{ad_logo_primary}

Quick favor: Our records indicate that you aren’t opening this email. But records can be wrong. Please click here if you’d like to remain subscribed to Fintech Takes. 

Happy Monday, Fintech Takers!

I hope you had a lovely weekend. 

Mine went from cold and snowy to sunny and 65 degrees. Spring in Montana, the definition of fickleness.

Before we get into it, a quick correction. Last Friday’s newsletter came out with the subject line “PMF 3.0,” even though the title of the piece itself was “PFM 3.0.” 

Now, I could try to convince you that this wasn’t a mistake, but rather a clever and intentional inversion, given that Personal Financial Management (PFM) tools often struggle because the early Product-Market Fit (PMF) that they achieve is rarely indicative of their long-term viability as venture-backed businesses.

But that wouldn’t be true. I simply fat-fingered it and failed to notice before sending the newsletter out to all of you. 

If you noticed the error, my apologies. 

If you didn’t, I completely understand 🙂 

- Alex

P.S. — It’s not too late to join us on Thursday for our virtual event. Our focus is on defining a lending strategy amid a period of extreme uncertainty. And the fun part is that we’re making it an open AMA-style event, so register now and bring your questions (or submit them in advance!) 

Was this email forwarded to you?

Design for Elevation of the Duchess of Newcastle's Bedroom, Hôtel Hope by Jules Lachaise and Eugène Pierre Gourdet.

3 FINTECH NEWS STORIES

American Express is throwing its hat into the agentic commerce ring:

American Express is introducing an agentic commerce developer kit and announcing Amex Agent Purchase Protection, an industry-first commitment to extend its backing to Card Member purchases made by registered AI agents across its network.

A simplified model of e-commerce is:

1. A human finds something they want to buy online.
2. The human gives the merchant money (often via a secure payment card credential) for the good or service they want to buy.
3. The merchant gives the human the good or service.
4. [Optional] If the human doesn’t receive the good or service, is unsatisfied with it, or just doesn’t want to pay for it, they can initiate a dispute with the merchant. 

Over the last couple of decades, we have built a lot of systems and a lot of rules to facilitate this process in the most streamlined way possible. When new technology comes along that can help make it even better, the card networks leverage their rules to incentivize adoption. 

A good example is 3-D Secure.

3-D Secure (3DS) is an additional authentication layer used in online card payments that lets the cardholder’s bank verify the buyer’s identity before approving a transaction (step #2 in our process flow above), often via a one-time code or app-based approval. In exchange for routing transactions through this flow and sharing more data upfront, merchants can qualify for a liability shift, in which the issuing bank (rather than the merchant) is responsible for disputes related to unauthorized card use.

The playbook is straightforward: introduce a new capability, attach liability to it, and use that to force a change in behavior among network participants.

Agentic commerce is, essentially, e-commerce disintermediated by AI agents. The process flow looks, theoretically, something like this:

1. A human instructs an AI agent to find something they want to buy online. They give the AI agent specific parameters for what they want (type, quantity, quality, price, etc.)
2. [Optional] The AI agent brings a set of options back to the human and works with them to select the product or service they want.
3. The AI agent gives the merchant money for the product or service that it is buying for its human.
4. The merchant gives the human (and/or its AI agent) the good or service.
5. [Optional] If the human doesn’t receive the good or service, is unsatisfied with it, or just doesn’t want to pay for it, they can initiate a dispute with the merchant (or instruct their AI agent to do so).

This new process is rife with potential problems, including:

1. How do the issuer and the merchant know if the AI agent involved in the transaction has been authorized by the cardholder?
2. How do they know if the authorized AI agent correctly understood the intent of the cardholder (i.e., what to buy)?
3. How do they know if the authorized AI agent, acting on behalf of the cardholder’s clearly stated intentions, took the correct actions?

American Express is leveraging its three-party model (it acts as both the card issuer and the merchant acquirer, thus creating a closed-loop) to launch its Agentic Commerce Experiences (ACE) Developer Kit. The purpose of ACE is to control for the first two questions above (Did the cardholder authorize this agent? Did the agent correctly understand the cardholder’s intent?) in order to isolate the third question (Did the agent take the correct actions?) and to protect consumers and merchants from losses relating to it.

Amex is still building out parts of ACE, but from what I can tell, here’s how it’s intended to work:

1. The provider of the AI agent enrolls in the ACE developer program, integrates Amex’s tokenized payment flows, and uses an agent identity standard, so that requests can be attributed to a specific, trusted bot/service.
2. The cardholder registers their card with a verified agent and links the account so the agent can access Amex-backed payment and membership features.
3. The agent sends Amex the cardholder’s authenticated purchase intent (in a structured format) so that Amex can explicitly confirm this intent with the cardholder (via the Amex mobile app) and authenticate and authorize the transaction.
4. The agent pays with tokenized credentials linked to the cardholder’s account.
5. The merchant passes back cart-level details so that Amex can use the data to confirm that it matches the cardholder’s intent and to assist in dispute investigations.

If all of those steps happen and the AI agent still somehow screws up the cardholder’s transaction, Amex (as the issuer) assumes the liability.

This is interesting to me for a few different reasons.

First, I’d imagine that if Amex is going to sign up for any amount of liability (a first in the world of agentic commerce, from what I can tell), it is going to have a very rigorous vetting process for companies that want to have their AI agents participate in ACE.

Second, it’s unclear to me how interested cardholders will be in registering their cards with verified agents or in confirming (for every transaction) their intent in the Amex mobile app. Seems a bit onerous!

Third, and most importantly, Amex needs merchants to share more data (what it calls “Cart Context”) so that it can confirm whether or not the agent-initiated transaction is aligned with the cardholder’s verified intent. Without this, there’s no way for Amex to know whether the agent made a mistake (in which case Amex is responsible) or whether it was the merchant or the cardholder who did.

But how will Amex convince merchants to do this, when, as a rule, merchants don’t like sharing more data than is absolutely necessary? 

My guess is that eventually (if ACE takes off) Amex will leverage the same tool that the card networks always leverage when they need merchants to do something they don’t want to do: A liability shift. It could be positive (reduce your liability if you do this) or negative (we will increase your liability if you don’t do this), but I can’t see how this doesn’t happen at some point. 

Speaking of American Express, it just made an acquisition:

American Express today announced that it has entered into an agreement to acquire Hypercard (Hyper), an agentic expense management company, adding to Amex’s AI expertise and capabilities across its commercial services business. Hyper’s team of AI experts will help American Express continue to build agentic tools and AI-powered solutions that help businesses automate processes and simplify operations.

Founded in 2022, Hyper has focused on transforming expense management from a manual process into more autonomous workflows, successfully developing native AI agents that auto-categorize and file expenses, check them against budget and policy and send reminders that submissions are due. In 2024, American Express and Hyper partnered to launch the Hypercard Rewards American Express card with embedded AI-powered expense agents leveraging the Agile Partner Platform. Since then, Hyper has continued to focus and refine its agentic expense management capabilities.

This acquisition comes on the heels of Amex’s acquisition of Center, a different corporate card and expense management provider, last year.

There are two ways to look at the Center and Hypercard acquisitions. 

The optimistic view imagines the internal deliberations happening at Amex as something similar to Moneyball, with the team trying to figure out how they can compete now that Capital One has acquired Brex:

Amex GM: Is there another expense management platform like Brex?

Amex Scouts: Not really.

Amex GM: And if there was, could we afford it?

Amex Scouts: Nope.

Amex GM: Then what the fuck are you talking about man? If we try to play like Capital One in here, we will lose to Capital One out there.

Amex Scouts: [Silence and Blank Stares]

Amex GM: We can’t acquire Brex. We can’t do it. What we might be able to do is re-create it. Re-create it in the aggregate. 

This is a fun hypothetical to imagine, but there are a couple of problems with it.

First, and most importantly, American Express isn’t the Oakland A’s in this analogy. It’s the New York Yankees. 

For decades, Amex owned the corporate card and expense management market, especially in travel-and-entertainment (T&E) spend, where its closed-loop model gave finance teams richer spend data, more standardized reporting, and tighter program administration.

When Concur came along in the early 1990s and started to grow into a new center of gravity within the T&E space, Amex chose to partner with the company rather than acquire it, instead allowing SAP to acquire it in 2014, after which Concur fell into obsolescence. This created a gap in the market that Navan (2015), Brex (2017), and Ramp (2019) all rushed in to fill. Rather than taking this second wave of disruption seriously and acquiring one of those companies, Amex waited for years before finally acquiring a much smaller and less valuable competitor (Center), which was co-founded by the son of one of Concur’s co-founders.

It’s worth being explicit about the timing here, because it’s kind of insane. Capital One announced its $35 billion acquisition of Discover in February 2024, giving it the same closed-loop network advantages that Amex has enjoyed for decades. After the Discover acquisition was approved in April of 2025, Capital One shifted its focus to bringing corporate card spend onto its new network by plunking down $5 billion for Brex in January 2026. 

In between those two events, American Express — which, to be clear, is significantly more profitable than Capital One — spent considerably less (presumably) to acquire Center. And now, seemingly because Center (unlike Brex and Ramp) had no meaningful AI capabilities to speak of, Amex is acquiring a different small corporate card and expense management provider that was already issuing a corporate card on its network and that just so happens to have Sam Altman on its cap table (he kicked in $2.5 million at some point) for an undisclosed (likely small) amount.          

Mechanically, what Amex is doing looks like Moneyball. However, financially, Amex is much closer to the Yankees than it is to the A’s, which makes me doubt the soundness of the underlying strategy.

Public launched AI agents for investing: 

Public … today began rolling out Agents, a new innovation that allows investors to automate their portfolio strategies with AI. Now, investors on Public can build Agents that actively monitor the markets and execute trades based on users' specific instructions.

For decades, investing meant manually entering orders—whether by calling a broker, placing a trade on the web, or tapping a mobile app. Agents on Public change that. Instead of monitoring the markets and executing trades, investors can simply describe what they want to do, shifting the experience from manual clicks to expressing intent. Agents on Public monitor conditions in real time and execute investors' strategies exactly as defined.

Agents can support a wide range of portfolio workflows, including trading strategies, cash management, and risk management. Before taking action, the AI asks follow-up questions to help map out a complete Agent workflow. The investor can refine timing, adjust triggers, and set precise conditions, all through follow-up prompts. When the logic looks right, the investor activates the Agent, and it goes to work.

When we apply AI agents to a task in (or adjacent to) financial services, the question we need to ask ourselves is this: Will removing human psychology from this workflow while retaining human intelligence lead to better outcomes, over the long term, for our customers?

With e-commerce, it really depends. Some human psychological traits, like our tendency to be deceived by charm pricing, are unhelpful and we could benefit by having AI abstract away the details of the transaction. Other traits, such as our tendency to add items to a wishlist or shopping cart so that we can consider the purchase rather than buying them with one click, are beneficial and should not be eliminated through AI (no matter how much retailers and payment providers want them eliminated).

In investing and wealth management, I think the case for agentic AI is unambiguously strong.

The most effective investing strategies all optimize for psychology first, and intelligence second. 

Take dollar-cost averaging (DCA) as an example.

DCA is the practice of investing a fixed amount at regular intervals, regardless of market conditions. Its power comes from removing decision-making entirely. Instead of asking “Is now a good time to invest?”The investor simply follows a hard and fast rule. 

DCA works less because it’s perfectly optimized, and more because it eliminates the behavioral mistakes — fear, greed, hesitation — that typically destroy returns. Its advantage is psychological, not informational.

The tradeoff is rigidity. By design, DCA ignores short-term opportunities — volatility, pricing dislocations, timing within a window — because incorporating them would require human judgment, and, historically, it has been impossible to introduce human judgment into the investing process without also introducing the biases and behavioral mistakes that come with it.

AI can eliminate this tradeoff. It can enable investors to be active without being in the market.

An agent can preserve the structure of DCA — automatic contributions, strict rules, no human intervention — while making small, bounded optimizations at the margins. For example, an agent could execute each scheduled contribution within a fixed 24–48 hour window — buying on minor dips if they occur, or at the deadline if not — preserving consistent DCA while modestly improving entry prices without drifting into true market timing.

The value proposition makes a ton of sense, which, in turn, makes me wonder why Public is (as far as I know) the only wealth/investing-focused fintech company to offer it.

The answer (I think) is the business model. If you make most of your money on fees and subscriptions, you won’t mind if your users are making better, less emotional trades using AI. By contrast, if you make a large chunk of your revenue by monetizing trade flow (via payment for order flow in securities and spreads in crypto), you might be incentivized to keep humans (and their emotional decisions) firmly in the loop.

On a completely unrelated note, here’s a quote from an interview that Robinhood CEO Vlad Tenev gave last year:

Most of the time you're not doing it just because you want to make money. You also love trading and you're extremely passionate about it. … I don't think there's going to be a future where AI just does all of your thinking, all of your financial planning, all the strategizing for you.  

2 READING RECOMMENDATIONS

Jev goes deeper on what Block is building with Moneybot, which was a key example I cited in last week’s PFM 3.0 essay. Great stuff, as usual! 

I really enjoyed this deep dive from Marc. Subscribe to Net Interest if you haven’t already!

1 QUESTION FROM THE FINTECH TAKES NETWORK

There are a TON of interesting questions being asked in the Fintech Takes Network. I’ll share one question, sourced from the Network, each week. However, if you’d like to join the conversation, please apply to join the Fintech Takes Network.

What’s another good example of financial services workflow where human intelligence − human psychology would = better long-term customer outcomes?

If you have any thoughts on this question, reply to this email or DM me in the Fintech Takes Network!

Thanks for the read! Let me know what you thought by replying back to this email.

— Alex  

Join 2,444 other finance and fintech leaders in the Fintech Takes Network

The conversation doesn't have to stop here
Keep learning and connecting in the Fintech Takes Network

EVENTS  |  FEED  |  LIBRARY  |  DIRECTORY

Get your brand in front of 62,800+ fintech and banking executives.

Workweek Media Inc. 
1023 Springdale Road, STE 9E
Austin, TX 78721

Want to ruin my day? Unsubscribe.