Can you believe Amazon created agent-to-agent payments 20 years ago? We called it machine-to-machine payments then. Here is the story:
Back in 2005, AWS was just starting to take shape. The idea was to give development centres full ownership of entire products. South Africa got EC2, Seattle got S3, and India got payments.
From Bangalore, we started building Flexible Payments Service (FPS). At its core, it let machines pay each other programmatically, long before today’s AI agents.
Incidentally, I was the first PM to join the India team, beating two of my PM batchmates by a few weeks :-)
We were solving the problems that are hot topics today. How do we adapt machine-to-machine payments into payments to merchants? Should there be a wallet? How do we embed permissions and limits into payment instructions?
Building payments so far from the US market was already a difficult choice, but we were stubborn about independence, making it worse. We even had our own data centre inside our office on Race Course Road in Bangalore. I remember a day the data centre was flooded after heavy rain, and all of us were carrying buckets to get the water out :-)
Then there was compliance. We had to build for PCI DSS, SOX, and OFAC requirements from the ground up. I remember needing to set up separate ATM-like booths where customer service representatives could access confidential information. We had to navigate US licensing requirements state by state, one by one.
The project was delayed multiple times. FPS eventually launched, but the original machine-to-machine vision never took off.
In the meantime, I moved to the US and built a simpler version for e-commerce merchants, this time working closely with the US platform team. That worked comparatively better. Our focus then shifted from developers to merchants - the direction that continues today as Amazon Pay.
I believe Amazon never realised its full potential in payments. Again and again, I watched ideas we had discussed internally become successful products elsewhere.
We were working like research scientists, drinking our own Kool-Aid, disconnected from the market.
FPS was ahead of the market, but being early is not always a good thing.
PS: The team was fantastic. Its alumni went on to build companies including Twilio, Flipkart, Juspay, BlueStone, and of course Silq. That’s a story for another day.
@ashen_one Beautiful video. My first job was in COBOL, and I’ve felt this a few times before.
Dear hand coders - what we enjoyed was not the coding, but the crafting. The creating.
We can now craft finer and faster. Nothing to feel sad about.
Beautiful video. My first job was in COBOL, and I’ve felt this a few times before.
Dear hand coders - what we enjoyed was not the coding, but the crafting. The creating.
We can now craft finer and faster. Nothing to feel sad about.
AI-written documents are destroying Amazon’s secret sauce of execution.
IMO, AI can be used for brainstorming and analysis, but not for writing.
At Silq, we stopped writing code by hand, but maybe we should keep writing memos by hand.
The usefulness of Amazon’s 6-page memo has been nuked by AI-generated text (employees sit in meetings suspecting text is AI and tune out).
Suspicious memos tagged as AI;DR (AI, didn’t read).
Agentic shopping is a mess, as agentic checkout is broken.
One option could be BNPL. Agents make the purchases, and the human pays later with one consolidated payment across all merchants, like a credit card bill.
Yes, the store has to accept BNPL for this to work. But it is easier and safer to let an agent pay through a BNPL MCP, where you set the limit and approve each purchase, than through a credit card number.
Yes, prepaid balance is another option, but most people don't like parking their cash.
Agentic shopping is a mess, as agentic checkout is broken.
One option could be BNPL. Agents make the purchases, and the human pays later with one consolidated payment across all merchants, like a credit card bill.
Yes, the store has to accept BNPL for this to work. But it is easier and safer to let an agent pay through a BNPL MCP, where you set the limit and approve each purchase, than through a credit card number.
Yes, prepaid balance is another option, but most people don't like parking their cash.
re: agentic shopping
I have NEVER in my life heard so much hype and discussion about a tech/product experience that basically does not work at all for me.
I feel like I’m using theranos
Exactly. When something fails, the customer suddenly has to understand the complexity and figure out whom to call. In my case, when a customer requests a refund, we have to work with the store leg to get the refund first, then provide it to the buyer leg. Customers can't understand such complications.
Agentic commerce is catching on, but Agentic payments remain a big issue.
The core problem: how do you let a bot pay without handing it your card number?
Two approaches are emerging. Both give the agent a number that looks like a normal card number, letting it use existing checkout forms.
1. Two-legged virtual cards (Ex: Stripe Link)
Stripe charges your card, then uses a separate virtual card to pay the store. Your payment to Stripe and Stripe’s payment to the merchant are separate transactions. That can complicate disputes, reward points and purchase protection.
2. Network tokens (Ex: Shop Pay)
Your existing card gets a stand-in number. It looks like any other card number, so it works with existing checkouts. The purchase is charged to your card, preserving the direct connection between you and the merchant. The catch is that these agent tokens are still rolling out, so they do not work with every card or bank yet.
Back in 2013 at Voonik, we used the two-legged model to build a universal shopping cart: like buying a shirt from Amazon and matching shoes from Flipkart in one click.
We saw firsthand the problems this model creates. So my bet is against two-legged models.
That said, Stripe has done many seemingly impossible things before. I wouldn’t bet against them figuring this out.
When we tried the two-leg model, we ran into issues with returns, refunds, chargebacks, etc. For example, when a customer requests a refund, we have to work with the store leg to get the refund first, then provide it to the buyer leg. Customers can't understand such complications.
Every employee now needs CEO skills.
You are the CEO of your own agent swarm.
A CEO’s primary job is to keep raising the bar, pushing the boundaries, and refusing to settle for “good enough.”
Now every employee has to do the same with their agents.
Agents often stop at the first answer that looks done. You have to keep asking: “How much further can we take this?”
My general advice for using AI for any purpose is to ask yourself: "how do I make what I am doing even more ambitious?"
You do this until you either hit a wall and then you know the shape of the frontier (useful!) or you find out that the AI can do more than you think (common!)
AI FOMO is real.
Every week brings something new. Every launch makes you feel you’re falling behind. Every new opportunity makes you question your direction.
You chase everything and finish nothing.
The people who stay focused will win, while everyone else keeps getting distracted.
My next experiment: “Inbox Development.”
What if every requirement arrived like an email?
Each engineer has an inbox. A requirement from the PM comes in. She forwards it to her coding agents.
A few hours later, she replies: “Done. Tested. Deployed.”
We got used to replying to emails within hours. What if we got used to shipping requirements just as quickly?
The goal: Inbox Zero.
We told our developers, “No more hand-written code,” six months ago.
Today, 99% of our code is written by AI.
Now Epic Games is telling its teams the same thing - and getting the same pushback.
JUST IN: Epic Games reportedly tells developers “No more hand-written code,” ordering them to use Claude to vibecode its upcoming Marvel & Star Wars game.
Since this post took off, I looked at Instinct, the AI assistant my whole family uses.
It was started roughly a year ago. Last month, it raised $1 billion at a $10 billion valuation. Team size: fewer than 20.
From what I could find, for every engineer, there is roughly one non-engineer across the product, strategy, and operations.
One person deciding what to build for every person building it.
The 1:1 team is already here.
At least, that is how it looks from the outside. Only @noahrshinn can confirm :-)
I got a lot of pushback for saying that one product manager and one engineer can now be the entire team.
Now @danshipper has a name for it: the pirate and the architect.
I ain’t crazy after all. :-)
I got a lot of pushback for saying that one product manager and one engineer can now be the entire team.
Now @danshipper has a name for it: the pirate and the architect.
I ain’t crazy after all. :-)
ALL CODE IS NOW OPEN SOURCE
Not legally. Practically.
For a long time, we assumed that compiling code and hiding the source code protected it.
AI changes this.
AI can take a binary, decompile it, and reconstruct readable source code that behaves like the original.
It may not reproduce the exact source file you wrote. But it does not need to.
All code you ship is now effectively open source.
Open source may have won without everyone agreeing to open-source their code.
This is the doom I predicted a few days ago, coming for Photoshop. A clean-room open-source reimplementation.
No prizes for guessing that they decompiled Photoshop to source code, processed that to some kind of non-code specification language, then fed the spec to an LLM with an instruction to generate Rust.
Adobe just got nuked. And closed source is dead, dead, dead.
https://t.co/0sxtGNaPtq