Best exit of my career. Shame nobody paid for it.
Co-founded it, majority shareholder, built it, sold it. A year later, still waiting for the money.
Some people are a good advert for why not everyone should be allowed to run a company.
A payment should make the next one more personal. That only works if the customer and the merchant both keep a record of what was bought, why, and on whose authority. Otherwise the agent starts from zero every time.
@LexSokolin@mercury@tryramp Software vs bank is the right fork. Multiples follow the data, not the balance sheet: Ramp sees every purchase and why it happened. The next test is agents spending on company cards. Whoever can show what each agent bought, and for whom, wins that.
@PhilipQiu5 It’s no surprise that merchants won’t take the friendly fraud risk by allowing a non-deterministic agent to make deterministic choices . They also kinda quite like having customers…..,
@vigstar1 The question is what true value add exists in the product and how you can charge for it. Adding agentic commerce over the top of existing rails does not make a lot of economic sense without data you can get $ for .
@OpenAI Claiming the listing is step one. The useful bit is when a Business Agent can help someone buy, and both the customer and the merchant can see what was recommended and on whose authority.
@soupsranjan This is why handing an agent a login and hoping the fraud stack copes isn't enough. The merchant needs a portable authority record: who the agent is acting for, and what it's allowed to do, sitting outside the model.
@andyfang Traction outside the tech hubs is the tell. The next question is whether the merchant can see it was an agent, and the customer can see exactly what was ordered on their behalf. That's how this gets personal, not just convenient.
@developer_dao@alexsrawitz@crossmint Whatever they buy first, the merchant still needs to see who the agent was acting for, and the customer needs a record they can trust. That's what makes the second purchase happen.
@sytaylor A card programme that can also buy stocks is still a card. The jump that matters is a record both the customer and the broker can trust, and that you can revoke.
An employee can leave without the company losing its identity. An agent should be replaceable too. Give it credentials you can revoke without rebuilding every relationship around it.
If your agent's "human approval" is an email OTP, and the agent can read email, you don't have a control. You have cosplay. Spend authority has to live outside the model.
@PhilipQiu5 The issue will be the merchant sees a Visa virtual not an Amex. So that damages Amex position with the merchant on acceptances. I.e less transactions = less incentive to pay the higher interchange.
@aa_y_ush@oost_marcel@Mastercard@OpenAI@MiebachMichael It shouldn't need a new regulator. It needs a portable authority record the customer and merchant both hold. Networks gatekeep because nobody else made spend permission portable. Fix that and the gatekeeping shrinks.
@LA_Iron_man This. A valid signature on a bad instruction is still a bad spend. Permissions and identity have to sit outside the model, or you're just automating obedience.
@sparkyfi_ai A spending limit enforced by an email the agent can read isn't a limit. It's a costume. Authority has to live outside the model or the agent can always vote itself a raise.
@ronniemgreen Right. Handing an agent a card is the easy bit. The hard bit is proving who it was acting for and what it was allowed to do. Until that's visible, it's just a faster way to create chargebacks.