Zubin: Welcome back to A to Z Fintech, part two of our
anniversary special, Double Digest. This is rewriting the
rules of agentic commerce, what we've called from the buy
button to the Machine Payments Protocol.
Aman: To the most exciting things in Zubin's life. If you've
just joined us, go back and listen to part one.
There is genuinely a thesis that runs across both of these
and the second half lands harder.
If you're with us start to finish, warning. We're now going
under the hood of the thing Stripe is betting the company
on, agentic commerce.
Zubin: Part one was the who, part two is the how,
specifically how money flows when the buyer is no longer
human.
Because as Patrick said this week, in the not-too-distant
future, agents will account for most online transactions.
Aman: And as we've said before, this is not the buzzword
version of agentic commerce. This is a fundamental rewrite
of how value moves on the internet.
The buy button, that little blue rectangle that defined 25
years of e-commerce, is about to be replaced by a protocol
you have probably never heard of.
Zubin: A year on, the architecture has actually landed,
right? The protocols have names, the wallets have launched,
the standards are being adopted. OpenAI, Google, Meta,
Microsoft, all publicly committed.
Aman: Which means part two's job is to give you the
playbook. Three deliverables for the next 30 minutes: the
Agentic Stack refreshed for 2026, the four-horsemen map of
who's playing for what, and a six, 12 and 24-month plan your
strategy team can execute.
Zubin: Obviously strap in, and I obviously need to do my
customary disclaimer. We are just two dudes riffing on this.
None of these opinions that we share on this podcast
represent anything for the companies we work in.
I'm gonna try my best, and we're gonna try to do our best to
get you out of class before the bell rings.
Aman: Okay, Zubin. Last season we built the three-layer
Agentic Stack. Discovery, execution, safety. Walk us through
what's changed in 12 months.
Zubin: I mean, it's the same three layers, but each one now
has production-grade plumbing where a year ago we were sort
of hand-waving. And layer one is discovery, the
merchant-to-agent handshake.
A year ago, an agent had to scrape your website like a Lycos
crawler. Today, there are two protocols doing the heavy
lifting. ACP, which is the Agentic Commerce Protocol, which
was co-published by Stripe and OpenAI, obviously now adopted
across the entire ecosystem.
And then UCP, which is the Universal Commerce Protocol,
which was co-authored by Google along with other industry
experts, which runs Gemini and within Google's AI modes.
Aman: Yeah, I guess the discovery problem was solved at the
standards level. The gap was adoption, right?
Zubin: That's right. That's right. The discovery piece is a
key part of it. But layer two is equally as important, which
is around the execution, the actual money movement, right?
Like, let's go back 12 months ago.
There was no real machine-to-machine payment standards.
Today we have MPP, which is the Machine Payments Protocol.
We talked about this in the prior episode, which is
co-authored by Stripe and Tempo, live since April, and
introduced a new
— and it introduced a brilliant new primitive called
Sessions. Think of it as OAuth for money.
The agent authenticates once, you authorise a spending cap,
and it streams micropayments continuously as it consumes
services.
Aman: It's the answer to the question we've just teed up.
The buy button is dead, MPP is what replaces it.
Zubin: What makes MPP special is that it's interoperable.
Stripe extended it for cards. Visa adopted it for their
Agentic Ready programme. Lightspark extended it for Bitcoin
Lightning. Coinbase has their own implementation.
MPP is genuinely becoming the TCP/IP of agent payments.
Rail-fungible underneath, protocol-shared on top.
Aman: All your best one-liners. Now, layer three, safety.
Tell us a little bit about that.
Zubin: Layer three, safety. Where the most interesting
evolution has happened. You know, if you go back a year ago,
the question was, did my human authorise this charge?
Today the real question is, is the agent itself even
legitimate? From Stripe's own data —
they shared this at Sessions as well — one in six attempted
sign-ups on AI services running on Stripe is a bad actor.
Free-trial abuse has more than doubled over the last six
months.
Aman: So the trust layer has gone from a permissioning
question to, I guess, an identity question.
Zubin: Exactly, right? And architecture answer, right? So
think of it as Verifiable Credentials for agents and
Verifiable Intent for transactions, new Radar models
distinguishing legitimate agents from fraudulent actors.
Anthropic, OpenAI and even Microsoft signing agent-identity
headers into their API calls.
You know, Visa and Mastercard, alongside the Fido Alliance,
have also sort of joined in building agentic commerce
identity standards.
Aman: You know, this is really interesting, Zubin, because
one of my jobs to do straight after this podcast recording —
my dad, as you know, passed away last year, and my mum just
got a message that, you know, his Amazon account is active,
right?
I have to go back there and figure out what the heck's going
on there. This is a real problem. And it's doubling and
tripling, as you say, all over the world. Now, now the agent
—
Zubin: Yeah.
Aman: You know, as you said, the agent is now an actor with
a reputation. That's kind of philosophically wild.
Zubin: Yeah, and it changes the entire threat model, right?
Like, the fraudster of 2026 isn't stealing your card, you
know, they're stealing your tokens.
Millions and millions of fake accounts to drain sign-up
credits, burn inference costs on free trials, rack up usage
bills they never intend to pay.
Stripe Radar's biggest upgrade this week is, explicitly,
defending against token theft.
Aman: That's incredible. So three layers, functional
plumbing on each. Protocol war isn't over, but the
standards, you'd say, are converging.
Zubin: Convergence is the operative word. A year ago, this
looked like Betamax versus VHS. Today it looks more like a
Wi-Fi standard scenario.
Multiple consortia, slightly different flavours, eventually
settling around an interoperable core.
Aman: So stack mapped. Now it comes to the strategic
question, who is playing for what? I like to call it the
four horsemen in this race, Zubin.
Stripe, OpenAI, Google, and then Visa and Mastercard, who
I'm just gonna count as one because they kind of play the
same hand. Each one is playing for a different endgame.
Zubin: Lay it out, Aman. Let's hear this one.
Aman: So horseman number one is Stripe. The endgame, be the
rails. They want to be the network underneath every agent
transaction, regardless of who issues the agent or who
fulfils the order.
Tempo is the chain, Bridge is the orchestration, Privy is
the wallet, Link is the identity. They are happy to play
with everybody. OpenAI, Google, Meta, Microsoft and
Anthropic, because it doesn't matter to them whose brain
initiates the buy, as long as the dollar flows over their
pipe.
Zubin: Dare I say, the Switzerland of agentic commerce play,
right? Like, neutral, picked by everyone, charges a small
toll on every flow.
Aman: I tell you, Zubin, you're on a roll with your
one-liners. It's a shame your wife is out of town.
Horseman number two, OpenAI. The endgame, be the brain, not
the back office. Last year, as we covered in episode 14,
they flirted with Instant Checkout on Etsy. They kind of
walked it back quite quickly.
The lesson they took, I thought, was payments are dirty
work, and they cannot afford to get distracted from the LLM
race. So they co-authored ACP with Stripe and shoved the
entire transaction layer back to the merchant.
They want to originate the buy, not settle it. Smart,
disciplined, very founder mode.
Zubin: But I'm gonna argue that it kind of leaves this huge
margin on the table, right? Like, which kind of plays
exactly with what Stripe potentially wants.
Aman: Yeah, and that brings us to horseman number three,
Google. The endgame for them is protect the search business
by extending it into agents. Google's AI Mode and the Gemini
app do native shopping powered by UCP.
JD Sports is a launch partner, and the strategy is don't let
ChatGPT steal the discovery moment. If users start asking
their AI agent to find them shoes, Google needs that moment
to happen inside Gemini, not OpenAI.
Zubin: That's right. And the genius is Google doesn't need
to own the payments, right? They just need to own the intent
of the internet, so to speak, right? And they're Stripe's
biggest beneficiary in this whole story, right?
Like, every UCP transaction settles through Stripe's Agentic
Commerce Suite under the hood.
Aman: As we both know, yeah. Yeah.
And that brings us to, I guess, horseman number four, the
collective horseman of Visa and Mastercard, or the networks.
The endgame there is don't get disintermediated.
Visa launched Agentic Ready, expanded into APAC and LatAm
this week, if I'm not mistaken. Mastercard, as we covered,
bought BVNK and launched their own stablecoin programme.
Both have signed up to MPP.
Both partner with the Fido Alliance on agent identity. Thank
—
Zubin: Yeah, but the problem is structural, all right? Like,
the card networks were built for a world where humans
entered sort of 16-digit numbers into some form fill.
In a machine-to-machine world, the card is just one of the
many possible rails, right? In this case, the Visa and
Mastercards are sort of fighting to remain one of the rails
rather than the rail.
Aman: I suppose it's a bit of strategic asymmetry. Visa
makes money on every swipe. Stripe makes money on every
outcome, you'd like. Swipe or not. Fiat or stablecoin. Card
or wallet, I guess.
Stripe is kind of agnostic to the rail. Visa, on the other
hand, is the rail. In a world where rails are commoditised,
I guess being agnostic wins.
Zubin: I don't know if you were following this weekend with
the Kentucky Derby. We did cover the four-horse race, but I
think there's a dark horse in this one as well, much like
the Kentucky Derby. What about Alibaba? What about China?
Aman: That's a really good point. Qwen the brain, Taobao the
marketplace, Alipay the rail. They processed 120 million
agent transactions in a single week last month. But most of
them, Zubin, were red-packet subsidies.
They were giving away like 10 million bubble teas to train
the models on the real intent. My kids would have been
really excited about that.
Zubin: Get out. Yeah, but you know, we've talked about this
a lot, command versus demand economy, you know, in this case
the protocol economy, would also say, right? All of these
can scale, but there's only one that can interoperate
globally.
Aman: True, Look. By the way, going back to my previous
comment, I just want to clarify, my kids are into bubble
tea, not into payments.
But we've talked about the Walls of WHAM. We framed it, I
think, last season. The protocol economy will eventually win
in the West.
The walled gardens will keep winning in China. India is a
bit of a wild card. UPI plus the new agent-payment stack is
starting to look like a third module — that sort of like
model that fuses both.
But I guess we've mapped this, so stack has been mapped.
Horsemen have been mapped. Now, Zubin, the deepest technical
move from Sessions 2026, and the one that quietly is
rewriting every business model in the next five years.
Zubin: Aman, I hate to admit, but I think it's time for a
Prof Z moment. It's been a while since we've had one of
those, right?
Because there's one technical concept from Sessions this
week nobody's talking about with the same level of awe,
which is streaming payments.
Aman: You lie! You look forward to this! I love it. Take us
through it. I'll be the smart-but-slow kid or student on the
uptake.
Zubin: Not to cold-call you. Cast your mind back on how
money actually moves today. You buy a coffee, merchant
batches the transactions at the end of the day, acquirer
settles with some sort of, what, T-plus-one, T-plus-two.
The whole system is built on this assumption that payment is
a discrete event that happens after the transaction.
Aman: Yeah. Holds for
almost everything we do. Buy a coffee, pay a salary,
subscribe to Netflix. Discrete events.
Zubin: But what happens when an AI agent burns $1,000 of
compute tokens in five minutes? Or when an agent calls
another agent's API, which calls a third agent's API, and
each costs a fraction of a cent?
Aman: Wait, batch settlement breaks. Either, I guess,
teacher, you either over-extend credit and you get burned,
or you sort of under-grant access and I guess the agent
stalls.
Zubin: You — Exactly, right? Like, the credit-card rails
simply cannot handle micro-settlement at machine speed,
which is why Stripe announced streaming payments.
It's really built on Metronome, the usage-based billing
platform that they bought in December, plus Tempo, plus
stablecoins as a unit of account. Here's the metaphor,
right?
Like, money is becoming a video stream, not a file you
download once, but a constant trickle of bits moving in real
time as value is delivered.
Aman: Kind of like Netflix instead of Blockbuster, I would
imagine.
Zubin: Like Netflix instead of Blockbuster. Or, for older
listeners, Spotify replacing the iPod. The unit of
consumption changed, same content, same value, completely
different commercial logic.
Aman: The implications are wild. If money streams in real
time, credit risk changes. Float disappears. Subscriptions
become outdated. Pay-as-you-go becomes pay-as-you-token. The
CFO's job kind of changes overnight.
Zubin: And the CFO of an AI company in 2026 is doing
something no CFO in history has had to do, reconciling
micropayments at the granularity of tokens consumed.
A single conversation with Claude or ChatGPT generates
thousands of billing events at a one-cent floor. These are
payments that cannot be processed on traditional rails.
Visa and Mastercard charge 30 cents per transaction. The
maths just breaks instantly.
Aman: And which is why I guess Stripe pushed streaming
payments onto Tempo. On stablecoin rails, on a
high-throughput chain, you could, I guess, do micropayments
at a fraction of a fraction of a cent. The economics just
work.
Zubin: Yeah, yeah, yeah. And here's the second-order effect
that ripples through every business model in the next five
years. Software-as-a-service was built for monthly
subscriptions. The next era is software-as-a-stream. You pay
for what the model just thought, in the moment it thought
it.
Aman: Zubin, you've done it. Software-as-a-stream. That's
gonna be on a billboard somewhere.
So we've covered the Agentic Stack, we've covered the four
horsemen, we've covered streaming payments. Let's do what we
always do at the end of part two and turn the camera back on
the listener.
If you're a founder, an operator, a CFO listening to this on
a treadmill in Singapore or a commute in London, what do you
actually do about all of this? Zubin, give us the playbook.
Zubin: You know, we've played this back over the last couple
of weeks, kind of culminated with this, but I would say
there are sort of three plays, right?
The six-month play, the 12-month play and the 24-month play,
at least based on what we're seeing in the market. The
six-month play, get your data hygiene right, right? This
sounds boring.
Single biggest determinant of whether your business will
exist to agents in 2027
is whether your product catalogue, inventory, pricing and
policies are exposed by clean, machine-readable APIs. Not
human-readable, machine-readable.
If your inventory still lives as a PDF and your prices
change via sort of manual updates, you don't exist to a
Gemini agent shopping on behalf of your user. You're
essentially invisible commerce.
Aman: You know, two shades of data hygiene is the new SEO.
Brands without clean APIs are the brands that essentially
won't show up when users ask the AI to find them a winter
jacket. Their existence is functionally optional.
Zubin: You know, the analogy I would give is when the
internet first came around, people would take their machine
magazines and just make them internet-ready.
This is — you can't just take your existing inventory and
just assume that the AIs would read it. You do need to make
it what now is the AEO, the new SEO, agentic engine
optimisation.
Aman: Yeah. By the way,
this is the benefit of having two dudes who've been around
the block. They can remember things from history.
Zubin: True, true, true fact. Let's play out what that
12-month play looks like. Pick your protocol stack. It
doesn't need to be on every single protocol. Pick one of
them, whether it's ACP, UCP, and ship a real implementation.
Maybe add MPP for execution, add Verifiable Intent for
safety. Cost in 2026, a small engineering team for a
quarter. Cost of not doing it in 2027
is when your competitors are getting agentic traffic and
you're not. Is your category position.
Aman: And implement it in a protocol-agnostic way. Don't bet
on a single horseman that we've outlined. Build that
abstraction layer. Zubin knows a thing or two about this in
his day job. Let the underlying rails win or lose without
breaking your own business.
Zubin: That's true. That is true. Yeah. Yeah. And keeping
church and state separately, this is exactly what we're
working with a lot of our clients as well in my day job, but
that's a different topic.
Six months discovery, 12 months really around execution. The
24-month playbook is really, reimagine your business model
for streaming. And this is a hard one, right?
Because most subscription businesses are going to face an
existential question, which is should our customers be
paying us monthly, or should they be paying us per event,
per token, or maybe even per outcome?
Aman: Yeah, and while Zubin struggles with the word
existential, most leadership teams are going to struggle
with something else and get it wrong. They will be
optimising for the current MRR instead of the future TAM.
The companies that win the streaming economy will be the
ones that voluntarily cannibalise their subscriptions before
someone else does.
Zubin: You know, my favourite story, and a lot of people
know this, my wife has been at Netflix for about 15 years.
And one of my favourite CEOs is Reed Hastings, right?
And if you kind of go back and look at the point in time,
actually, my wife was there, Reed Hastings burning
DVD-by-mail to essentially build streaming. And he did that,
right? Steve Jobs killed the iPod with the iPhone.
Andy Grove pivoting Intel out of memory into
microprocessors. The pivot's gonna be the key.
Aman: And there is a fourth play we've been — the
founder-mode play, the cultural play, you'd like.
Across every company we've covered this season, the ones who
win this transition are the CEOs who treat AI not as a
feature but as a re-platforming event.
Patrick Collison's two imperatives, reinvent your product,
reimagine how the business works.
Zubin: Which means you don't put AI in the corner of the org
chart and let it grow. You essentially put it in the centre
and let everything else reorganise around.
Aman: And you know, quite frankly, this is the difference
between every legacy bank we've covered this season and
every neobank. The neobanks are re-platforming around AI at
scale and pace.
The legacy banks are bolting AI onto their existing
platforms. The gap, I predict, is going to be widening by
the quarter.
Zubin: All right, Aman, we've covered a lot this episode,
but before the mic drop, let's just acknowledge the thing.
This is our anniversary episode. Be weird not to, right?
Yeah, yeah.
Like, 12 months ago, we were two operators with strong
opinions on this chaotic shared Google Doc. Sometimes I feel
we still are, though. But today,
Aman: Alright, go for it!
Zubin: 30 episodes shipped. Stripe, NuBank, PayPal, OpenAI's
DevDay, the protocol wars. What am I missing? The unbankable
creator economy. The most recent one with Mastercard and
BVNK. Oh my God, these are endless, right?
And when I think about this episode, we close the circle on
where the whole thing starts.
Aman: You're making me nostalgic, Zubin. And here's the
thing I genuinely did not expect.
The lessons we have taken from doing the show are roughly
identical to the lessons we have given. 30 episodes have
been, in a way, a 30-week MBA on what happens when a small
idea, taken seriously, compounds.
Zubin: And I'm gonna be — just getting started, right? This
flywheel is real. Each episode has taught us how to be
slightly better hosts. Each guest, each frame, each mic drop
makes the next one easier, right?
Our audience has grown, and I'm loving it, right? The
LinkedIn comments get smarter. The WhatsApp messages from
former colleagues get even more interesting.
And my favourite, bumping into fellow industry nerds, fellow
payment nerds, fellow fintech nerds, all talking about how
they're listening to our episodes and loving the content.
Aman: Yeah, I think for me the most rewarding part has been
hearing from listeners who are actually builders. The
founder in Lagos who messaged us about the indy worker
episode.
The product lead in Mumbai, who remains nameless but we both
worked with, who used the WHAM framework recently in a board
deck.
The CFO in Dubai in the middle of all of this chaos who
quoted the Telco Trap framework back to me while on a call.
And actually even more recently than that, I bumped into a
student in Singapore who said that episode 15, I don't even
remember what it was about, was the first time that the
history of banking made sense to her.
Zubin: And this is the founder-mode payoff, right? You make
the small things, you ship every week, you don't quit when
the analytics dip, and eventually the small things become
load-bearing for a few thousand people.
Aman: Which is, in miniature, exactly what Stripe did from
2010 to 2026. Just with several more zeros.
Zubin: And on that note, the second mic drop of the night.
Aman, come on, let's take it home properly.
Aman: Okay, here's what I've been holding out for the end.
Every payments company that came before Stripe had a
ceiling, Zubin. Visa had a ceiling. Mastercard had a
ceiling.
PayPal, as we said two episodes ago, hit its ceiling and is
now slowly being, you know, broken apart by your man, Mr
Lores.
Square hit its ceiling and is now Block. Adyen arguably hit
its ceiling and is now an excellent but bounded business in
Amsterdam.
Zubin: You know, Aman, you're onto something here, right?
Stripe doesn't have to have a ceiling.
Or rather, Stripe's ceiling is the size of the entire
economy itself, because what they have actually built
underneath all the announcements and the conferences and the
developer love is the index of global economic intent.
Every business on Stripe is a node. Every transaction is a
signal. Every API call is a data point.
Every agent is a participant, and the more agents come
online, the more useful that index becomes.
Aman: You know, Zubin, I couldn't have said it better
myself. And this is why I love doing this with you. This is
the Google move. Page and Brin built the index of the web.
The Collison brothers are building the index of the economy.
Same playbook, bigger TAM, more compounding, quieter because
money is less glamorous than search, but in the long run,
more lucrative.
And I think 100 years from now, when people write the
history of how financial services were re-platformed in the
2020s, they will not write a chapter called the rise of
payments.
They will write a chapter called the rise of economic
infrastructure. And the protagonist of that chapter will not
be a bank. It will not be a network. It will probably not be
a wallet.
It will be a company that started with seven lines of code
and ended up indexing the world's commerce the same way a
company 25 years earlier indexed the world's information.
Zubin: So what you're saying is, Stripe is not the next
Visa.
Aman: Stripe is the next Google. And the moment we, as the
fintech industry, wrap our heads around that, the next 10
years suddenly becomes a lot more sensible.
Zubin: On that, I'm out of words, which frankly never really
happens.
Aman: Anniversary mode, Zubin.
Zubin: Anniversary mode, my friend.
Aman: 12 months, 30 episodes, 128,000 words and counted of
script. A genuinely embarrassing number of unfinished
WhatsApp drafts, and the most fun two adults can have
without playing Premier League fantasy football.
Zubin: To our listeners in San Francisco, New York,
Singapore, Lagos, London, Mumbai, Sydney, São Paulo, Dubai,
and the increasing number of you who message us from places
we have to Google to confirm exist, thank you, genuinely.
Aman: And if you've stuck with us from episode one, you're
one of the reasons we keep shipping. If you've just joined,
welcome back to the back catalogue. You have a beautiful
Sunday afternoon ahead of you.
Zubin: Well, two quick favours before we go. If this episode
makes you think, hit that subscribe button. We try and drop
every Wednesday or every other Wednesday.
If it makes you laugh, share it with one person at your firm
who needs to hear it. And if it makes you angry, leave a
comment. We read every single one of them. Ping me on
LinkedIn or an email. The best ones become, usually, our
next episodes.
Aman: The Protocol War is just getting started. The agents
are just learning to spend. The Collisons are not stopping.
And Zubin and I, foolishly, are not stopping either. Year
two starts next week. Stay curious. And on this anniversary…
Zubin: Stay purposeful, my friends. Stay founder mode.