
there’s been some great discourse over the past week on @OpenAI’s ChatGPT Wallet. i’ve kept this one short(er) and focused on the core issue with agentic commerce: misaligned incentives. as always, please share your thoughts and disagree with me!
✓agentic commerce + payments @meta all opinions are my own

there’s been some great discourse over the past week on @OpenAI’s ChatGPT Wallet. i’ve kept this one short(er) and focused on the core issue with agentic commerce: misaligned incentives. as always, please share your thoughts and disagree with me!

The future of commerce is an AI agent that doesn't just answer questions, but closes sales. Excited to launch @Razorpay’s partnership with @Meta to make this real on @WhatsApp business agent! AI agents can now recommend products, generate payment links and complete sales, 24/7. All within a conversation.


yes, i wouldn't be surprised if ChatGPT Wallet falls back to PAN injections, especially when @instinct, @bot and @hark_labs do it already! but "you don't have to ask permission from anyone" is exactly the problem. openai launched this in deep partnership with visa, mastercard, amex, and chase. payments is a partnership business with a long memory, and quietly routing around the partners you just launched with won't align with "Building trusted agentic payments with networks, banks, and businesses." but let's entertain the fallback scenario anyway (its openai, who gon stop em?): - issuers and networks go blind and lose control. - one more juggernaut favors stealth checkout. several agents already autofill raw card details from vaults. add chatgpt, and most agent volume flows where the networks can't see it's an agent delegated purchase. - tokens and rules don't mature on their own. they mature on volume and edge cases. if the volume goes to the fallback, issuers have less reason to support tokens, tokens stay unreliable, and more agents fall back. it isn't really vision vs. execution. the real question is what's the agent company's incentive to ever leave the fallback? for openai, it's probably long-term leverage. the token path is the only way to prove to the ecosytem that agent volume is lower-risk, and that's the only argument that ever gets issuers to MAYBE share economics. falling back trades that away for reliability today. perhaps they receive a merchant take rate, but without network buy-in, the path to resolving disputes for merchants (which will get worse with more volume) is even harder. for smaller agent companies, they care most about user experience, reliability and a path to monetization. today, tokens offer none of the three. vault + PAN ships fast and works at more checkouts, and stored-value balances/VCC issuance even share economics. if even openai falls back, no smaller agent has a reason to choose tokens. until the token path is the better business for agent companies, not just the right one, agent volume will keep flowing around the networks instead of through them. and we are back to square one.

i’m getting dms from many folks at the networks that believe that a lot of the issues i mention are on track to being resolved sooner than i’d think! that’s so great to hear, and i’m all for the networks to increase the reliability of agentic tokens. network tokenization is one of the most critical payment innovations of the past decade, and it should be the end state of agentic payments for reducing fraud rates, increasing auth rates, and, of course, distributing intent. but it will be a tall task! it will require alignment across competing networks, issuers, acquirers, merchants, and agent companies. and no: protocols, frameworks, and playbooks won’t work!

yes!! y'all already know i have some strong opinions on debit especially in the US (fun fact: it's the same durbin-exempt lead bank issued prepaid debit card as link balance). feels like there's a new product every day, but i promise i will share my thoughts on it as well :) already behind on publishing my chatgpt wallet article (i just dozed off after date night!)


my girlfriend insists we do date night and record a fun tiktok, so my article won’t be out until later tonight! my girlfriend (and her parents) are concerned i love agentic payments a bit too much, so gotta do what i gotta do!!

and it promises to be much shorter and more digestible based on y’all’s feedback 🫶 goodnight!

this is the exact framing we need to use to judge any payments product. i finally got some time to write this evening and stayed up to finish my first draft! stay tuned for my next article dropping monday, nov 16th: CHATGPT WALLET IS THE RIGHT BET ON THE WRONG TIMELINE

great read, would highly recommend builders on the agent and merchant side to check this analysis out!

@rooqster @raincards i’m sure you would agree we need high speed stablecoin rails too 😂🫡

read this and immediately thought this referred to stablecoin rails. am i cooked?

“Solving payments is the most critical and difficult piece for agentic commerce to scale. Not because making a payment is hard. Getting an agent to pay is the easy part. […] So when I say ‘payments,’ I mean balancing everyone’s economic incentives so that they benefit from legitimate, profitable payment volume. The real question is how we create the ‘chaordic’ conditions it desperately needs to be sustainable.” quoting the most important part of my recent article since a lot of twitter seems to think chatgpt wallet has solved agentic commerce. i’m sure @strongriley and the @OpenAI team would themselves agree that they haven’t! @eric_seufert thanks for always being based!

not true. most agentic network tokens have clear agent identifiers

+1. this will NEVER work for consumer payments in the US. US commerce runs on credit and incentives are so aligned across the ecosystem that you can’t change it.