemetry data that was
51:42
coming out of systems feels a lot more structured now.
51:46
But back then that felt unstructured.
51:49
You know, we built systems like AWS Glue, yeah, this AWS, you
51:53
know, flagship ETL service. We built systems like Open
52:00
Search, which is search over your, your unstructured logs and
52:05
your unstructured documents. And so unstructured data has
52:08
been sort of the underpinning for AI for a very long time.
52:12
We bring a very different take on unstructured data than, you
52:16
know, traditional AI folks because we come from the
52:18
database community. And, you know, the way we think
52:21
about this is that anytime you want to do something that is,
52:25
you know, significant, it's not just going to be the model
52:29
that's going to be doing it. It's going to be the model
52:32
that's working over the data that is personal to you,
52:35
personal to your, you know, specific to your enterprise, to
52:38
your business. And what I think is going to
52:41
happen, Ryan, over the next 10 years is, and you're actually
52:45
seeing it happening right now, AI is getting faster and faster
52:49
and unprecedented rate, unprecedented rate, OK, it's
52:53
getting faster, 10X per year. That is faster than Moore's Law.
52:59
And So what that means is imagine what you can do now
53:03
today, right? So I've got some, you know, sort
53:06
of statistics off the top of my head.
53:08
And, you know, I can certainly, you know, get you more detailed
53:11
statistics at another time. But if you kind of take a look
53:14
at, you know what AI is able to do, if you, if you take a look
53:20
at like all of Shakespeare's today, it can, it can basically
53:23
read and respond to you all of Shakespeare in under 10 minutes.
53:27
OK, in a few years it's going to be under 10 seconds.
53:33
Unbelievable. In 10 years, you're going to be
53:35
able to go through all of the PDFs, OK?
53:39
All of the PDFs that are there on the web, OK.
53:43
In under an hour, yeah. Let's imagine you had access to
53:47
all the all the PDFs on the large in the largest repository
53:52
of documents today, which is SharePoint.
53:54
OK, In 10 years, you're going to be able to go through all of
53:58
those in under a day, right? So what?
54:03
And you're going to be able to do it with AIA is going to be
54:06
literally, you're going to ask a question and, and you're going
54:09
to be able to sweep through all the data in the world, not
54:12
through indices, but actually reading everything each time you
54:14
ask the question in under a day and in 20 years, maybe in under
54:18
a few hours. And that kind of power right now
54:22
is unprecedented. OK.
54:25
And, and, and really it comes from the ability for AI to
54:30
unlock the information in your unstructured data.
54:33
And that's what we help people do.
54:34
And that's the that's the ride that we're on.
54:38
Beautiful. The company is called Aaron dot
54:40
AI. That's a RYN dot AI.
54:45
I want to give a shout out to the second Harvest food Bank
54:49
that is contact or is down there in Silicon Valley and the second
54:54
Harvest Food bank collects, stores and distributes food with
54:56
one driving belief that our communities thrive when hunger
55:00
is minimized. That's the shfb.org to learn
55:04
more about them. That'll be in the show notes may
55:06
hold. This has been awesome.
55:08
What's the best place that people can connect with you
55:10
online? Feel free to just send me an
55:12
e-mail. Aaron dot AI is our is our
55:16
website and you can just click on, you know, ask for more info
55:20
and we're happy to. Also you can DM me and LinkedIn
55:23
or on X either one. Excellent.
55:26
And we'll link all that up in the show notes.
55:27
Thank you so much, Sir. Thank you.
55:30
Thank you for having me, Ryan. This episode is brought to you
55:35
by Moeburst, the AI first growth engine trusted by iconic
55:39
startups like Calm, Robin Hood, Reddit, Uber and Google.
55:44
Moeburst is a digital, full service, mobile first marketing
55:47
agency that helps companies to scale and become category
55:51
leaders. Whether you're finding product
55:53
market fit or ready to accelerate growth, Moeburst
55:56
delivers creative, data-driven solutions that help with scale
56:00
and efficiency and impact. At Moburst, we're not just
56:03
guessing what works, we're predicting it.
56:06
As an AI first agency, Moburst uses cutting edge tools to
56:10
analyse user behaviour, adapt creative in real time, and
56:14
uncover hidden growth opportunities.
56:16
The result? Marketing that's faster,
56:18
smarter, and built to convert from early stage startups to
56:22
global brands. We've helped hundreds of
56:24
companies overcome complex challenges, unlock new
56:27
audiences, and optimize every metric that matters.
56:31
At Moberst, we believe that creative thinking, advanced
56:34
technology, and data are the core drivers of success at
56:37
scale. If you're serious about scaling
56:39
your business with precision and purpose, it's time to talk to
56:43
Moberst. Visit moberst.com and let's
56:46
build something category defining together.
56:53
Hey founders, let me ask you this.
56:55
Are you building with AI or getting left behind by it?
56:59
If you're serious about scaling smarter, faster and bolder, you
57:03
need to check out AI for founders.co.
57:07
It's your go to resource for cutting through the noise and
57:10
actually using AI to grow your startup.
57:13
Each week, AI for Founders drops a sharp, no fluff newsletter,
57:18
dives deep with expert guests on the podcast, and hosts monthly
57:22
workshops where you can get your hands dirty with the real tools
57:25
and strategies. Whether you're precede or Series
57:29
B and beyond, launching something new or just staying
57:32
sharp, AI for Founders gives you the edge you need to win in 2025
57:37
and beyond. Hit AI for founders.co to
57:40
subscribe to the newsletter, tune into the podcast, and grab
57:44
your spot in the next workshop, and it's all free.
57:47
Build smarter, move faster, think bolder with AI for
57:51
Founders. Now, back to the show.
57:57
All right, all right, coffee is great, but coffee won't prep
57:59
your meetings for you. It'll just keep you awake while
58:02
you panic, hastily scrolling through LinkedIn at 6:59 AM.
58:07
Ambient, however, will prep your meetings.
58:09
Each morning you get a founder tuned briefing sourced overnight
58:13
from your calendar, your meeting history, and the web.
58:16
It's like walking to every call with a CHEAT SHEET that you
58:18
didn't have to write. No more scrambling, no more
58:21
guessing. Just context, confidence, and
58:24
focus. And if you're raising money,
58:26
Ambient investor dossiers are an absolute must have.
58:30
You'll know the investor's portfolio, their interests, and
58:33
their latest moves before you even step into the room.
58:36
You'll even get a heads up on red flags.
58:39
Think of it as the caffeine upgrade that your calendar
58:41
actually needs. Prep while you sleep, perform
58:44
while you're awake. Sign up for free at ambient dot
58:48
US. Go check it out at ambient dot
58:50
US. That's ambient dot US.