Machines of Buying and Selling Grace - Adam Behrens, New Generation

Channel: aiDotEngineer

Published at: 2025-07-23

YouTube video id: zlZz0mDF2eg

Source: https://www.youtube.com/watch?v=zlZz0mDF2eg

[Music]
So, as a philosopher turned engineer, I
have, for better or for worse, been
obsessed with two questions my whole
life. Uh, what is a thing and why does
it exist? So, we're here talking about
AI and the Fortune 500 and the future of
AI commerce. So, let's ask a hopefully
straightforward question. What is a
store?
A 100 years ago, a store looked like
this. Something we don't really
recognize today. Inventory was in the
back of a shop. You had to talk to a
clerk, tell them what you wanted. They
went and then fetched it and brought it
to you.
It wasn't until the 1950s and60s with
information systems that we were able to
actually scale the concept of a store
and you saw big box retailers like
Walmart and Costco emerge. The inventory
moved from the back to the front and you
now had this concept of browsing
with the internet. We took that store
and then we put it online. And so we
took the scale of merchandise and
matched it with the scale of
distribution. You can now browse
anywhere in the world 24 hours a day.
A lot of people think that websites are
dead, uh, but a shocking number of
people still shop online. Walmart had
almost 500 million shoppers last month.
Home Depot had 170 million shoppers in
its online store last month.
But the thing that we sacrificed was we
got a sea of sameness. Can you really
tell the difference here between Adidas,
Reebok, Brooks?
And so maybe we can start piecing this
together to answer this question. What
is a store?
A store is a location for and a protocol
that facilitates transactions.
You have merchants that want to sell
something, buyers that want to buy
something, and then a system to
facilitate that interaction.
So great. Now we can start asking the
question, what happens to the store with
AI?
If e-commerce digitized the merchandise
and the distribution, AI digitizes the
participants and their interactions.
We go from static websites to merchant
agents. We go from consumers browsing to
consumer agents. And we go from
low-level payment infrastructure to
higher level intent infrastructure.
But we still have the same goal that's
not changing. We're still trying to get
to a transaction.
And so what is this actually going to
look and feel like qual, you know,
qualitatively?
Uh that's something a new generation
that we spend a lot of time obsessing
over. How will a new generation of
consumer both human and agentic
interact with a new generation of
interface that is dynamic and real time
and generative built on a new generation
of infrastructure that needs to serve
those new needs.
So we see two possible futures. You'll
have either AI agents that go to
websites or you'll have agents that have
programmatic access.
Option one, when they go to websites
will look something like this. Maybe
you're in in chatbt, you want a new TV.
You you're into gaming, so you say,
"Hey, I want a new TV." Your agent goes
to a new type of website optimized for
that agent that can richly express your
intent. It dynamically cuts the product
catalog and the brand style guidelines
and returns that content directly into
the chat interface that you're that
you're operating in. when you want to
buy that agent then can go back to the
website and go through the entire
checkout process for you.
It gets interesting when you think about
programmatic access in this world. So
maybe you have the same starting place.
You're in chat, but instead of actually
going to a website, you have an MCP
server or an API that's programmatically
accessing every merchant on the
internet.
that API endpoint can then reason over
the API call and return back a rich set
of UI elements generated on the fly.
Similarly, when you're looking to buy,
instead of going to a website, it's just
going to hit an API endpoint.
And so, if that's the future, how do we
get from today, this sort of static
world of consumers browsing and websites
to this future world of agentic
interactions between buyers and sellers?
So let's start with the goal. The goal
is a transaction.
We want high quality conversion with
users that hopefully are happy.
Hopefully they they don't return the
item that they get.
And so in code, a payment is represented
with a payment uh intent when a user
clicks a buy button.
That intent then goes through a set of
transformations through the checkout
process that results in success and
money movement.
Our first challenge in this agentic
world is what if software is the one
that's clicking this button? Uh if you
use operator today, it'll mo mostly
error out on these e-commerce websites.
And so there's two solutions that people
have come up with. Uh one is the
solution that exists today. Uh this is
kind of the Stripe SDK solution. So
instead of checking out with the
merchant, you check out with ChatGBT or
the software provider. That software
provider then spins up a virtual card
and buys the item on behalf of you using
that new card. We think the more elegant
solution is what one of our partners,
Visa, is working on, which is delegated
authentication.
The agent is able to use your actual
credit card and go through the checkout
flow for you.
So that solves transactions,
but it doesn't get us very far.
What if we can move up a a level of
abstraction? What if we can actually go
to the buying and selling intent and the
preferences?
So a buyer intent today is expressed and
it's inferred via keyword searches,
click data, and site metrics.
In the agentic future, we think that
this is going to be explicitly captured
rather than inferred. Convers
conversation data is rich with user
intent and you can actually just ask a
user agent what it's trying to do. You
don't have to statistically infer it.
This gives us our second challenge.
If we have fuzzy intent, often people
are just searching for, hey, I want I
want a pair of running shoes. How do we
actually get that to the skew level
item, which is the inventory
representation of that product?
Today, the main solution is to force it
explicitly. So, you actually have to
provide a product detail page URL to an
agent in order to buy.
We think there's something very
interesting that's emerging with the
merchants that we're working with, which
is users that come from an AI channel
are much higher conversion, much higher
dollar value, and much higher lifetime
value. That opens up a whole new set of
possibilities for how to rethink the
cost structure of fulfillment.
Maybe it doesn't actually matter if the
user gets the wrong thing if it's easy
to return and they're high value.
On the seller side, that selling intent,
as I mentioned, is really represented
with a product detail page. We all know
these. They're pretty static. There's
just a buy button, a price, maybe
there's some discounts, maybe there's
some bundles.
In the future, this is going to be very
dynamic. We think merchants will need to
show real-time product availability.
They'll have contextual pricing and
discount that they can that they can
serve to a user in line and the ability
to infinitely bundle products across
multiple merchants.
This starts to get complicated. So then
we encounter our third challenge.
How do we know if a specific item is
actually available across all the
thousands of stores that exist? So we
have a user that wants to buy something
and now we actually have to go find the
store.
There's two solutions today for this
that we think are suboptimal. One is to
use existing product feed
infrastructure. So a lot of folks use
this with Google. Um that requires chat
products to individually work with every
single merchant to get this data. The
other alternative is you scrape product
data from every website in on the
internet. Uh we've done that. Uh we
think it's suboptimal. Uh it's both
repetitive. you have to do it a bunch
and then you end up clogging websites
with bot traffic.
The more elegant long-term solution that
we're working on is to actually create a
unified API to access product data
across every merchant. You can think of
this like a plaid but for product data.
So instead of aggregating over products
or instead of aggregating over financial
institutions, you can aggregate over
merchants.
And then the last piece at this layer is
how do we represent buyer and seller
preferences? Today this is one-sided and
very narrow. You have siloed user
accounts, transaction data. You have
limited LLM memory and then businesses
don't share really anything about their
intent beyond quarterly and annual
reporting.
In the future, we think this will be
two-sided and very expansive. You'll
have rich context on users not just
across the purchase but across every
aspect of their life and businesses will
be able to uh express their real time
strategic goals. What's low on
inventory? What users do they care
about? What uh strategic change has
happened with something like tariffs?
The challenges here are very wide open.
These are market design challenges. You
have the challenges that preferences are
very complex. They change over time.
They often conflict between buyers and
sellers. And there's a disincentive to
honestly report your preferences.
The current solution is uh you naively
trust the information that you're given.
Uh other folks have talked about prompt
injection or manipulation of LMS. And so
that's kind of the world that we live
in.
When I worked at Bridgewwater on our
trading system and market systems, uh
the the world of finance solved this
with third-party institutions and market
makers that manage those different
differences between uh buyers and
sellers. We think that's the world that
this needs to move to.
And so great, we've gotten a little bit
further, but nothing is really agentic
yet. You sort of need this last piece.
This is the frontier which is you need
to add intelligence to the
decision-making at each of those
components. And so consumers and
merchants need uh not just intents and
preferences, but they need intelligence
that can reason over them and negotiate
them.
And then on the infrastructure side, you
need to move from just market making to
actual coordination and reasoning over
these participants.
the logic for generating in real time
the interfaces that each of these uh
people need.
Okay, so let's see how this is working
in real life. I think a lot of people
look at the Fortune 500 and they think
oh these are big old uh slowmoving
companies.
Um but what people fail to realize is
they had to survive the last and 150
years of dramatic technological and
societal shifts. And so we find that
they're actually quite forwardthinking
in terms of this challenge.
And so if you take an example of
Samsung, they started 150 years ago as a
fish merchant in Korea with 20 or 30
employees. It wasn't until the 70s that
they started selling televisions. And it
isn't really till the 90s that they
became what we know them today, this
technological behemoth.
and they're at the forefront now of
thinking how does the brand of Samsung
evolve in the world of AI
and how are they going to bridge
e-commerce to this agentic future.
So uh the first step is we create an API
and MCP server for chat any chat client
to use.
A lot of Fortune 500 companies have
complex product systems. Uh Samsung, for
example, has 10 different verticals,
each with their own uh inventory
representations of products. And so we
do the work of abstracting that into a
consistent API with cohesive endpoints
that will work across any merchant.
The second step is then to connect that
product data with other data sources at
the company.
This is the first step in con in
constructing that seller intent.
The natural starting place with what
we're doing is let's just connect the
brand and design system that the company
has to actually wrap the products in how
they want to be represented. So we're
moving away from these carousel and
static representations of products.
That third third step then feeds into uh
container for experimentation. And so we
make an AI subdomain
that allows for rapid experimentation of
generative interfaces
that can ingest both the product and
brand data to serve customers.
And so we're also experimenting with
what does conversation look like when
it's not just a bullet point and text
list but actual images and and products
and content.
And then the last piece is handling
aentic transactions.
This is actually enabling the payment
flow to work on this new surface for bot
traffic which is a real inversion of the
typical posture of a dot website.
And the reason that brands are excited
about this is again because users from
AI chat while they might be small are
much higher intent. they're deeper in
the funnel and they convert much better.
And so we think every retail brand,
every merchant needs to adopt this
posture.
And with any big change, we think the
right approach is to start with the
question of what and how.
Uh we don't think stores go away. We
just think that they evolve. We think
they return to the original form that
they were in. And we think that form is
actually a conversation.
Thank you.
[Applause]
All right, we do have time for questions
if anybody wants to chat for a bit. Uh,
and if not, yeah, go ahead. Yes. Can you
use the microphone over here, please?
Thank you.
I'm not that tall. I don't I don't know
if you noticed that. So but uh we are
talking about machine customers maybe
just like there is a book called when
machine become customers. Yeah. Yes.
From the Gartner. So I work in a bank
and now it's it's very usable what you
said. So but what is your projection in
terms of how this is going to be just
like the break even right when it's
going to start seeing this in our daily
basis because today is only
Yeah. Yeah, it's it's happening fast. I
mean, even even if you look at a product
like chat GPT, you know, it it has it is
starting to bring shopping experiences
into the product. The thing that's
missing is you don't actually have the
full journey within within that
application yet. And so, you still link
out to a website. And that's a thing
that we're interested in exploring,
which is is will you still link out to a
website? You know, I don't think that
goes away immediately. And then every
brand that we talk to, there's a strong
desire to own a surface in this new
world. And so we think you can start
with a webl like surface that is built
in a way that's transportable so that
when these chat products want to bring
shopping into the application that you
can actually just bring that data and
components directly in, you don't
actually need to rearchitect and
rebuild. Um, and so we're really excited
about that. It's almost like this
inversion where instead of going to the
website, the website is going to go into
a thousand different places.
Thank you. You know, there is Oh, you
had a question. Sorry. Yeah. Are credit
cards the right payment mechanism for
the agentic economy or do we need
something new? I didn't want to open
that can of worms uh in the talk. Uh
I think uh conceptually
there's a strong argument for stable
coins and crypto to be the native uh
payment rail for AI mostly because the
the agents can actually live within the
wallet. Um I think practically consumers
use credit cards. Uh and so it's the
most likely bridge to get to that world.
Um and then there's a third alternative
which is the agent itself just owns a
perpetual credit card and you top it up.
Um which is is interesting. Thank you.
There's an interesting parallel in
places like uh China and Brazil where
they have the form of you know super
apps where everything happens in that
one place. Do you see claude and chat
GPT trying to become the super app and
shopping just takes place entirely over
that platform? I think I mean I think
that's the their goal. Um and then it
becomes this question of
how do the rest of us how do merchants
have have have control in that
environment and I think the I think the
one thing that model providers are very
open to and what's different than the
internet is is the goal is to get the
user to the right outcome.
Um, and so they have a an interestingly
they have a different incentive than
tech companies in the past. And so we
think there's actually kind of a like a
nice alignment between the goals of a
merchant and the goals of these chat
products. And do you foresee any form of
revenue share happening there?
Definitely. Um, I think we don't think
it's going to be advertising. um it'll
probably look a little bit more like
um either affiliate revenue or um you
know if you provide high quality data uh
and you can be attributed to a good
answer I think there'll be some portion
of that that these model providers will
will give back to merchants.
Great. Uh Adam thank you so much that
was a great talk. Thank you.