A panel discussion on the future of agentic commerce with Ben Brown (Partner, Flagship Advisory Partners), Craig DeWitt (Chief Product Officer, Skyfire), Robin Gandhi (Chief Product Officer, Lithic), and Colin Luce (Chief Executive Officer, Basis Theory).
Together, we explored how agentic commerce could transform payments, digital identity, and risk, and discuss the opportunities and challenges ahead as this emerging landscape takes shape.
This podcast episode is based on a webinar Flagship hosted on 14 October 2025: https://insights.flagshipadvisorypartners.com/webinar-the-rise-of-agentic-commerce .
Do not hesitate to reach out to Ben Brown, Ben@Flagshipap.com with any comments or questions.
Listen to the latest episode on the following platforms:
Elisabeth Magnor
This is Fintech Conversations with Flagship, a podcast production by Flagship Advisory Partners. In this podcast, we discuss the big topics shaping the world of payments and fintech, and each episode will bring you sharp insights and fresh perspectives from the industry. Whether you're building, investing, or simply curious about fintech, you're in the right place.
And welcome to this episode of Fintech Conversations with Flagship. In this episode, we'll dive into one of the biggest buzzwords in fintech and payments: agentic commerce—the new world where AI agents act and transact on behalf of consumers and businesses. Over the past months, we've seen exploding interest in the field, and everybody's wondering: will agentic commerce change the game of payments like the internet did, or is it simply hype? In this episode, you'll first hear Ben Brown, Partner at Flagship, sharing an overview of what agentic commerce is and how it works for around 15 minutes.
Then, for the rest of the episode, you'll hear a discussion panel with industry leaders from Basis Theory, Lithic, and Skyfire. This episode is based on a webinar we hosted, and if you'd like to watch the full video recording and presentation, this is available on Flagship’s website. Let's get started.
Ben Brown
Thanks, everybody, for joining. We'll kick off today's call with a short overview of what agentic commerce actually means, how it works, and the impacts on the payments ecosystem.
Once I've set the context, we'll spend most of the time on a discussion with three innovators who are on the front lines of building new payments and infrastructure for agentic commerce: we’ve got Craig DeWitt at Skyfire, Colin Luce at Basis Theory, and Robin Gandhi at Lithic.
It’s a cross-section of companies building infrastructure for the agentic commerce economy across things like end-to-end payments platforms for agentic commerce, digital identity for agents, virtual cards, and supporting some of the cutting-edge evolution of network tokenization we see. I really appreciate all the panelists for joining us today. I'll try to keep the Flagship part of the discussion as brief as possible because we want to set good context for the discussion to come.
One place we wanted to start is: does this matter, and is this a real thing? We hear from clients a variety of perspectives all the time. There aren't many forecasts out on agentic commerce yet, and the ones that do exist vary widely.
Some say as high as 40% of all global transactions by 2030—five years from now—will be agentic-influenced. Our point of view isn't that aggressive, but, like Bill Gates said, people tend to overestimate what will happen in a year and underestimate what will happen in 10. To that end, we do think agents are likely to be meaningful and perhaps influence ~20% of digital payments in the U.S., ~30% of total payments in the U.S., in 10 years. That means we're at the early stage of a once-in-a-generation emergence of a new form of commerce and payments. It’s exciting to be here at the ground floor—for everyone in payments, and broader merchants, online advertising, and AI builders as well.
While we're early, we're also seeing a parade of announcements across the industry, from many different players. Merchants like Amazon and Walmart have rolled out conversational shopping assistants inside their apps. Walmart has even rolled that out on the seller side with its “Marty” agent for marketplace seller onboarding.
That isn't something we talk about a lot, but a big potential impact of agentic AI is expanding participation in the digital economy—smaller businesses, more custom goods, the circular economy—all the things that can be hard to put online. Agentic AI can really help.
We're also seeing CDNs like Cloudflare and Akamai partnering with companies like Tollbit—and even Skyfire, who’s with us today—so publishers can authenticate and monetize AI bot traffic, which is increasing at an enormous rate. Today, roughly half of all web traffic is bots and half humans. And of those bots, it's almost 50–50 between good bots and bad bots.
One big impact: if your fraud models assume all bots are bad, you're going to have a lot of false positives as agentic traffic takes off. We see big tech companies like Google announcing protocols for agentic-led payments with long lists of merchants, FIs, and payment networks participating.
Visa and Mastercard have both announced solutions to evolve tokenization for an agentic world. Even today—six hours ago—Visa announced its latest proposal around agentic identity and trusted agent identities. And these early announcements are quickly turning into live experiences, such as shopping on Perplexity or buying from Etsy sellers natively within ChatGPT, which you can see here as an example on the page.
As this all comes together, across the customer journey AI promises to impact everyone: the search engines and social platforms that drive digital advertising, merchants, payment processors, card issuers. Some impacts are obvious—zero-click search and embedded buying diverting consumers from visiting merchant sites, or agents confusing fraud tools tuned to recognize bot traffic. Some are less obvious—explicit agentic consent trails could reduce certain dispute types if we can verify who authorized what, especially at SKU level. At the same time, issuer costs for card benefits (extended warranties, price protection, etc.) could change if agents systematically trigger claims at dollar amounts none of us would manually monitor.
Every company in payments—and the broader digital commerce ecosystem (merchants, ad tech, commerce platforms)—should think about second-order effects, not just the obvious headlines.
There are multiple ways agentic commerce can happen in retail. It won’t all be shopping within a conversational interface of an LLM. A few models:
Where it gets interesting is when agents act as facilitators or aggregators—either initiating payments through the merchant’s commerce/payments stack, or acting as the marketplace themselves and owning the full customer relationship.
Beyond these, we may see specialized agents for parts of the journey. For example, agents that take care of the checkout form for you—even if the rest of the shopping journey is normal. That could make every product page a checkout page and give every merchant one-click checkout, without reconfirming shipping info and CVV codes.
We’re payments people at Flagship, so: how are payments working in these scenarios? We see four primary models:
Networks are steering toward an agentic token model. There’s a range of protocols and frameworks already announced. Common theme: evolve the tokenization infrastructure (billions of tokens over the past decade underpinning digital wallets) for an agentic world, with more data. Differences remain in digital identity and intent signals. And these aren’t the only solutions; for example, Worldpay announced a partnership with Trulioo on a “Know Your Agent” framework. Panelists here have also built solutions in this space. Lots of proposals and early work—unclear exactly what “wins.”
Double-clicking one example—the Google AP2 agentic payments protocol—you’ll hear terms like intent mandates, cart mandates, and payment mandates that aren’t familiar today. Think of them as notarized breadcrumbs of who asked for what, what items and price were approved, whether a human was present, and what payment method to use. That breadcrumb trail will be gold for issuers, acquirers, and merchants to tune fraud models and adjudicate disputes. Those signals will create impacts for every player.
Three takeaways:
With that, let’s shift to our panel—people hands-on, building the future. I’d love to bring in Colin from Basis Theory, Robin from Lithic, and Craig from Skyfire. It’d be great if each of you could quickly introduce yourself, what your company is doing in agentic commerce, and what makes you most excited about the space.
Robin Gandhi (Lithic):
My name’s Robin. I run Product at Lithic. Lithic is a next-generation modern issuer processor—similar vein as the Marketas of the world. We issue cards.
We’ve been around in many iterations for the last 10–12 years; in this iteration—selling infrastructure—for the last four. For agentic commerce, we’ve been experimenting, especially with folks here on this call. Where we as an issuer really play is around virtual cards. We’ve been exploring how to generate highly controlled virtual cards and give them to an agent to make purchases. It’s early days, but we’re pushing hard because we believe virtual cards are one way to solve this. There are other mechanisms, and we’ll get into those, but virtual cards are the obvious step one.
Colin Luce (Basis Theory):
Colin Luce, co-founder and CEO at Basis Theory. Our platform puts merchants and platforms back in control—with vaulting and tokenization as the underpinnings—so you can imagine how that maps to agentic commerce. Many use cases we’ve enabled over the past 3–4 years are distributed, embedded shopping experiences; agentic shopping is the next evolution. Social shopping has grown (TikTok Shop, Instagram links); ChatGPT is just the next medium where attention lives. We focus on enabling commerce where attention lives.
Most in-production use cases today are non-AI “agentic” commerce experiences; we’re at the precipice of the agentic-AI phase.
Craig DeWitt (Skyfire):
I’m Craig DeWitt, co-founder of Skyfire—an agentic commerce platform. We serve AI platforms/agents and merchants who want to accept them. On the AI side, we provide identity and the ability to use payments to buy what’s needed—both stablecoins and, now with what card networks are bringing, tokenized cards. On the merchant side, we let merchants identify bots, identify not just the agents but the principals behind them (the user or business), and accept them as paying customers—through the website (most volume today) and headless via a variety of API approaches and protocols. That’s Skyfire.
Ben Brown
Awesome—thanks to all three. Great complementary panel: Craig is all-in on agentic AI end-to-end, Colin enabling tokenization that underpins a lot of this (and evolving into broader solutions), and Robin/Lithic from the cloud-based issuer-processor perspective. These components come together to make the future happen.
Let’s jump in. Colin, you said we’re not seeing a lot of truly agentic experiences yet. What are the most mature ones today—verticals, apps, merchants?
Colin Luce
I like your friend’s take that we should celebrate building blocks, although it can sound like a participation trophy. The most mature “agentic commerce” today is non-AI agentic. Example: Roku’s shoppable ads platform (we helped enable ~3.5 years ago): one-click buy from the Roku device. That relies on tokenization, checkout APIs—the same backend ideas we’re contemplating, just without AI.
Important nuance: agentic commerce vs agentic payments. ChatGPT Commerce and Perplexity are agentic commerce (efficiency in discovery/comparison). The payment flows are existing infrastructure. So when people say “what about risk/fraud?”, if payment flows mirror today’s, existing tools should apply. Transaction volumes on AI platforms are very early (think tens/hundreds per month at some), so we’re at day one.
On disputes/chargebacks and legal frameworks—it’s like self-driving cars. Who’s responsible in a crash? We’re years into autonomy; accidents didn’t stop adoption, but adoption is still not universal. It’s amazing we can autonomously drive across San Francisco, but we can’t autonomously book an airline ticket reliably. “Atoms vs bits.”
Robin Gandhi
Colin and Craig probably see more consumer examples. I think B2B use cases will be more interesting. People like buying fun stuff; removing the person from that experience is less compelling. The transformative impact is on things people hate buying—marketing spend allocation (“here’s 100k, spread across platforms”), supplier payments, low-consideration replenishment. Smaller ecosystems where trust can be built. That’s where we’ll see traction first.
Craig DeWitt
On the e-commerce use case people imagine—new interface where an agent purchase happens—the coolest I’ve seen is Consumer Reports: they already have millions of users who read reviews; add a “Buy it” button. Want a toaster? Click buy. They spin up a headless browser, navigate to the merchant, find the right SKU, autofill card details, and buy on behalf of the user. That uses card tokenization, etc.
The hard part so far is access: how does the agent get access to the site? HTTP 403 errors are going through the roof as sites block programmatic access due to AI traffic. A big part of our business is helping agents identify themselves to get access. We’re seeing many B2B use cases here, because the way the internet worked (open crawling) is changing as bots flood it. It’s much harder to get access than two years ago.
Ben Brown
I see Cloudflare “are you human?” a lot—false positives are up. That’s a fundamental change and creates opportunities. Let’s talk where this moves faster/slower. a16z wrote that AI plays different roles by purchase type (impulse, essentials, lifestyle). Some will be algorithmic impulse nudging (TikTok); others price trackers; others interactive AI coaches (mortgages, autos). Some sectors move faster.
If we’re here in 1–2 years, what moves the fastest?
Colin Luce
It depends on context. The value of AI is proportional to the context it can access. Procurement has written rules—high context. Consumer Reports—very acute context. Social platforms have massive data about us. ChatGPT/Perplexity are well-positioned: they have conversational context (are you value-seeking, loyalty-driven, etc.).
Payment transaction descriptions are still messy (15 years on). Networks saying they’ll provide context—maybe—but “$3,000 at United” doesn’t tell you if I bought one business-class seat or seven economy seats. Or “book me a flight to Miami under $X over the next two months”—what if my plans change? Without calendar/travel context, it breaks. So it’s not B2B vs B2C; it’s wherever the agent can get the most context.
Ben Brown
What’s the single biggest blocker today—payment plumbing, digital identity, data quality, consumer trust, or something else?
Craig DeWitt
Procurement: we automated buying a specific resin for a large OEM partner. Identity and payments worked; internal processes were the hurdle—ERP, wires, CSVs, manual steps. For e-commerce, the blocker is user adoption—is it 10x better? For B2B content publishers, the blocker is acceptance: ad models are breaking, but they need to understand protocols/services to allow agents and accept programmatic payments.
Net: “agentic commerce” is a huge umbrella; each use case has its own hurdles. Some will react by turning things off (banks, publishers, merchants). Others will endure a painful transition while putting new infrastructure in place. If everyone “turns it off until infra is ready,” we get a chicken-and-egg problem.
Robin Gandhi
Merchant acceptance is the biggest piece. We’re building enablement, but the merchant needs a clear signal: “this is good traffic, we captured intent, and we know the agent.” Once we agree on a protocol and acquirers support it, we can solve it. We tried a simple experiment—making a donation to the San Mateo Library—and it got rejected as fraud. That shows how broad the “block bots” posture is. We need acquirers/PSPs to enable the right protocol end-to-end.
Ben Brown
We’ve seen a flurry of frameworks—Visa, Mastercard, Google (AP2), Stripe (ACP), etc. Are they fundamentally different or flavors of the same thing?
Colin Luce
My biggest concern is competitive dynamics and incentives. Collaboration is needed among natural competitors. AP2 is pushed by Google; ChatGPT is well-positioned—are they going to play nicely? On PSP/acquirer side, Stripe is behind ACP; will Adyen or Checkout be comfortable supporting a rival’s protocol? Specs might be open, but decisions aren’t always purely technical.
If merchants have to integrate 12 protocols, adoption stalls. Even within the networks’ offerings (e.g., VIC/AgentPay), issuer support is limited. You can’t ask consumers to pick their issuer behind the scenes. We need convergence.
Craig DeWitt
Common threads: identity exchange and a payment construct. But many specs try to position who holds power. AP2 (Google) wants to collect mandates; networks say “we carry chargeback risk, we hold mandates”; ACP says “just integrate OpenAI and we’ll handle it.” That’s not great for industry simplicity; it is an opportunity for startups to “future-proof” merchants.
On SKU-level data: ACP/OpenAI asks for full product catalog/pricing APIs. Big brand merchants say no way—they won’t become a pure fulfillment back-end. They want a direct relationship with the end user and visibility into what the agent is accessing. Some will turn it off and lose sales; others won’t give it all away. Etsy/Shopify are more likely to lean in, given their merchant base. A Home Depot-type merchant wants the contractor’s agent to buy 10 feet of pipe—but they need identity exchange to preserve the relationship and history.
Colin Luce
Merchants are asking: what’s the tax? Everyone pays a tax today (ads CAC, take rate). Is the agentic platform tax 2% or 20%? If the tax is zero and traffic is free, merchants might share SKU data. That decision hinges on the economic model.
Ben Brown
We’re close to time. One piece of advice for the ecosystem—what should they do tomorrow? Robin (issuers), Colin (PSPs/acquirers), Craig (merchants).
Robin Gandhi (Issuers)
Be in the conversation on protocols. Work with the card brands to support emerging token frameworks. Most modern issuers can spin up highly controlled virtual cards already—connect the dots and be ready.
Colin Luce (PSPs/Acquirers)
You’re the gatekeepers to merchants. Do the work to make adoption feel like adding Apple Pay/Google Pay—as close to a flip-a-switch as possible. Your one-to-many leverage is key.
Craig DeWitt (Merchants)
(Easy answer: reach out to Skyfire! And call Flagship for strategy.) Practically: start authenticating good agents, run controlled pilots, and choose a partner that abstracts protocol complexity so you’re future-proof.
Ben Brown
Thank you to the panelists and to everyone in the audience for joining us.
Elisabeth Magnor
That’s it for now. Thanks for listening to Flagship. We’re always happy to chat. Stay tuned for more exciting fintech conversations.