From Stripe Link, Agent Token to Merchant Chargebacks, Transaction Authorization and B2B Payments, delve into the underlying payment infrastructure of Agentic Commerce

This year's Money20/20 conference saw Agentic Commerce become one of the hottest topics.

As more and more AI Agents begin to search for products, book services, and even complete payments on behalf of consumers, the payment industry is facing a completely different type of transaction compared to traditional e-commerce:

If an AI Agent uses your credit card to purchase a product, how does the merchant confirm that the transaction was authorized? If the Agent buys the wrong item, who should be held responsible? Can the traditional credit card chargeback, dispute resolution, and risk control mechanisms be applied to transactions initiated by AI?

To gain deeper insights into these issues, I invited Lithic's Chief Product Officer (CPO) Robin Gandhi to guest in this episode of Fintechnize Leaders.

Robin was also a former colleague of mine at Adyen, and technically, he was my boss back then.

From Adyen's issuer processing system to TripActions (now Navan), Nium, and now Lithic, Robin has worked for many years in the payment infrastructure and issuing technology fields. Today, with the rise of Agentic Commerce, he and Lithic are at the core of this shift.

We discussed the technical architecture of Stripe Link and Agent Token, how Visa and Mastercard are promoting Agentic Payments, why merchants still refuse transactions initiated by AI Agents, and how to control risks through programmable credit cards.

We also discussed a noteworthy prediction: Compared to consumer shopping, corporate procurement and commercial credit cards may become the market for Agentic Commerce to scale up quicker.

The following is the full transcript of our interview, appropriately edited for readability.

1. From Adyen to Lithic: Has Agentic Commerce finally shifted from concept to reality?

Charlie: Welcome to the Fintechnize Leaders Series.

Today, I'm glad to invite an old friend Robin. We worked together at Adyen many years ago. Technically, he was my boss back then.

I still remember my first day at Adyen. The Partnerships team was very small, just you, Karo, Kalyani, and a few others, and I squeezed in as part of your team.

Years later, you've worked at several top Fintech companies. If I'm not mistaken, you were responsible for issuing products in the later stages at Adyen, then went to TripActions (now Navan), followed by Nium, and now you are the CPO of Lithic.

Lithic is now at the center of many discussions on Agentic Commerce.

Could you briefly introduce your experiences over the years?

Robin: You’ve summarized it very well; what more can I add?

At Adyen, I was responsible for building Issuer Processing, which is the issuing processing system.

After that, at Navan, then known as TripActions, we were using Stripe's issuing tech stack. At Nium, we used FIS, Fiserv, and Thredd.

I’ve been at Lithic for two years, and I think we've built something really special.

When it comes to Agentic Commerce, some of the technical architecture decisions we made earlier turn out to align perfectly with the capabilities needed by today's AI Agents.

Interestingly, when we made those decisions, we never considered Agentic Commerce as a possibility. I can assure you that we didn't envision these application scenarios at that time.

But today, these capabilities have become a significant foundation for our participation in Agentic Commerce.

Charlie: The last time we met, it was by chance at a Side Event of the Money20/20 conference in Las Vegas.

You have participated in Money20/20 many times in various roles over the years.

This year, my immediate impression is that the AI stage was prominently positioned at the main venue entrance. Almost everyone is talking about Agentic AI.

Agentic Commerce seems to be a hot term for the whole industry, much like stablecoins were two years ago.

What was your biggest takeaway from this year's Money20/20?

Robin: Can I be direct? Agentic is incredibly hot.

It's the topic everyone wants to discuss.

Of course, that makes sense. Everyone is focused on AI Agents, and with new products from Anthropic rolling out, the market has a strong sense of expectation and momentum.

Charlie: But have you seen real applications? Not demos or PR announcements, but real usage and transactions.

Robin: Yes. Agentic Commerce has indeed begun to happen.

Previously, it was mostly Agent-to-Agent scenarios, and now we are seeing some new application cases emerging.

Of course, starting from scratch, any number looks fast. Where it goes from here remains to be seen.

For a long time, we were discussing the concept of Agentic Commerce, but in reality, no one knew what it would ultimately look like.

Now we have products like Instinct, Muse, Dots, and I feel it's starting to become real.

As for whether you and I are already using it, I know I definitely am. As for whether everyone is putting their credit cards into these products, I still don’t know.

But at least now, we really are buying things through these products.

So the question becomes: how much of this is hype, and how much is genuine demand?

A year or two ago, I thought it was more hype, and the market had a strong sense of bubble.

But now, the excitement comes from the fact that Agentic Commerce seems to have finally entered a phase where it's really starting to take off.

2. Why is this time different? From OpenClaw to consumers willing to hand over their credit cards

Charlie: I often compare the current situation with when OpenClaw was trending earlier this year.

At that time, many people purchased Mac Minis to run OpenClaw. There was a brief explosion phase: Agents became smarter and more personalized, and people began to try to let them complete transactions.

But today, the technology architecture is quite different.

Products like Instinct and Dots have packaged the entire product experience effectively, making users feel safe entering their credit card information, which allows for the creation of virtual credit cards on this basis.

What do you think is the biggest change compared to earlier this year?

I'm not talking about the early Agent Payment attempts like Skyfire a year or two ago. Back then, it was more Agent-to-Agent, but now there are real consumers behind the Agents.

Robin: Let me introduce some background first.

Some may not know, but Lithic originally evolved from Privacy.com, and we still operate the Privacy product.

Simply put, Privacy provides a virtual credit card that can be linked to a user's account balance or bank account.

It naturally aligns closely with the needs of Agentic Commerce today.

When OpenClaw first emerged, we actually saw a significant increase in the usage of Privacy. Many users registered new accounts and began creating virtual cards.

However, OpenClaw was mainly targeted at tech enthusiasts.

These users are very tech-savvy and are willing to forgo many security mechanisms because they believe they can handle the associated risks.

But such products never really entered the mass market.

I'm not saying that Muse or Instinct have become mass-market products yet, but at least they are tools that anyone can use.

No matter whether you fully trust these companies, your level of trust in them is certainly higher than that of tools like OpenClaw.

In my view, that’s one of the biggest differences between today and before.

Now you are dealing with relatively trustworthy platforms.

You might still have concerns, but you may feel:

“This is a legitimate product. I can put my credit card information in it, and I can also give it my OpenTable or DoorDash login information.”

Previously, for the vast majority of ordinary consumers, this might have felt too scary.

Charlie: Even now, ChatGPT and Gemini are starting to enter the financial services field.

In the past, people would hesitate for a long time before linking their bank accounts to Plaid.

But now, many are eager to let ChatGPT help them manage personal finances.

So it really depends on each individual's sensitivity to privacy and security, and on how long you are willing to endure before handing over your most sensitive information to these platforms.

Robin: I completely agree.

Each person's risk tolerance is different.

My risk preference isn't particularly high, but at least I am willing to give my OpenTable login information to an Agent.

3. The underlying architecture of the Agentic Card: Stripe Link, Network Token, and Mandate

Charlie: Speaking of credit cards, many people have heard that Muse is collaborating with Stripe.

When users shop through Agents, Stripe Link can generate a virtual card number.

But for many, even those working in the tech industry, their understanding of the Agentic Card might stop here.

From the perspective of an industry insider, could you explain how the whole Agentic Card ecosystem operates?

Robin: I believe Stripe Link's success is due to its comprehensive solution.

There is also strong demand in the market for similar solutions to emerge.

But there is a very important component here, which is the Agent Token from the card organization network layer.

Our idea is to hope to combine the Agent Token with the Mandate, which is the authorization instruction given by the user to the Agent, for transmission.

That way, if there is a problem with the transaction, the user will have clear dispute rights.

I believe this must adhere to the existing rules and mechanisms of card organizations.

At the same time, you also need to be able to control where these cards can be used, to prevent the Agent from going out of control.

So our idea is:

You can give the Token to the Agent while also providing it with a Mandate.

If a transaction has a problem, the user still has a way to seek recourse.

But beyond that, we will also provide a more powerful rule control system.

For example, if a merchant doesn't know how to handle the Agent Token, what should be done with this transaction?

How should the transaction be processed?

Should there be a Human-in-the-loop at some stage to confirm with the user?

Or is there any other way to complete the transaction smoothly?

These are all issues we are addressing.

Charlie: I understand. So I guess many of the potential partners you are currently engaging with overlap significantly with the clients the Stripe Link team is in contact with.

Robin: I believe the market indeed hopes for a few different choices.

We are fortunate that we have invested considerable effort into the Rules Engine and Network Token architecture in the past.

Therefore, we now have the capability to engage deeply in these discussions.

Charlie: Do you think this market will become highly competitive in the future?

One analogy I think of is the stablecoin credit card market.

For example, early entrants like Rain, and those in the Asia-Pacific region such as Reap and RedotPay.

These companies don't necessarily build complete technology stacks themselves. Some provide BIN Sponsor services to each other, while others focus on different segments.

To some extent, this still follows the traditional payment industry's gameplay: some do Gateway, some do Acquirer, each handling part of the process while cooperating with one another.

Now, many small and medium-sized issuers have appeared in the stablecoin credit card space.

Do you think that Agentic Card will replicate this development path? A multitude of players entering the market to seize the opportunities that Agentic Card proliferation will bring?

Robin: In terms of infrastructure, I don’t think there will be many players.

Think about it: to get the entire system up and running, what components are needed?

First, you need a third-party Token Vault, which is a secure storage and management system for Tokens.

In this field, we mainly consider two or three large suppliers.

Then, you need issuing capabilities and the ability to create Tokens.

This field includes Lithic and a few other companies, but the numbers are limited.

So I think it's not entirely the same as the stablecoin credit card market.

However, I believe that many different Agents will appear in the future.

Agents will not just be Muse, Dots, and Instinct. Many companies will launch their own Agents.

But from the perspective of underlying infrastructure, I don't think there will be as many suppliers as in the stablecoin credit card space.

4. Why are Visa and Mastercard promoting it, yet AI Agent payments often fail?

Charlie: What do you make of the current response from card organizations?

Visa, Mastercard, and even American Express seem very proactive.

Visa introduced Visa Intelligent Commerce very early.

However, I still often hear complaints about Adoption in industry discussions.

Some are from the consumer side, some from the merchant side, and some are about issuing banks not adjusting their authorization logic promptly to accommodate the very different transaction behaviors of AI Agents versus traditional consumers.

What situations have you observed?

Is it the card organizations driving these changes, or are issuing banks pushing it? Which participants are the most proactive?

Robin: I think card organizations have invested deeply in this matter.

They absolutely want to find ways to make Agentic Payments truly work.

There are many reasons for this.

About a year and a half ago, we hadn't seen many applications at the consumer level.

Back then, it was more Agent-to-Agent, and the discussions were largely led by stablecoins.

People were asking: Can payments between Agents be achieved through stablecoins?

But now the situation has changed.

Today we are discussing a genuine cardholder who wants to purchase goods and services that you and I normally would buy through Agents.

So naturally, the card organizations will push this matter vigorously.

You are right, there are currently many issues concerning Acceptance (transaction processing).

Card organizations are gradually identifying these issues, and we can also see some problems from the perspective of issuing institutions; everyone is researching which issues can be resolved and how.

Simply put, I think they are actively addressing it.

Although today’s transaction acceptance is still imperfect, I can see that they are trying their utmost to improve it.

However, there’s an interesting question:

When the acceptance mechanism for Agentic Tokens matures and all merchants know how to recognize Agent Tokens, will merchants end up seeing this as a higher risk type of transaction?

If all merchants can recognize Agent Tokens, will they adjust the risk scoring for these transactions to be higher, considering them more dangerous than regular e-commerce transactions?

I think the answer might be yes.

But we will address that issue when the time comes.

Charlie: I also read an article you wrote this year discussing the issue of AI Agents and credit cards functioning properly, yet transactions still being blocked in the end.

You began paying attention to these barriers early on.

What is the biggest obstacle now compared to when you wrote that article?

Robin: The main issue at that time was on a CDN level, with services like Cloudflare directly blocking access from Agents.

I'm not saying these issues have completely disappeared now, but I believe that's no longer the biggest obstacle.

Especially considering that some Agents’ operational methods have changed today.

Currently, the main issue we see is Token Acceptance.

The merchant's systems were not designed to handle these types of transactions.

Agent Tokens often look just like a regular credit card number, or PAN (Primary Account Number).

But there are still some differences between them.

Card organizations are working to address the processing issues that these differences cause.

In the early days, I conducted a Proof of Concept, trying to have an Agent complete a purchase for me.

As a result, I couldn’t even donate to the library in San Mateo because the transaction got blocked earlier in the process.

But if today I hand over a card to Muse to donate to the San Mateo library, I believe it won't be blocked at that earlier step anymore.

If a transaction fails, it’s more likely because:

“You sent me a Token, but I don't know how to process this Token.”

So the earlier main obstacle has shifted.

5. Will this shopping season be a breakthrough moment for Agentic Commerce?

Charlie: As the year-end shopping season approaches, what are your expectations?

I have seen many surveys from both consumers and merchants, as well as revenue forecasts for Agentic Commerce, and the conclusions aren't entirely consistent.

On one hand, some relatively concerning numbers stand out. For instance, about two-thirds of US merchants are not yet prepared.

Does this mean that this year's shopping season will expose many problems?

Or do you think this will turn into a significant opportunity for Agentic Commerce?

Robin: I believe this will be an opportunity.

In the long run, the impact of these technologies will definitely be positive.

Moreover, I even think that the fact that two-thirds of merchants are unprepared might not be as important.

Because we actually live in a bubble of the tech industry.

I’m not sure how many ordinary consumers will suddenly start having Agents purchase large amounts of goods this Christmas.

Another point is that people generally enjoy shopping.

Especially buying gifts for others, which is inherently the meaning of the holiday.

That's also why most people won't ask their Executive Assistant to buy Christmas gifts for the whole family.

Of course, some will do that.

But generally speaking, you would want to see and touch the products yourself, at least understand the specific specifications, then decide:

“Yes, this is what I want to buy.”

Personally, I would definitely use these tools.

For example, I might tell Instinct or Muse:

“This is what my daughter likes, this is what my son likes. How old are they, what do they usually like? Help me find some suitable gifts.”

But I might not directly let the Agent place the order for me.

Of course, that doesn't mean it won't develop to that point in the future, nor does it mean that there are no people doing that right now.

But I think it’s unlikely that this shopping season will see a surprising explosion.

It’s more of an experimental phase.

People will try out these tools to see what they can do.

But I don’t believe that Agentic Commerce will take over the entire shopping process this year, nor do I think it will represent a significant proportion of transactions.

6. If an AI Agent buys the wrong item, who is responsible?

Charlie: Many discussions in the industry are now focused on authentication, risks, and dispute resolution.

I’ll give you a real example I recently encountered.

This week I wanted to schedule a doctor’s appointment.

Since my last appointment made through Muse was for my daughter, Muse automatically registered this appointment under her name.

Later I realized that this appointment was not even under my name.

However, if I want to cancel and reschedule, there might be a cancellation fee.

So here’s the question:

If I cancel the appointment, should I bear the cancellation fee?

Or is it because the Agent didn’t truly understand the Mandate I provided, which is the authorization instruction?

In the current card organization rules and various new payment agreements, these situations have not been well defined.

This new potential dispute doesn't entirely belong to the traditional credit card dispute types.

Are your users and partners discussing similar issues?

Is there a good solution for the average consumer?

Robin: I think it's still too early.

Most of the partners we currently engage with are mainly considering:

“Agentic Commerce has already started happening; how do we make it truly work?”

Rather than:

“If disputes or chargebacks occur in the future, what should we do?”

I'm not saying they completely ignore the latter, but for now, everyone is more focused on how to seize this opportunity.

As for the question of liability you asked, if you hand your credit card to an Agent without clearly explaining your needs, then to some extent, the responsibility lies with you.

It’s hard to explain otherwise.

This is why embedding the Mandate into the Agent Token would be very helpful.

Of course, it can’t solve all problems.

But if you clearly specify in the Mandate:

how much you are willing to spend on this appointment, and that the appointment should be made for yourself, not for your daughter.

Then when an issue arises, you can at least present a more persuasive argument:

“You see, the instruction I provided was exactly this.”

However, even the Mandate system isn’t currently defined that finely.

It might not be specific enough to say “this is for my appointment, not for my daughter.”

More likely, it just states:

“This is how much I am willing to spend.”

As long as the Agent doesn’t exceed that amount, you may not have particularly strong grounds to dispute the transaction.

7. Not just spending limits: What kind of credit risk control do AI Agents need?

Charlie: Another common practice is to set spending limits.

Do you think this is an effective solution? Or will it overly restrict many new uses of the Agent?

Robin: That’s exactly what we are currently doing through technology.

The credit cards we offer can be controlled along many different dimensions.

Not just the spending amount, but also when and where they can be used.

For example, if previous cards had spending amounts within a certain range, but suddenly a transaction appears that deviates by four standard deviations from historical behavior, that obviously represents a higher risk.

At that time, we can send a text message to the user asking them to confirm:

“Are you really intending to make this purchase?”

I think these are things we should do.

This is also why the technology architecture we built earlier is particularly valuable now.

On one hand, there’s the Mandate.

But as we discussed earlier, the Mandate itself cannot solve all the issues you encounter.

Ideally, additional control conditions should also be set for the credit card.

This way, you can better control when the Agent can use the credit card.

When certain transactions trigger your pre-set risk parameters, the system can ask for confirmation again.

Instead of casually handing the credit card to the Agent and completely neglecting it.

Ultimately, the responsibility will likely fall back on you.

So having these control mechanisms genuinely in effect aligns with everyone's interests.

Charlie: Besides these, what are some other common or creative control methods?

Robin: I can only discuss what we are doing. I believe other companies are also exploring many interesting methods.

For instance, we can use fuzzy matching based on the transaction description to assign different risk scores to different transactions.

We can also identify unusual behavior based on the historical spending standard deviation previously mentioned.

If a transaction originates from an IP address abroad, we can treat it as another risk factor.

I'm not trying to boast, but this indeed demonstrates the value of modern issuing infrastructure and a strong Rules Engine.

Most other issuing systems have been primarily built around real-time authorization flows.

Of course, we support real-time authorization as well.

But a robust rules engine allows you to set rules in advance, minimizing the need for ongoing intervention in each transaction by the Agent.

This is exactly the capability that Agentic Commerce requires today.

So we are fortunate that the technology we built earlier neatly fits the current market.

8. Why should merchants accept transactions from AI Agents?

Charlie: Besides the technical infrastructure, from a business perspective, what factors can drive more merchants and PSPs to accept Agentic Commerce?

Robin: From the merchants' perspective, they naturally worry about the risks of accepting Agent Tokens.

Because they are uncertain about what happens if they approve a fraudulent transaction.

This isn’t entirely a technical issue.

The entire credit card industry, especially card organizations, needs to continue to play a role.

Card organizations need to tell merchants:

“You should accept more transactions like this. We will inform you how to obtain protection and how to identify potential fraud.”

Merchants need to build this sense of security and confidence.

On the other hand, I don’t believe merchants can forever refuse this type of transaction.

I mentioned earlier that Agentic Commerce transactions during this shopping season might not see explosive growth.

However, the whole market is developing rapidly, and more transactions will certainly emerge in the future.

If you, as a merchant, proactively reject this kind of traffic, you are essentially relinquishing a potential significant revenue source.

The ongoing demand for transactions and purchases initiated by these Agentic Commerce platforms will, in turn, drive the development of the entire payment industry.

Platforms will approach card organizations and say:

“This is the capability we need.”

And the card organizations will respond.

So if we are not discussing specific credit card control technologies but rather broader commercial adoption, I think that is ultimately the logic.

9. Why might B2B and commercial credit cards explode earlier than consumer shopping?

Charlie: I'm also curious about the application of commercial credit cards.

In the Agentic Commerce scenarios you're discussing now, are they more coming from consumers or is there also a lot of demand from commercial credit cards?

Robin: I believe a significant portion of future Agentic Commerce will occur in the business domain.

Look at what the Ramp team is doing now.

B2B transactions are inherently easier to be seen as legitimate, reasonable transaction needs.

Going back to the earlier example of Christmas shopping.

When I want to buy gifts for my children, I might still prefer to see and touch those products myself.

But if it's just an office printer cartridge that needs replenishing, then there's no problem at all.

Just hand the Agent a commercial credit card and let it make the purchase.

I can tell the Agent:

“You have a budget of $5,000. As long as office supplies are running low, or if you know we need to replenish certain items, just make purchases within those parameters.”

Compared to spending my own time managing it, I prefer to let the Agent handle these tasks.

So I believe we will see more applications in the business field.

However, at present, the market is evolving too rapidly, making it difficult to make accurate judgments.

But perhaps, this will become a real opportunity.

Charlie: I am currently seeing many startups working on Agentic Travel.

You have worked at TripActions before, so you should be quite familiar with the actual needs behind these scenarios.

I think this is a very natural combination.

It's no longer the pandemic era. People are traveling frequently again, happily flying to Las Vegas for various conferences, and corporate travel has returned to normal.

So the combination of Agentic Travel and Agentic Commercial Cards is a very natural application path.

Additionally, I see companies like Visa and Ramp exploring these directions as well.

There’s even non-credit card AP/AR related to Bill.com, which is corporate accounts payable and receivable, as well as other traditional corporate financial processes.

So the demand for B2B Agentic Payments doesn't just involve credit cards but is very broad.

Robin: I completely agree.

We will definitely see more applications like this in the future.

Moreover, I think these scenarios are ones that people are more comfortable letting Agents manage.

Even though today the market's focus is mainly on Personal Agents, I believe it will eventually lead to people realizing:

It turns out these personal Agents can help us complete many procurement tasks we don’t want to spend time on but need to finish.

10. What is still missing from the Agentic Payments ecosystem?

Charlie: Looking at the entire Agentic Payments ecosystem, what do you think is still lacking?

It could be areas in the payment technology stack that are not yet完善, or product functionality, or certain key participants.

Where do you think the biggest gaps are?

Robin: I think what still needs to be resolved are issues related to transaction acceptance, dispute resolution, chargebacks, and fraud risk.

It’s not that these things don’t exist at all.

It’s just that we need to improve them more thoroughly in the next year and a half to two years, or even sooner.

If not, the entire ecosystem might become somewhat chaotic.

For example, users have completed many transactions via Agents but find they have no way to initiate disputes even though they believe they have adequate grounds.

Or, many users try to make purchases through Agents but get rejected at the merchant side.

Or users wish to convert their credit cards into Agent Tokens, but issuing banks refuse to support it.

There are many similar issues.

The industry is still in its early stages, and I don’t find it strange to encounter these problems.

But this time, the pace of development is faster than most past technological revolutions.

Everything is accelerating.

So we must resolve these issues faster than ever before.

I think this is also why the entire industry is actively investing.

11. How much real business is there in Agentic Commerce? Bull and Bear market scenarios

Charlie: Lastly, I really want to understand the real situation in the industry.

From your various communications with clients, how much of it pertains to Agentic Commerce, and how much is still ordinary issuing business?

Robin: The vast majority of our client communications still revolve around regular credit card business.

But Agentic Commerce is very interesting.

And we do see a clear market opportunity as well as the role that Lithic can play as an Issuer Processor.

So we have invested a lot of time in research and are collaborating with some potential large platforms to help them realize these capabilities.

If Agentic Commerce truly develops, it will be good for everyone.

But if you ask me how big a portion Agentic Commerce makes up of our business right now?

The answer is very, very small.

But this is exactly why we choose to work in a startup.

Because startups give us the opportunity to continuously attempt these new directions.

Charlie: How do you see the Bull Case and Bear Case for commercial adoption of Agentic Commerce in the coming year?

For instance, by this time next year, will over 30% of your client communication be centered around Agentic Commerce?

Or might it still be around 10%?

Robin: From an Issuer Processor's perspective, we mainly serve the large platforms we are all aware of.

And the number of such players is inherently limited.

So will Agentic Commerce make up a significant portion of our client communications?

Probably not.

I hope our sales team has sufficiently many client communications across other business areas, rather than having the majority focused solely on Agentic Commerce.

But I believe we will ultimately establish connections with most large platforms in this field, if we haven't already.

When it comes to actual transaction volumes and the Bull Case and Bear Case, that’s another question.

Let’s start with the Bull Case.

This is certainly the direction the entire industry hopes to see.

Having a real Personal Agent, such as products like Instinct, Muse, or Dots, is indeed very exciting and innovative.

It seems to have the opportunity to take the entire market to a new stage.

Honestly, I find myself sending many tasks to Muse every day.

But I don’t know how many people in society are using it like I am.

If things develop in line with our expectations, then this market could be very large.

Now, onto the Bear Case.

If we continue to fail to resolve dispute handling, liability attribution, and other related issues, and suddenly the market sees a large number of reports like:

“Users’ AI Agents bought a lot of things they didn’t want at all, resulting in hefty credit card bills, and users must bear the cost themselves.”

Then many people may say:

“No, I won’t let the Agent directly pay for anything anymore.”

For example, restaurant reservations.

You might tell the Agent:

“Help me find a suitable restaurant, but notify me when the reservation is ready; I won’t hand you my credit card. Just send me the booking link, and I’ll complete the reservation myself.”

If too many incidents like this occur and we have no way to address them adequately, that would feed into the Bear Case.

This is also why card organizations are currently very anxious and want to resolve these issues as quickly as possible.

Charlie: In fact, many people like me are just testing what Agents can help us purchase.

As you said, these are not normal shopping behaviors.

So if someone attempts to glean patterns from the current transaction data, they might see most of it as noise.

If risk profiles or risk control models are established based on this data, it may not be applicable in the future.

People need to gradually move out of the hype and trial phase, and start to use Agents like ordinary consumers.

Only then will the industry receive genuinely valuable transactional signals.

Robin: I believe people are starting to use these tools in their daily lives.

Just in a very ordinary way: “help me get this done.”

Of course, the demographic currently using them might still be a very particular small group.

But it indeed has potential.

Conclusion

Charlie: Thank you very much, Robin, for taking the time to talk with me today.

I really enjoyed this discussion. We've delved into the core of the Agentic Card and, to some extent, unveiled some of the mysteries of this industry.

Before we end, is there anything you would like to add?

Robin: No, I had a great time talking today.

It’s always interesting to discuss these issues with you.

We delve into what’s really happening in the industry, not just skimming the surface.

I hope this discussion also helps everyone.

I am very much looking forward to seeing how this industry develops next and what changes we can participate in and promote.

Charlie: Alright. See you at the team reunion in a few weeks!

Thanks again for your time today.

Robin: Thank you! I had a great conversation today.

免责声明：本文章仅代表作者个人观点，不代表本平台的立场和观点。本文章仅供信息分享，不构成对任何人的任何投资建议。用户与作者之间的任何争议，与本平台无关。如网页中刊载的文章或图片涉及侵权，请提供相关的权利证明和身份证明发送邮件到[email protected]，本平台相关工作人员将会进行核查。