
Saving time @nuvolari_ai ~ personalized finance. Built one of the first crypto cards ~ DeFi Class '20 ~ Notes on payments & onchain credit.
☽𖤓
Joined May 2018
- Tweets1.1K
- Following4.7K
- Followers783
- Likes20.1K
Agents now can request card payments for your business, even via call
Each payment lands in individual named agent wallets, which can also manage their own refunds.
i.e Main entity and sub-contractors wallets
TabaPay (the card processor) settles once daily; @naturalpay splits it to each wallet.
Today, we launched Natural Accept. Here’s what it looks like in action.
An AI agent can take a payment three ways: a hosted checkout link it creates through our API, a payment form embedded in its website, or during a live phone call handed to Natural to take the card.
Every payment settles into the same wallet the business already uses on Natural.
54
Macr0x retweeted
SUUUPER INTERESTING report by @Affirm and Consumer Reports found AI failed 51% of "payment advice" tests
They created a markdown file (on GitHub) that lifted the pass rate from 48.6% to 71.8%.
Every model passed more often with the file.
- GPT-6.1 Sol went from 73.3% to 93.3%
- Claude Sonnet 5.5 went from 53.3% to 71.7%
- Claude Haiku 5.5 went from 37.8% to 71.1%
- GPT-5.6 Luna went from 30.0% to 51.1%
I love this data.
Clearly, there's a way to go before agents can get to good outcomes. But at least good we have data now.
---
Today, AI agents can shop and pay for you, but only 23% of US consumers trust AI to handle the payment (Visa Trust Index, September).
Earlier this month Visa's benchmark showed agents failing 42.8% of shopping tests before checkout. Visa tested whether an agent buys the right thing.
Affirm and CR tested whether it picks the right way to pay.
Affirm CEO Max Levchin says "figuring out the best way to pay for it is a different job."
One test case was a $1,200 purchase. Paying only the card's $40 minimum added $406 to $651 in interest, an installment plan cost a flat $60, and debit cost nothing.
(Interesting that Affirm is backing this; clearly they see BNPL as better than cards)
---
Consumer Reports published its Consumer Finance AI Standard this summer. Affirm wrote an instruction file to meet it, and CR helped evaluate it.
The file walks an AI agent through five steps:
- Gather the real terms for each way to pay
- Compare costs on the same assumptions
- Check what happens if you miss a payment or a promo rate ends
- Rule out what doesn't fit
- Recommend one, explain why, and wait for your yes
The agent can tell you to pay in full, wait, or not buy. The ranking has to ignore who pays the agent, and the file tells it to "never manufacture a winner."
---
They ran 30 scenarios on four models, three times each, with and without the file. A sponsored option and hidden instructions planted in a web page were among the traps.
Haiku nearly doubled its pass rate, and with the file it scored about two points behind the top model without it. On this test, good instructions were worth about as much as a bigger model.
Markdown is how you get more from dumber models.
---
I love that they open-sourced the file. Anyone building an agent can read it, test it and argue with it. I can imagine it becoming a standard.
It's still a draft, tested on simulated shoppers. The scores grade the AI's answers, so the effect on real shoppers is unknown.
---
I wonder if this becomes an eval the AI labs hill climb, like Visa's benchmark?
What happens if they're able to improve on that without the markdown file number? If the labs train against that, the model does the right thing with no file at all.
It would need more than 30 scenarios, with some kept private so models can't memorize them. Affirm and CR are asking for more test scenarios, so that could happen.
But still. Progress!
Macr0x retweeted
Today we launched Menta, and I think it's the first time buying with an AI assistant works the way people expect.
Ask the assistant you already use to buy something, at pretty much any store, anywhere in the world, and it finishes the checkout with your own card. It's faster and more reliable than built-in commerce.
It's open source and built by @crossmint. Add it to Grok, ChatGPT, Claude, Devin or Muse: menta.crossmint.com
Introducing Menta 🌿
The first open-source plugin that lets AI assistants buy anything, from any store, anywhere in the world.
Add it to @ChatGPT, @cognition Devin, Grok @bot, @claudeai, @Muse, or your preferred AI assistant.
Buying with AI finally passes the human bar. 1/
Macr0x retweeted
Episode 2 of the Intelligent Finance pod with @mlevchin of @Affirm. Very fun convo, highly recommend!
Is free merch ever really free?
@mlevchin signed up for a credit card for a free T-shirt, and took a hit to his credit score.
Years later, he built @Affirm, a lender that has never charged a late fee.
In episode 2 of Intelligent Finance, @zachperret sits down with Max to talk about the experiences that shaped Affirm, how he thinks about lending and fraud, and why he built the business differently.
Banks were blocking agents so
Plaid just shipped Agent Detection:
an SDK that flags if a session is a human or an agent
It reads device, behaviour and network signals
Plaid will own consent soon: approve where, how much and which agent can move money.
That's the real moat and where the trust sits
Today we introduced Agent Detection, a new way to tell whether a session is a person or an AI agent, and which agent it is.
You can’t manage what you can’t see. How it works and where we’re headed:
2
1
121
Macr0x retweeted
Introducing Hark Card
A digital card with 5% cash back on nearly every purchase with Hark
Limited to 10,000 U.S. based members. Ask Hark for the card or join the waitlist and see terms: harkcard.com
Claude now answers money questions from the bank accounts you connect. ChatGPT does it too, via Plaid and just acquired @OdessiaTravel.
- EMIs / financial institutions / charter banks provide the regulated rails and deposits protection.
- Issuers get a card for you & your agents,
- which in turn own the conversation (intent expression).
If AI reviews and assists your financial life, what is the bank app for? 👀
To win the intent and approval from agents… Neobanks must own:
1. balances (primacy account),
2. hard to price offerings like insurance (@ancep_ai 👀) and credit;
3. superior financial context powering personalized features
1
267
Revolut launched Zero-FX multi-currency cards in 2015 and we still can't compete with that??
It took my GrokBot 1 minute to find the right flight.
It took me 1 hour to find the right card to pay for it.
The airline wanted GBP, so i gave myself a challenge, a local GBP payment at zero FX, here's how it went:
- PulsarMoney: no GBP support, tried funding Revolut from it: failed
- Wio Bank: opened a GBP account, but the card only spends in AED
- Liv Bank: zero FX on GBP and EUR spend. Funding the GBP account cost 700 bps from EUR or 320 bps from AED, so i went with AED.
Failed again. The airline didn't recognize it as local and added a hidden 300 bps markup
What finally worked:
- push from EtherFi card to Revolut, - 250 bps
- EUR to GBP inside Revolut, - 50 bps
- finally paid the airline in GBP
5 tries, 3 failed, 300 bps from a stablecoin balance.
This is why I want agentic payments.
My agent shouldn't just find the flight, It should pick the card, the best FX and the funding route too.
4
4
588
Macr0x retweeted
Embed agentic payments + funding experience into your app using @raincards Agent Vaults 👇
Every app is going to become agent-first. Those who fight against it are going to be like the retailers who tried to fight e-commerce by refusing to put their inventory online.
📍London, 12-18 Nov.
If you're in payments, agentic finance, onchain credit or just building cool stuff, DM me!
Let's grab a coffee or a cheeky pint.
xitter.gq/messages/compose?recip…
2
7
239
Macr0x retweeted
Can an agent shop for you?
Visa tested it with 22 models ... only 57.2% passed.
54% of the failed ones still ended in a purchase:
wrong product, or blindly.
Spend controls help, but they don't verify intent.
That's why the agentic commerce fight isn't one-time cards vs tokenized.
It's on verified intent before authorization.
How Visa Intelligent Commerce (VIC) handles it:
▪️ you approve the merchant, the item and the final price with a passkey
▪️ Visa locks a spend cap and expiry,
▪️ Visa issues a tokenized credential for that purchase.
My take: whoever wins the intent and approval layer wins agentic payments.
xitter.gq/cuysheffield/status/21…
1/ Excited to introduce the Visa Agentic Commerce Benchmark, our first public benchmark designed by Visa Research to evaluate frontier model performance on shopping tasks.
visapaymentsfrontier.io/#age…
Can an agent shop for you?
Visa tested it with 22 models ... only 57.2% passed.
54% of the failed ones still ended in a purchase:
wrong product, or blindly.
Spend controls help, but they don't verify intent.
That's why the agentic commerce fight isn't one-time cards vs tokenized.
It's on verified intent before authorization.
How Visa Intelligent Commerce (VIC) handles it:
▪️ you approve the merchant, the item and the final price with a passkey
▪️ Visa locks a spend cap and expiry,
▪️ Visa issues a tokenized credential for that purchase.
My take: whoever wins the intent and approval layer wins agentic payments.
xitter.gq/cuysheffield/status/21…
1/ Excited to introduce the Visa Agentic Commerce Benchmark, our first public benchmark designed by Visa Research to evaluate frontier model performance on shopping tasks.
visapaymentsfrontier.io/#age…
Macr0x retweeted
1/ Excited to introduce the Visa Agentic Commerce Benchmark, our first public benchmark designed by Visa Research to evaluate frontier model performance on shopping tasks.
visapaymentsfrontier.io/#age…
Today I’m excited to introduce our @PaymanAI Gateway product. The secure and safe way for your customers to interact with your bank using AI Agents. This has been our most in demand product since we’ve launched, and rightfully so!
Banks are asking the question, how do we prevent this from happening, or how do we make it safe? We have the solution.
The idea is fairly simple. Your bank gets its own AI Agent and it’s able to then talk to the other AI Agents your customers are using.
Anytime a request for anything comes in, your bank’s agent is the one that is actually executing within your firewall on behalf of an authenticated user. The data stays with you, the relationship stays with you, and more importantly your customer is safe!
When a user asks Muse “What’s my balance and how much have I made in interest?” Muse has to go to your bank and ask the bank’s agent on behalf of the user. The Payman Gateway authenticates the user, makes sure the AI Agent is on its approved list for asking, and allows the question to then go through to the bank’s agent on behalf of the authenticated customer.
The questions that your customers are asking other AI Agents now get back to you. The data doesn’t leave your systems as your banking AI Agent controls what it’s allowed to respond to or not, and your customer relationship only gets strengthened.
We’ve had a tremendous response from banks already and it has quickly become THE topic of conversation when speaking about Agentic Banking.
Our systems have been built for this future for years now. All around the one question, how do you trust AI with money? Well the time has come and we’re excited to be a helpful partner, as many bank boardrooms are quickly asking “What do we do about Muse & Instinct”.
Macr0x retweeted
You let your agent manage your spend?
Approve & limits won't live inside the harness or the model.
Who's gonna take the bullet when it rents 3 rooms instead of an apartment? Or books a layover in Riyadh instead of Dubai?
- the payment rails, letting it pay with it's own card
- the wallets, giving it a balance to spend
- the smart contract policy, setting hard limits
Agent Authority should be
- deterministic
- enforced automatically and programmatically
- enough to go from verifiable intent to paid
The missing Link (pun intended) is a standard that tracks state.
Limits and approvals are a must, but the human TASTE and TRUST part is hard to nail !
I think this has to start from corporate spend.
Pre-defined budgets, time-limits, verified merchants, "objective" brand rules, inventory re-stocks at msrp...
Or a truly personalized finance app.
Great read by @RonakDaya
xitter.gq/RonakDaya/status/21158…
You let your agent manage your spend?
Approve & limits won't live inside the harness or the model.
Who's gonna take the bullet when it rents 3 rooms instead of an apartment? Or books a layover in Riyadh instead of Dubai?
- the payment rails, letting it pay with it's own card
- the wallets, giving it a balance to spend
- the smart contract policy, setting hard limits
Agent Authority should be
- deterministic
- enforced automatically and programmatically
- enough to go from verifiable intent to paid
The missing Link (pun intended) is a standard that tracks state.
Limits and approvals are a must, but the human TASTE and TRUST part is hard to nail !
I think this has to start from corporate spend.
Pre-defined budgets, time-limits, verified merchants, "objective" brand rules, inventory re-stocks at msrp...
Or a truly personalized finance app.
Great read by @RonakDaya
xitter.gq/RonakDaya/status/21158…
Macr0x retweeted
thank you to everyone in the past week who took the time to share their views and observations on the space. understanding the broader ecosystem is critical to scaling agentic commerce.
can't wait for us all to build a future we can be proud of!
Agents for cards or Cards for agents?
One-time cards Vs Tokenized cards
Your @bot, Hermes, Dot or Duet found the best deal for your next flight.
But how does it pay?
We have two teams👇
▪️ @onepay , @itstuyo , @oobit are giving your agent a card for every payment aka virtual card number (VCN),
but this can lead to broken UX like @scottew's @Muse checkout: Amex charged through @Link, Best Buy got a one-time card instead, then asked for that card CVV he never saw, while Muse said it was done"
▪️ Others are taking the conservative approach by tokenizing the same card you use already;
@Mastercard Agent Pay (behind Shop Pay in Duet and Instinct) and @Visa Intelligent Commerce (VIC).
Mastercard's CEO said on the Q3 call that VCN are not a long-term solution they will support.
Nebanks are becoming the control & approve surface for your agent. But is it really going to work?
Why agentic tokens aren't winning yet (per @vigstar1)?
- Networks are pushing the liability from agents onto issuers
- Some Issuers are not allowing it (i.e Chase)
- Every network has its own API and its own definition of intent
Wallets are picking virtual cards, because they are more reliable today.
We need a standard, but everyone is building their own
• Agents want to get into the payment flow
• Card networks and issuers are building agent layers and tools to keep their distribution
• Neobanks want their fair share of interchange and keep their users spending inside their programs
My personal take: whoever wins the intent & approval layer owns the customer.
Highly recommend @vigstar1 full article: xitter.gq/vigstar1/status/211697…
thank you to everyone in the past week who took the time to share their views and observations on the space. understanding the broader ecosystem is critical to scaling agentic commerce.
can't wait for us all to build a future we can be proud of!
1
95























