There are nearly 17 billion cards (corporate, debit, credit, prepaid) that confirm payments instantly. These are today used by businesses to subscribe to SAAS (data, online) products.
The same will be used by agents, card rails need to capture the “human intent and permission” which card schemes have / are now building - allowing AI agents to prove their identity, authorization, and purpose to merchants, distinguishing them from malicious bots.
1) Fully autonomous agentic shopping has little, near non zero demand, semi autonomous ie searching and presenting the “pay button” will work.
2) Agents don’t need new forms of money or rail. Existing ones suffices ie cards, fiat wallets suffice with with human permission / intent capture.
It's simpler to authorise your agent to use your fiat money, via the existing form factors (ACH, Cards, Fiat Wallets etc). Users consent to the agent needs to be captured to make the transition final.
vs
building a new form of money (x402, crypto etc) and then authorising your agent to use it.
@jonoringer Exactly.
If the cost of code → ~0 and the cost of distribution → ∞, the game flips.
The incumbent companies (saas etc) that already have distribution, customers, and trust still have the advantage.
Incumbents doesn’t automatically lose — but they must become AI-first fast.
Feels like agentic payments right now has way more supply than demand.
Payments usually work the other way around ie demand shows up first, then the rails get built.
PayPal worked because eBay sellers needed a way to get paid.
BNPL took off because consumers wanted credit at checkout.
You can’t really do “build it and they will come” in payments.
Crypto payments tried that for a decade.
Agentic payments today feels similar : lots of startups building, not many users actually asking for it yet.
@noyesclt Everyone’s building the supply side of agentic commerce — protocols, rails, fiat + stablecoin payments, x402, agent wallets.
But where is the demand?
@perplexity_ai feels a bit like the Uber of the AI era. The asset light pioneer.
Instead of owning the LLM fleets, it simply routes each job to the best model for the task.
Meanwhile OpenAI, Anthropic, Google and xAI are building the LLM taxi fleets. Billions in capex, infra, safety and regulation, with models increasingly converging and profits nowhere in sight.
If that trend continues, the routing layer above the models may end up being the smarter business.
@noyesclt Only Visa n Master benefited from this VC fuelled Neo-bank fad.
Given the ease of launching a neo with no entry barriers) - dozens of neo banks sprouted in each country with little differentiation.
Likely 2-3 Neo’s globally (Revolut, Wise) and 1-2 per country will flourish
@akm1410 India is in the classic "end of a credit life cycle" - It's the borrowers who are king -> loose underwriting, loose collection, 'banks via Fintechs' willing to lend tonnes of money (bank risk mitigated by "significant First Loss Default Guarantees (FLDGs)" by VC fuelled Fintechs
@andymukherjee70@ananthng@Sabya_K Unfortunately, Not that simple
Card Schemes (Visa, Master) cannot be replaced with Fast Payment like (UPI), since
Both have a different value proposition for both customers & merchants, across multiple dimensions - Economic Model, Trust, Network, Speed, Finality, Experience.
@noyesclt Great note, Tom !
ps: I wrote a note on India’s UPI, which has a gigantic fraud problem and currently the buck stops with no one !
https://t.co/G52ICIlaLa
@soupsranjan@Zelle@nytimes - Card Schemes have half a century of experience in dealing with fraud - Chargeback system works well.
- Alipay, PayPal also manage dot tackle fraud at scale.
Fast Payment Systems need to catch up: Issue is no one is willing to accept responsibility - Neither FPS nor Bank.