Dan Shipper talks with Emily Glassberg Sands, Stripe's Head of Data and AI, about what data from Stripe's network—which processes roughly 2% of global GDP—reveals about AI's impact on the internet economy. The conversation covers AI companies scaling three times faster than the top SaaS cohort of 2018; fraud moving from checkout to the full customer lifecycle, including compute theft as a new payment fraud; the AI-on-AI arms race between bad actors and fraud detectors; where AI revenue growth is coming from; outcome-based billing replacing seat-based pricing; how the developer experience changes when agents build software; the agentic commerce spectrum from assisted buying to autonomous purchasing; and how Stripe is rebuilding payments infrastructure for agent-driven commerce, including the Link consumer wallet for delegated agent purchases. Timestamps: 00:00:45 Introduction 00:01:27 New rules for an agent-driven economy 00:03:57 Compute theft is the new payment fraud 00:10:00 Fraud detection expanded from checkout to the full customer lifecycle 00:19:48 Why AI companies are scaling faster than top SaaS companies 00:23:27 Outcome-based billing is replacing seat-based pricing 00:29:57 Where AI spending is coming from 00:36:45 How the developer experience changes when agents are the builders 00:41:00 The agentic commerce spectrum, from assisted buying to autonomous purchasing 00:51:06 Link: a consumer wallet for delegated agent purchases # Channel Identity Observations — "Getting Agent Ready With Stripe's Head of Data and AI" ## Explicit Self-Description Evidence - **Channel name: "AI and I"** — stated in the closing segment: *"You absolutely, positively have to smash that like button and subscribe to AI and I."* - **Host identity: Dan Shipper** — the outro names *"Dan Shipper as the captain of the spaceship,"* and the guest addresses him as **"Dan"** at the open (*"Emily, welcome to the show… Thanks so much, Dan"*). - **Promotional framing of the show** — the outro (apparently a comedic or AI-generated bit) describes the show as delivering *"knowledge bombs about chat GPT"* and *"a journey into the future."* This is self-description, though its exaggerated tone suggests parody. ## Observable Presentation ### Format - **Long-form expert interview podcast/video** — a single one-on-one conversation with a senior industry guest (Emily, introduced as *"head of data and AI at Stripe"*), covering strategy, product detail, and data points. - **Host as curious operator, not neutral journalist** — Dan discusses his own company's setup: *"I'm sitting here… doing millions of dollars a year in Stripe transactions"* and checks his own Radar dashboard live mid-episode (*"I'm doing that right now… We are at 0% fraud"*). ### Demonstrated Audience - **Builders, founders, and technical/business operators in AI** — questions center on practical decisions: *"If you're running an AI company and you don't even know what your fraud rate is… what are the top things you need to do?"* - **Audience assumed to be AI-native and somewhat technical** — casual references to "ramp," "vibe coding," token metering, CPQ, and the aside *"with the exception of maybe some nerds on the call."* ### Claimed/Demonstrated Scope - **Focus on the AI economy and its infrastructure** — topics span agent-driven fraud, monetization models, agentic commerce protocols, and developer tooling. - **Interviews with executives shaping AI infrastructure** — the guest's role and access to Stripe data (*"Stripe looks at 2% of global GDP"*) position the show around insider macro views. - **Recurring-show framing** — Dan closes with *"We'll have to have you on again,"* suggesting an ongoing interview series. ## Host vs. Guest Evidence Separation - Claims about Stripe's products and data (Radar, Link, agent commerce protocol, ARR growth stats) are **the guest's claims**, not channel claims. - The channel's own demonstrated claims are limited to the show name, host, subject-matter scope, and audience posture described above. # Delivery Observation: *Getting Agent Ready With Stripe's Head of Data and AI* ## Overview Two-speaker interview format: **Dan (host)** interviewing **Emily (guest, Head of Data and AI at Stripe)**. Delivery is conversational, data-heavy, and punctuated by recurring verbal tics and structural habits. --- ## Tone - **Conversational-analytical**: The discussion of fraud and billing stays informal despite dense statistics — *"You wouldn't believe how many lazy parents there are in the world."* - **Playful interludes**: Humor breaks up technical content — *"I think a good CRM seat is… three LLMs are for sure more tempting."* and Dan's *"I'm Yolo vibe coding my payment infrastructure."* - **Enthusiastic host delivery**: Dan repeatedly signals engagement — *"That is fascinating,"* *"Fascinating. I feel like I'm learning so much. This is amazing."* ## Vocabulary & Register - **Business/VC jargon**: ARR, CAC, opex, unit economics, freemium, ex ante/ex post — *"free compute is kind of the new CAC."* - **Economist's self-framing**: Emily flags her framing explicitly — *"I'm an economist by training, so I'll be a little nerdy."* - **Stripe-internal product names used casually**: Radar, Link, shared payment token (SPT), agent commerce protocol, token billing. ## Recurring Phrases & Verbal Habits ### Emily (guest) - **"Right?" as a checking tag**, very frequent: *"the internet economy is becoming more autonomous, right?"*, *"they don't care about, you know…"* - **Filler clusters**: "you know," "um," "sort of" appear constantly — *"it's sort of this like and you probably feel it as a consumer."* - **"Kind of" as a hedge**: *"kind of the larger frame of the moment,"* *"kind of a whole new world."* - **Ordering habit**: Enumerates points — *"is threefold. One is multi-account abuse… the second trend… the third type."* ### Dan (host) - **Repeated restarts/stammering on questions**: *"How are you guys even able to I So So I totally understand…"*; *"And is the is it is it the the volume of fraud constant?"* - **Self-referencing as a customer**: repeatedly works his own usage into questions — *"we have a bunch of virtual cards,"* *"I'm doing that right now,"* *"We are at 0% fraud."* ## Teaching / Explanatory Habits - **Concrete numbers as anchors**: Emily attaches figures to nearly every claim — *"about 7% of their sign-ups,"* *"$625 per payer,"* *"250,000 fraudulent free trials a week,"* *"58% of their payment volume runs through Link."* - **Explicit magnitude framing**: *"just to give you a sense of the order of magnitude,"* *"order of magnitude, you can think of like exponential growth."* - **Old-vs-new contrast structure**: repeatedly frames points as historical shift — *"fraud used to be a transaction thing, now it is a customer thing,"* *"We used to think of fraud as sort of payment fraud."* - **Spectrum framing**: *"I prefer to think about it as a spectrum"* when explaining agentic commerce, then walking through levels sequentially. - **Live demo moment**: Dan checks his own Radar dashboard mid-interview — *"Uh 0.02% early fraud warnings, total fraud rate 0.2%. So, we're… doing pretty good, right?"* ## Interview Structural Patterns - **Host tees up numbered/layered questions**: *"A, I want to know what that means for Stripe, but B, I want to understand…"* - **Guest product plug folded into advice**: *"I would just adopt our highest tier Radar plan, but but the actual mechanics of that are…"* - **Canned outro**: The episode closes with an exaggerated, hyperbolic subscribe pitch — *"smash that like button,"* *"Dan, I'm absolutely, hopelessly in love with you"* — a stark tonal break from the interview. ## Stable vs. Isolated - **Stable**: Emily's data-anchored enumeration, "right?"/"you know" fillers, and before/after framing recur throughout. - **Isolated**: the live dashboard check, the "Halloween costume" exchange, and the outro monologue appear once each and shouldn't be treated as patterns. - **Uncertainty note**: pauses, laughter, and overlapping speech are marked in the transcript but exact frequency of interruption-style habits cannot be verified beyond quoted instances. # Observation Document: "Getting Agent Ready With Stripe's Head of Data and AI" ## Participants - **Emily** — Head of Data and AI at Stripe (identified via host's introduction) - **Dan Shipper** — Host (identified in transcript; also named in outro) --- ## 1. The Agent Economy Frame **Emily's core thesis:** The internet was built on the assumption that the main actor is a human sitting at a screen — browsing, filling forms, checking out, writing code. That assumption is breaking in stages: - Humans still in control but interacting through AI interfaces instead of websites/apps - Agents acting on humans' behalf - Software interacting directly with other software **Claim:** "Over time, this actor, these agents will become the predominant actors on the internet." — Every layer of the stack needs evolution: product discovery, purchasing, developer tools, and underlying economic infrastructure (payments, billing, fraud, identity). **Stripe's two-sided mandate (per Emily):** get Stripe itself "agent ready," and help businesses get agent ready. --- ## 2. New Fraud Landscape **Key redefinition:** Fraud used to mean payment fraud (stolen money/card credentials). Now, per Emily (citing a meeting that day with a large AI user), **fraudsters are stealing compute** — a very different problem. **Reasoning:** - In traditional SaaS, free tiers cost little and were worth little to steal - In AI, every prompt, image, and API request carries real cost — "free compute is kind of the new CAC" - Intelligence is getting cheaper but "still very far from free" ### Three abuse vectors (Emily's data points) 1. **Multi-account abuse** — repeated sign-ups with new identities claiming new-user credits. ~**7% of sign-ups** across AI companies on Stripe are multi-account abusers. 2. **Free trial abuse** — Example: a large AI company with only **4% trial-to-paid conversion**, each trial costing **$25** in LLM spend → **$625 cost per payer** before first revenue. "Vast majority" of trial users were abusers with no intent to pay. - Free trial abuse has **4x'd in 6 months** - One large AI user: **250,000 fraudulent free trials blocked per week** - Some companies dropped free trials (throttles growth); others blocked virtual cards — but ~**15% of legitimate AI-company card transactions on Stripe are virtual cards**, so blanket blocking harms real users 3. **Non-payment abuse** — consuming thousands/tens of thousands of dollars in compute via invoicing or overages, then failing to pay. Loss has already happened by billing time. **Fraud is now full-funnel, not transactional** (Emily): it starts at sign-up. **On fraud growth (Emily answering Dan):** Fraud is increasing both because fraudsters have AI and because the value of what can be stolen (compute, tokens) is higher than traditional SaaS ("three LLMs are more tempting" than a CRM seat, she joked). ### Radar product evolution - Radar moved from checkout-time to sign-up-time integration: metadata at sign-up, scores passed back, evaluation "every moment subsequently" - Four protected moments: **sign-up, payment, overage, refunds/disputes** - Extended beyond cards to **ACH, SEPA, all payment methods with disputes, and crypto**; a **Radar API** allows screening transactions not processed on Stripe (e.g., on Worldpay or Adyen) - Emily: Stripe sees **2% of global GDP**, growing **34% year-on-year**; views fraud mitigation as a **public good**, justifying investment beyond direct Stripe value - Overall stance: "bullish but certainly not complacent"; agents review anomalous patterns hourly/daily to burn down new vectors **Live demo moment:** Dan checks his Radar dashboard mid-interview and reports ~0.2% total fraud rate. Emily quips, "Maybe the fraudsters don't want to come after you." --- ## 3. AI Company Growth and Monetization **Growth claim (Emily, from top 100 AI companies on Stripe):** - Companies reaching **$30M ARR do so in ~18 months — 3x faster** than top 100 SaaS companies of 2018 - Similar orders-of-magnitude speedups to $1M and $5M ARR **Monetization model iteration:** - Seat-based SaaS fit humans and near-zero marginal cost; real inference costs force change - **Usage-based billing** on tokens, API calls, workflows, and outcomes — "whatever unit best reflects both the customer value and the cost structure" - Hybrid models common: subscriptions + overages, prepaid credits, real-time top-ups - **Lovable example:** started with Stripe Billing subscriptions, added usage-based token billing above thresholds (Lovable Cloud, Lovable AI) - **Token billing:** real-time pricing tracked to underlying model token costs plus markup — protects "wrappers" whose margins "can disappear" if model costs rise and prices don't ### Emily's pricing predictions - **Model providers:** will meter tokens (with usage/overages dominating; seats disappearing) - **Vertical AI solutions:** in steady state, will meter **outcomes** — because end users will demand ROI accountability from vertical providers. Infrastructure is ready (Intercom/Fin charges per resolved support ticket). She notes outcomes will be **multidimensional and complex** ("objective functions": complexity, quality, CSAT, cost of the automated worker) and will take time to get crisp - "I just have a hard time imagining a year from now most vertical providers are literally charging on tokens" - **Seat-based billing will disappear in enterprise:** if agents make developers 10x productive, "don't you need 1/10 the developers?" She'd be "super surprised" if six months from now enterprises have half the seat-based licenses of today - **Consumers differ:** most individuals remain uncomfortable with anything but a fixed monthly fee, maybe with overages ### Dan's counterpoint (preserved) Dan describes his company's model — per seat **with overages** — explaining seats persist because buyers know how to model costs, and overages protect margins now that software has real run costs. Emily agrees usage/overage charging won't go away for model providers, but predicts seats will. --- ## 4. Where the Money Comes From **Emily's analysis:** - AI growth so far has been largely **net new spend**, not substitution — due to experimentation, slow license turnover, and AI starting near zero - Going forward she expects substitution from three sources: 1. **Traditional SaaS** (including SaaS companies' own AI arms cannibalizing their legacy revenue) 2. **Headcount opex** — example: an engineer costing $300K salary + $30K in LLMs; companies will reason about budgets "on a plus-10% basis" 3. **Other AI providers** — category retention is higher than SaaS (B2C and B2B within e.g. AI dev tools), but individual-provider retention is slightly lower ("curious travelers hopping" between providers as models improve) **On valuations:** Top-100 Stripe AI companies mostly solve disjoint vertical problems with no competitor yet — enough "blue ocean" that "overall AI valuations are probably okay," with a few "frothy" crowded spaces. Micro evidence of competition: sales-led contracts shifting from sticker pricing to negotiated CPQ-heavy billing as direct competitors emerge. --- ## 5. Agent Experience / Developer Tools **Reframing "developer" (Emily):** The persona now includes non-technical founders describing apps in plain language (Vercel, Replit), coding assistants scaffolding integrations, and agents provisioning infrastructure. Goal: a coherent, trustworthy end-to-end product experience regardless of whether the actor is human, agent, or human-through-agent. **Data point:** **LLM traffic to Stripe docs is up 10x year-over-year** — signal that machines are becoming users of developer infrastructure. Human docs use is flat-to-climbing (not substitution); humans also check docs to sanity-check agent output for payment integrations. (Dan admits he is "Yolo vibe coding" his payment infrastructure.) **Stripe Projects** (launched ~2 weeks before interview, per Emily): - Lets humans or agents create and manage parts of the software stack from the command line; resources provision in accounts you own; credentials sync back - Launched with Vercel, Supabase, PostHog, Neon, RunLoop; **100+ companies reached out post-launch** wanting to join - Motivation: coding gets easier faster than setup/provisioning does --- ## 6. Agentic Commerce **Emily's spectrum** (warning against jumping to extreme visions): 1. AI removes friction — human still decides 2. Descriptive search replacing keywords/filters 3. **Real delegation** — human gives constraints; system purchases on their behalf (the "minimum viable bar" for agentic commerce) 4. **Ambient commerce** — unprompted, system knows seasonal needs (e.g., summer camp planning) ### Agent commerce protocol (ACP) - Co-created with **OpenAI**; also used by Microsoft Copilot and Meta's in-ad shopping - Merchants integrate once with Stripe (catalogs, prices, checkout), then toggle on across agent surfaces from the dashboard - **Merchant remains merchant of record** — critical because businesses want channel access without giving up customer relationships, trust, or fraud control - **Shared Payment Token (SPT):** lets payment credentials pass securely from agent to merchant (credentials not viewed by the agent); includes Radar fraud scores so merchants can act even when processing off-Stripe **Dan asks whether he can integrate directly** (not just via OpenAI/Meta). Emily: yes — premise is there won't be "one agentic shopping experience to rule them all," and merchants would "literally break" if forced to integrate with every storefront. Stripe's decade-long pattern: sell wherever customers are (websites → apps → marketplaces → in-person → agentic). ### Current volume and categories - Agentic consumer commerce is **"relatively small" as a percentage** of commerce Stripe sees, but growing quickly - Dominated by **commodities** — known, observable, low-priced items. Example Emily cites: **Halloween costumes** ("You would not believe how many lazy parents there are in the world") - Emily's personal test: tried booking a full family summer trip via an agent; wasn't satisfied enough with flights/hotels/itinerary to one-click buy. She expects models, interfaces, and experiences to improve; Stripe is agnostic to experience evolution and focused on underlying primitives ### Link consumer wallet - ~**250 million consumers** use Link - Dense in AI: **58% of Lovable's payment volume runs through Link** - Evolving Link for delegated authority with guardrails: consumers decide which agents can request credentials, under what conditions and limits, and whether purchases require approval — "not give a random agent your card and hope for the best" --- ## 7. Notable Jokes / Asides - Dan jokes an agent might "steal someone's credit card... you never know with Grok." Emily: "No comment." - Dan on seat-based SaaS fraud: "Who doesn't love a good CRM seat?" - Outro contains an over-the-top promotional plug and a declaration "Dan, I'm absolutely, hopelessly in love with you" (program/comedic outro segment). --- ## Tensions and Unresolved Points (preserved) - **Seat-based pricing:** Dan defends per-seat-with-overages (predictability); Emily predicts seats vanish in enterprise within ~6 months while persisting for consumers. Disagreement left open; Dan invites a follow-up episode. - **Valuations:** Emily hedges — "probably okay" overall, "a little frothy" in pockets — no definitive claim. - **Outcome-based pricing:** Emily frames it as a steady-state destination with multidimensional measurement still to be worked out; timeline uncertain. - **Agentic commerce maturity:** Emily explicitly notes we are not at ambient commerce and that current volumes are small; her booking experiment failed her satisfaction bar. # Speaker & Role Attribution Record ## Video - **Title:** Getting Agent Ready With Stripe's Head of Data and AI - **Video ID:** `01a0f15a-8874-786d-ac02-13595efe6c81` - **No curated metadata was supplied**; speaker identities below are inferred solely from the transcript. ## Speakers & Roles ### Dan Shipper — Host / Creator - Identified by name in-transcript ("Emily, welcome to the show"; "with Dan Shipper as the captain of the spaceship"). - Conducts the interview, asks questions, describes his own company's Stripe usage ("millions of dollars a year in Stripe transactions"), and checks his Radar dashboard live during the episode. - Also identified himself as "Yolo vibe coding" his payment infrastructure. ### Emily — Guest (Head of Data and AI at Stripe) - Identified in-transcript: "You are the head of data and AI at Stripe." - Primary source of claims about Stripe data, Radar, agentic commerce, Link, and AI-economy trends. Also states she is "an economist by training." ### Promotional Narrator / Outro Voice — ⚠️ Promotional Segment - End-of-episode voice urging viewers to "smash that like button and subscribe to AI and I," praising the show and host, ending with a declaration of affection for Dan. - **Flagged as promotional content**, distinct from the interview. ### Unattributed Narration (Opening) - Cold-open lines ("The internet has this new kind of actor on it…") appear without speaker attribution; they repeat lines later spoken by Emily, suggesting a highlight reel, but attribution is **uncertain**. ## Third-Party Entities & Claims Referenced All claims below are attributed to **Emily (Stripe)** unless noted: - **Stripe internal data claims:** LLM traffic to Stripe docs up 10x YoY; ~7% of AI-company signups are multi-account abusers; free trial abuse up 4x in 6 months; 250k fraudulent free trials blocked weekly for one large AI user; ~15% of legitimate AI-company card transactions are virtual cards; top 100 AI companies reach $30M ARR in ~18 months (3x faster than top 100 SaaS from 2018); Stripe sees 2% of global GDP, growing 34% YoY; one company's free trials converted at 4% at $25 cost each ($625 per payer). - **Named third parties:** OpenAI (co-creator of Agent Commerce Protocol), Microsoft Copilot, Meta, Intercom/Fin (outcome-based billing), Lovable (58% of payment volume via Link), Vercel, Supabase, PostHog, Neon, RunLoop, Worldpay, Adyen, Ramp (mentioned by host re: virtual cards), Mesa, Ship. - **Host-sourced claims:** host references Claude/ChatGPT per-seat-plus-overage pricing at his own company and Grok remark (joking, unverified). - Anonymous example: "a large AI company" and "one of our very large AI users" — identities not disclosed. ## Provenance & Availability Notes - **Comment availability:** Not indicated in source material. - **Sources/citations:** All figures are speaker assertions from Stripe internal data; no external documents or links provided. - **Uncertainty:** Guest's surname not given in transcript; metadata absent. The internet has this new kind of actor on it. Over time, this actor, these agents will become the predominant actors on \[music\] the internet. These AI companies are just growing from a revenue perspective faster \[music\] than any previous cohort we've seen. LLM traffic to Stripe docs is up 10x year-over-year. And that's just a useful signal that machines are becoming users of developer infrastructure, too, including Stripe's developer infrastructure. \>\> \[music\] \>\> Emily, welcome to the show. Thanks so much, Dan. So, uh really excited to have you. You are the head of data and AI at Stripe. Um and I feel like this is such a good time to have someone from Stripe on because you all famously are increasing the GDP of the internet. And the internet is changing so much right now. And and therefore the like economy of the internet is changing from something where uh humans are buying and selling from each other to a thing where uh to an economy where agents are buying and selling from humans and agents are buying and selling from each other. And I feel like A, I want to know what that means for Stripe, but B, I want to understand, since you have this macro view of the agent economy, what does that even mean and what are you seeing? Yeah. So, a big shift I think we're in the midst of is that the internet economy is becoming more autonomous, right? So, for a long time, for forever, right, the internet was built around a extremely simple assumption that the main actor was a person. And the person sitting in front of a screen and they're browsing and they're filling out forms and clicking through checkout. Um but also they're writing code and setting up tools. And that assumption is starting to break um in various ways, right? Sometimes the human is still totally in control, but they're interacting through an AI interface instead of through a website or a traditional app. Um, sometimes the agent is acting on their behalf, and then sometimes software now is just out interacting directly with other software. Um, and as all of that starts to happen at all of those layers, a lot of things need to be rethought. Um, so, you know, there has been rethinking of how our products discovered and how our products bought, but also what should developer tools look like? And, you know, in our world of Stripe, like what is the underlying economic infrastructure? So, the payments and the billing and the fraud detection and the identity layer um, that's needed in this world where, you know, actors are no longer just humans. And so, that for me is kind of the larger frame of the moment. It's not just, hey, AI's making search better or AI's helping people code or AI's, you know, evolving commerce on the margin. It's really like, actually the internet has this new kind of actor on it. Over time, this actor, these agents will become the predominant actors on the internet. Um, and as that's happening, basically every layer of the stack um, starts to need an evolution. So, for Stripe, it's like, okay, Stripe, how are we getting agent ready? Um, but then also, how are we helping businesses get agent ready? And both of those are happening in in a number of ways. Yes, in commerce, but also um, just in how builders build. And can you give me some specific examples of of the kinds of things you're seeing? Like, I'm almost wondering, for example, I know at at Stripe, one of the things you do a lot of ton is fraud. Um, A, I assume there's a whole new type of fraud happening, but B, I'm almost wondering what what even counts as fraud now in the sense of it's possible that my agent could go steal someone's credit card and check out. I don't think that Claude would, but like, you never know with Grok, you know? Um check out on \>\> No comment. No comment. But you're right that sort of AI introduces very different fraud problems. You asked what is fraud? We used to think of fraud as sort of payment fraud. Someone was stealing money. Someone was stealing your card credentials. Increasingly actually, and I was in a meeting with one of our very large AI users today, fraudsters are stealing compute. And that's a very different type of problem. So in earlier software models, if you think of like sort of traditional SaaS, letting someone into a free tier didn't cost you very much and stealing a free tier wasn't very valuable to the fraudsters. Now giving someone credits, freemium offering, free trial, you know, letting them, you know, rack up a bunch of tokens and pay at end of month except maybe they choose not to pay, actually is a major fraud vector and an existential risk to a lot of these businesses, right? Because in AI, every prompt, every image that gets generated, every API request has a very real cost attached to it. And you know, people are talking about intelligence getting cheaper. Yeah, but it's still like very far from from free. And then also when you look at sort of the growth model for many of these AI companies, free compute is kind of the new CAC, right? You used to cost of acquisition. Like you used to spend a bunch on paid media and now you spend a bunch on your free trials and your credits and your self-serve onboarding as sort of a major lever for growth. And so the abuse we see in that context where compute is the new CAC and compute is very expensive, is threefold. One is multi-account abuse. So this is like bad actors come in and they sign up like over and over again and they create a new identity every time on a new email address and they claim their new user credits um, and they stay ahead of detection by like iterating across a bunch of different aliases. Um, and just to give you a sense of the order of magnitude, across the AI companies running on Stripe, about 7% of their sign-ups are these multi-account abusers. So, non-trivial share. Um, the second trend that we see in sort of new vector of abuse is free trial abuse. And this is often sort of the most urgent issue because the unit economics break really quickly. To give you a sense, we had a large AI company, um, who was seeing only 4% of their free trials convert to paid. And each free trial cost them $25, um, you know, in LLM spend. And so basically, it was costing them $625 per payer before the first dollar of revenue was brought in. And when we double-clicked on those sort of free trial folks, the vast, vast majority of them were actually abusers. So, they were actually stealing the compute, they never had any intent to pay. These weren't people who were genuinely trying out your service and then chose not to buy. These were people who were literally, uh, abusing, abusing your systems. Um, and so, you know, some companies, uh, just dropped free trials altogether. Of course, that's not great cuz you're throttling growth. Um, others, uh, responded by blocking virtual cards. So, um, I don't know how often you've been marketed virtual cards. I'm often marketed virtual cards, right? Get this, you know, one-time use card, it expires after 24 hours, so you never have to pay for the service. Um, you know, in the hands of a good consumer, fine. In the hands of, uh, a fraudster, like very much not fine. The problem with blocking all virtual cards is for AI companies, about 15% of legitimate card transactions on Stripe are actually virtual cards. So, \>\> User only for ramp, for example. Like we have a bunch of virtual cards. \>\> you don't want to be in the same way you don't want to be turning off free trials, you don't want to be throttling um virtual card um virtual cards either. Um and sort of order of magnitude, you can think of like exponential growth in free trial abuse over the last 6 months, it's four x'd. Um and uh for um one large AI user on Stripe, we're currently blocking 250,000 fraudulent free trials a week. Um so, the magnitudes here are quite high. \>\> And is the is it is the the volume of fraud constant? It's just it's shifting shape or is just fraud going up because they're more powerful now cuz they can just use AI agents to do it. Fraud's going up because the fraudsters have AI on their side, although it's also on the side of the detectors, but also because the value of the services they can steal is higher. Right? Like what I don't know. When you steal traditional SaaS, like what good do you get? Like you steal some inference, you steal some compute, you can resell it, you can do all sorts of stuff. \>\> love a good CRM seat, you know. Yeah. Don't you? Who doesn't love a good CRM \>\> Emily. \>\> \[laughter\] \>\> I think a good CRM seat is three LLMs are for sure more tempting. And by the \[laughter\] way, the the third type of sort of new abuse we see is this non-payment abuse, right? So, like you you um incur an overage or you have like, you know, 30-day invoicing, except you never pay um your invoice and you know, in in many cases, customers are consuming thousands or tens of thousands of dollars in compute during a month or a day or sometimes an hour, and by the time they get billed and fail payment, you know, that loss has already happened um and and these AI companies are left holding the bag. And so, for us, like fraud used to be a transaction thing, now it is a customer thing. It is a full funnel thing. It starts at the time of sign up is this multi-account abuse, should they get credits, is this free trial abuse, should we give them a trial in the first place, and then when they have overages, should we be throttling them, should we be requiring top up, should we be blocking service completely? Um and it's just kind of it's kind of a whole new world because the thing to steal is much more valuable, and the cost of having it stolen is much more existential. \>\> How are you guys even able to I So So I totally understand how you need to be in that full funnel in order to detect fraud, but my understanding of, you know, whenever we've integrated Stripe, it's usually like on the checkout. We're not necessarily putting you in there when when someone puts in their free their email address for free. So, have you changed the product to do the full funnel, or how do how does that actually work? Yes. So, Radar, which is our fraud protection product, used to be at the transaction level, right? So, at the at the at the moment of checkout, as you note. Um but because so much of the fraud risk was coming up funnel, um you know, AI companies are now increasingly integrating Stripe Radar at the time of sign up. Um and so, you know, we see the metadata at the time of sign up, we pass back scores at the time of sign up, and every moment subsequently, again, because fraud is now a full funnel problem, not a transaction problem alone. If you're, you know, asking for a friend, if you're in running an AI company and you are you don't even know what your fraud rate is, and you're you want to protect yourself from this kind of abuse, what are the like top things you need to do in order to make sure that you're reasonably safe? Yeah. So, I would just adopt our highest tier Radar plan, but but the actual mechanics of that are at sign up, you want to know if your customer is good before you give them any access to any credits, you want to make sure they're good. At the time they pay, you want to make sure that charge is good. And anytime they have an overage, you want to make sure they're good for their money. And that it and you know, there there's other stuff around refunds and disputes that we also support, but I think those are the four kind of major moments in the AI company's customer life cycle um where we're just maniacally focused on protecting cuz that's where we're seeing um the biggest cost and and the fastest fraud growth. And at each point, that's just a call to the Radar API? Yes, correct. And what if I'm sitting here, which I am, doing, you know, millions of dollars a year in Stripe transactions, but I actually have no idea what my fraud rate is other than there's like that little there's that little thing where it's it's not even I don't even know if it's a necessarily a fraud rate. I think it's our our card chargeback. Anyway, our fraud rate is low enough as marked for me to not care about it. But I guess I don't really know if there's some amount of free trial fraud that I'm not totally understanding right now. So, what are the things I should be looking for to uh know if I should dig deeper and potentially do some sort of, for example, Radar integration? Yeah. So, so thing one is you can go to your Radar dashboard and see if you see anything that looks spurious there. If not, you can also ask the Radar assistant, which is in the dashboard. Um and as you're doing that, you can describe your business model. So, you can say like, you know, I have a high marginal cost business, in which case you know, you care more about certain types of fraud than others. Um but you can also just take a stab at integrating a funnel and and see how it performs. We we can certainly um you know, share with you based on backtesting what we think the big issues are. But the but the fastest way um to get a clean read is is just to integrate. Got it. So, I would So, I I don't think that we're integrated right now. So, I would just go look at go look at Radar and see and see if it does it does it say anything. I'm doing that right now. Uh just to it'd be really funny if I found that we had a ton of fraud that I didn't know about. We are at 0% fraud. How is that possible? Um \>\> know. Uh 0.02% early fraud warnings, total fraud rate 0.2%. So, we're I you know, we're doing pretty good, right? That's pretty low. That's pretty low. I mean, yeah, you're a pretty good human. Maybe the fraudsters don't want to come after you. I'm doing you this episode and then they'll be like, yeah. \>\> \[laughter\] \>\> Okay, that's really interesting. Okay, so that's that's fascinating. Um I I want to go back a second to the to the AI economy because one of the one of the things you said earlier is fraud is increasing overall on the internet. Um and that it's also it's increasing because the fraudsters have AI, but that it's also you all and everyone else on the side of good in the in the AI economy also has AI to defend against these sorts of attacks. I think you're getting a you're getting an interesting window into the arms race that I think is playing out in lot in in lots of different areas that have this kind of threat vector. Um a really simple one is cybersecurity. Um not just for like payments, but for hacking and stuff like that. But, there's there's all these other similar types of things where AI makes one part of the process much easier and then another part of the process has to use AI to compensate to to to catch up. So, how is that race going? What is that like? What are the early reports that you're that you're seeing feeling uh being in being in a race with AI armed fraudsters? I think the interesting thing about fraudsters is they don't really care about boundaries. They don't care about, you know, whether this transaction is processed on Stripe or off Stripe. They don't care about whether this transaction is on, you know, fiat or crypto, whether it's on a card network or a buy now, pay later. They're just going to figure out sort of how to work around the system um to get through. And so, one of the important levers, and I appreciate you calling us the good guys, one of the important levers I think the good guys have for winning is to be comprehensive. And you know, a simple example in our world sort of Stripe Radar used to only work for cards transactions. And then last year we added ACH and SEPA, right? So, other payment methods. Um but this year we've extended to all payment methods that have disputes. And we added crypto. And we added the Radar API. So, guess what? You can screen transactions even ones that aren't processed on Stripe, right? So, you can process on Worldpay or Adyen or whomever and through the Radar API get the same fraud signals. Similarly, and we haven't talked about agent at commerce yet, but as we've built out our agent at commerce suite, one of the new primitives we designed is the shared payment token, which allows agents to safely pass buyer credentials onto merchants for the merchants to process the transaction. And as part of those shared payment tokens, we pass over the Radar fraud score so that the merchant, again, whether or not they're processing on Stripe, can action them appropriately. You know, when it comes to fraud, we really see fraud defenses, fraud mitigation as a public good. And that allows us to invest disproportionately above and beyond the direct value to Stripe because protecting the internet is important for growing the the internet economy. So, I would say like overall, like yes, fraudsters have AI in their favor. Stripe looks at 2% of global GDP and is growing 34% year-on-year and sees a broader swath through our multi-processor solutions like the Radar API and so luckily not only do we have AI on our side just like they do but we also have data on our side um and the more comprehensive we've gone in our fraud protections I think um uh the more we've been able to kind of eke ahead. Now, that's not to say that we're not constantly surprised by um the new creative vectors they come up with but you know, you can have an agent every day or every hour taking a look at anomalous patterns on the Stripe network and identifying new vectors that are popping up across processors, across payment methods, across merchants um and burn them down pretty quickly. So, uh I'm I'm overall bullish uh but certainly not complacent. What about other parts of the AI or agent economy? So, we've talked a lot about fraud. What are the other things that you see as sort of having this bird's-eye view of of what's going on that people might not realize? Um I mean I think you know, the the AI economy is is broad. I think there is a set of horizontal model providers that have a very interesting view into where is AI being adopted and with what intensity um throughout the economy. Um there's a number of sort of vertical AI solutions. People like to call them wrappers and I say that not condescendingly just as in like it's not their models, it's someone else's models but they have um domain specific data and relationships and context and they're solving problems in you know, health care or architecture or whatever who have a pretty unique view into vertical level adoption of um of AI. Um but I guess I'd be curious like what you have in mind on who has the best the best horizontal view. You're asking me? Yeah. Well, I'm I you know, uh I want to know what it what it looks like on the payment side, but uh I imagine I imagine them the model companies have have the best one overall cuz they're that's where all the tokens are going. Yeah. Yeah, I think they see a lot of the tokens. I think the AI gateways also have a pretty unique perspective into you know, who's buying what from whom. Um you know, as I step back and look at the AI economy from the Stripe vantage point and and we see you know, who's buying what from whom for how much, who's retaining and churning their subscriptions. Um there's a few a few themes that stand out. One is just and I think people feel this intuitively, but not everyone has like seen it in the data. These AI companies are just growing from a revenue perspective faster than any previous cohort we've seen. Um I was looking at the top 100 AI companies on Stripe and the ones that reach 30 million in ARR get there in about 18 months, so a year and a half. And that is like three times faster than the top 100 SaaS companies from 2018 and and by the way, that's the 30 million number, but even if you look like how fast do they make it to 1 million ARR or 5 million ARR, they are scaling like orders of magnitude faster than high-performing SaaS companies from less than a decade ago. The second kind of meta trend is this like and you probably feel it as a consumer. I know I do. This like very fast iteration across monetization models, right? So traditional SaaS had a lot of you mentioned the seat had a lot of seat-based usage, you know, fixed monthly subscriptions. That made sense for them because they were being used by humans primarily and their marginal costs were basically zero, but you know, we've talked about the very real inference costs in the context of fraud, those also have very real implications for how you price. And so, usage-based billing um has become very important very quickly. Companies are metering, you know, tokens and API calls, but they're also metering workflows and they're metering outcomes. Kind of like whatever unit best reflects both the the customer value and the cost structure. Um and then they're charging with like very high precision, right? They literally want to know every event, how is it rated, and what's all the metadata that sits on that rated event. Um way more hybrid monetization models, right? So, I talked about subscriptions, but subscriptions aren't dead. They're just subscriptions with like usage overages or like prepaid credits that burn down or real-time top-ups, which gets to my comment earlier on um this non-payment abuse issue. Um and very kind of multi-dimensional pricing and monetization. Lovable is um a a really good example, right? So, they they use Stripe Billing for their initial launch, which was fairly simple subscriptions, sort of more traditional pricing, and allowed them to monetize very quickly. And then they added a bunch of products like Lovable Cloud or Lovable AI, and they moved with those into usage-based billing, right? So, customers are actually charged based on token consumption, but it's a hybrid model, so it's above a certain threshold. Um and that just, you know, helps companies like Lovable align revenue with usage and value and the actual cost of running the models. Um and in the limit, you know, we actually have a solution called token billing, which is underlying model costs change a lot, sometimes very quickly. Um and if you are a wrapper on top of someone else's LLM and your pricing doesn't keep pace, then basically your margins can disappear, right? So, uh you know, costs go up and your price stays where it is, then you're in the red. And so, token billing is just, "Hey, let's um in real time track and price to the cost of the underlying tokens with some markup as set by the business." Um and so, you know, Mesa and Ship and Lovable are are all examples of this kind of kind of infrastructure. I love all of these points. I want to go through them one by one. So, a big one that you're talking about is fast iteration across monetization, and it feels like there's this hyper experiment experimentation going on right now where people are like, "Well, we could charge we could charge per token. We could charge on a token basis. We could charge um per completed request." Like, I think Fin, the customer service platform, charges per case resolved, which has been a thing in customer service for a long time, but it feels like that it that could come for a lot more types of software as LLMs make it easy to tell, "Did we actually do the work to get paid?" What do you think is the If we're If we're going to pick one, there's a whole range of exploration going on, but if we're going to pick one new pricing model as the like, you know, if if last year's pricing model or last decade's pricing model was just straight-up per seat, what do you think is the new standard pricing model that that is starting to emerge from from the Stripe customers that you that you see? If you are buying the model. So, if if you're primarily a model provider, let's say your customer is primarily buying the model, I think you're metering tokens. Like in an API, like OpenAI API, Cloud API, yeah. Exactly. For these vertical solutions, I think in steady state, you are metering outcomes. But, it's going to take us some time to get there. Not because of the billing infrastructure, actually, that's totally ready. You mentioned the Fin example, Intercom does the same thing actually on Stripe billing. They they have an outcome-based uh meter for for support tickets resolved. Um why do I say for vertical solutions it's going to be on outcomes? Because I think end users are going to want to hold those vertical solutions accountable for outcomes, and they're going to want to know that they have positive ROI on their spend. Now, when you and I buy a model, we feel like we ourselves are accountable for the ROI that we get on the whole plethora, wide range of applications we might have for that LLM. But if you're a vertical provider, if you're really focused on like solving a concrete need in a given business domain on top of someone else's LLMs, it seems like the core value is sort of it's sort of on you to ensure the ROI is there, and I think outcome-based pricing is is the most efficient way to hit that. Now, I don't think all outcomes are created equal, and so you could imagine these like I'm an economist by training, so I'll be a little nerdy, but like you could imagine these like complex objective functions where it's not just did you resolve the support case, but how complicated was it and with what quality and like what was your CSAT, and you know, how expensive was the person that you were automating in that task, and so that's why I say in the limit, like I think it'll take time for us to be very crisp on the outcomes we care about and how we measure those outcomes, and those outcomes will be multidimensional. But I just have a hard time imagining you know, a year from now most vertical providers are literally charging on tokens. That's really interesting. I'm I'm very curious to see that because what what I felt, so I think you can see this a little bit in in the example in the LLM example you gave, but also in the Claude and ChatGPT examples and and some of the pricing that we've ended up doing is it's per seat, it's per user with overages. Um because we're we've started to exist in this world where we used to charge per seat, so people know how to model it. It's pretty easy to kind of figure out how much I'm going to pay, but software used to be free to run, and now it's not. And so we have to cover our butts, basically, protect our margin by adding the like the overage so that customers basically know what they're going to pay unless there's some sort of special special circumstance. Do you see that Where do you see that fitting in the examples that you gave and and and I guess I guess you would you would say eventually that might go away. I'm curious why. So I don't think the the charging for user charging for overages will go away for like most of like the model providers. If anything, I think that will dominate and the seat-based billing will go away. Like, you know, we can go back to like the Finn or Intercom example. Like you like you Well, you and I would think it's silly to charge based on number of customer service reps that were using the tool because obviously a lot of what the tool is doing is automating customer service reps. Um in today's world it isn't perceived as silly to do seat-based usage of developer tools, but I think it's a fair question since basically November or December to say, "Wait, why isn't that silly? That seems a little silly." Because, you know, if what these agents are doing is um you know, making every developer, let's say, you know, I don't know, 10x more productive at some point, then don't you need 1/10 the developers? And why would you want your revenue pegged to the count of developer as sort of the base price? So I suspect that we will see uh seat-based disappear. Um now, it in in the enterprise context. Um now, I think it's quite different in the in the consumer individual context. I think um you know, with the exception of maybe some nerds on the call, most people are actually pretty uncomfortable as individual consumers um with anything but a fixed fee monthly maybe with some overages if they want to spend like crazy. But in businesses I just I would be super surprised if six months from now we have half of the seat base licenses that we have today. That is fascinating. Well, we'll have to have you on again to talk about that one. I'm so curious to see and I would love to see more straight data coming out about that. One other thing that you brought up before this was your your also seeing these companies scale faster. Like you said that the time to get to 30 million in in ARR is like 18 months, which is significantly faster than any other cohort of companies you've seen. I'm curious where is that coming from? Like presumably the spend or the growth from their customers is coming from somewhere. Either it's spend that people weren't spending before it was like, you know, on the on a company balance sheet just waiting to be deployed or they're pulling it from an another provider and then going really rapidly into these new ones. Do you have a sense for what's happening here and where's all the why are they growing so much faster and where's all the money coming from? Yeah, so I think a lot of the AI growth that we've seen is actually like net new spend being pumped into the economy. I think it has largely not been a substitute for traditional SaaS or for sort of head count up X because it's been experimental, because people are still learning, because organizations are somewhat slow to drop existing licenses often because they're contracted into longer durations, but also because AI was starting not literally at zero, but at near zero. So like there weren't other AI companies to to sort of uh soak up the spend market share from. Yeah. Um I would say now going forward, um, I I I I don't have a crystal ball. I can't tell you exactly what the dynamics will be, but I expect that some of it will be a substitute away from traditional SaaS. Just And And by the way, I I I don't say that in a, you know, old company and new company sense. Like some SaaS companies are doing an amazing job reinventing themselves as AI first, and so you will have AI arms of traditional SaaS companies that are eating some of the revenue from the traditional version of the same company. But some will come from SaaS. I think some will come from headcount opex. Like I I is It is very hard to believe that companies will start spending single-digit, sometimes double-digit percentages of their headcount opex in LLMs and not step back and say, "Okay, well, my headcount opex My headcount cost just changed. You know, it used to cost me $300,000 for an engineer, and now it costs me $330,000 for an engineer cuz 300 of them is salary and equity, and 30K is LLMs, and so I better reason about my budget on the, you know, plus 10% basis and make headcount decisions accordingly." Um, and by the way, ROI decisions as well. Um, and then some of what we are seeing is definitely substitution now across AI providers. So I was looking at, um, retention rates for AI companies, and what you see is actually the within the domain. So for example, within AI dev tools or AI coding tools or AI model providers. Um, the retention rate, both B2C and B2B, is higher than it was for SaaS. Um, Interesting. I'm shocked. But but but within within Okay, got it. Got it. Yeah. But for the individual provider, it's slightly lower. Yeah. Which is intuitive. Like once you start Which is ex post intuitive. Although ex ante, I actually literally didn't know and needed to query the data. But ex post is intuitive. It's like once you start using, you know, AI dev tool, like a coding assistant, like you love it, you're not going to stop using it, but you very well may iterate across providers as, you know, models vary in their quality or um Anytime a new model comes out, you're just like I got to try this. And there's a there's a high percentage of um curious travelers basically just hopping from one thing to the next within a category. But they're definitely going to stick using a tool like that for a long time. Yes, exactly. And so I would say like a lot of the sort of crazy fast AI growth we've seen is like net new dollars spent, but I think businesses are going to start to reason about that as a substitute for SaaS or that as a substitute for head count opex or that as a substitute for other AI companies, and it will be less purely additive in the go-forward year than it was in the past year when you know, people were really just starting to ramp up on on their AI spend. \>\> Does that imply anything to you about the valuations of current, you know, hot AI companies? Like let's let's let's accept from this like the opening eyes and Anthropic's of the world, but like the in the 30 million cohort from of this cohort in the coming in the coming up ones, does that say anything to you about their prospects or their growth rates or their valuation? Well, so actually if you look at like the top 100 on Stripe, like there are little pockets of twos and threes that are directly competitive, but a bunch of them are like solving totally disjoint vertical problems with no competitor yet in the space. And so I do think there's like enough blue ocean sort of vertical solutions that I think overall AI valuations are probably okay. I think there's like a couple of crowded spaces that you and I could intuitively reason about. Where, you know, you might think it would be a little frothy. And by the way, you see this at that's sort of the macro view, but you see this in the micro view, too. Like, if you look at um sort of the the sales-led growth contracts, right? Like, when there's a new you know, when you were the first um uh AI dev tool, you basically charge people sticker and you do very little negotiations, and enterprise pay you sticker and whatever. And then all of a sudden you have to have these like much more complex um I mean, you hire a bunch of sellers and you have your, you know, CPQ configure price quote system and you have this nuanced billing because you're competing against two or three other providers who have like uh you know, competitive-looking monetization models and you're reacting to that. So, on the micro you start to see some of those uh some of those competitive reactions creeping in as well. But, I I think the I think the overarching kind of next year will continue to have a bunch of sort of blue ocean vertical stuff that didn't exist before. Um but, there will be some pockets where uh it's it's a little more it's a little more heated. Fascinating. I feel like I'm learning so much. This is amazing. Um I want to go into Stripe. Instead of talking about the AI economy, I want to go into Stripe a little bit. Specifically, Stripe is you know, it serves the you serve developers. And you're built for a world where humans are the ones buying and selling, and also humans humans are the the ones making the software. Um now agents are buyers, they're sellers, they're builders. Um and you're So, you have to sort of serve agents. And I'm curious how that has changed how you think about the products that you offer and the you know, moving maybe from just thinking about developer experience to agent experience, all that kind of stuff. Do you want to start with agent experience or agent I think they're both they're they're kind of different, but they're they're both really interesting. Which one are you most excited to talk about? Maybe agent experience and then we can work backwards to agent economy. \>\> Right. Yeah, let's talk about agent experience. Okay, so you know, the the developer story, well, so the the whole idea of developer experience is changing. And historically when I said developer experience, you thought like, hey, making it easier for a human engineer who's at a keyboard, right? So like you need clear APIs and you need better docs and you need less setup work and um all of that still matters, like it's not going anywhere, but I think the the developer is now sort of a a broader swath of persona, right? It could be um a non-technical founder who's just in, you know, Vercel or Replit like describing an app in plain language or it could be um a coding assistant who's like scaffolding an integration or it could be an agent who's like out trying to provision infrastructure on on a human's behalf. And so um I think it's it's less about just like, okay, how do we help a developer, human developer write code and more about how do we have a coherent and trustworthy product experience sort of end to end that acknowledges that at some moments the actor is a human, at some moments the actor is an agent, and at some moments the human the actor is a human working through an agent. And so um you see this shift in in some really concrete ways. Very simple example, um LLM traffic to Stripe docs is up 10x year over year. And that's just a useful signal that machines are becoming users of developer infrastructure, too, including Stripe's developer infrastructure. Um All right, human use of the uh Stripe docs. So human use of Stripe Docs is actually like flat to climbing. It's not like a straight substitute. I think there is just like more developer acti- developer activity happening and LLMs are growing dramatically within that share. That makes sense. Cool. Um I would also say that humans continue to check on the docs to sanity check what the agent is coming up with because your payments integration is actually like a pretty big decision that you're making. \>\> I will say, yeah, better humans than I are sanity checking, but I'm glad that someone is san- sanity checking. \>\> Are you Yoloing it? I'm I'm Yolo vibe coding my payment infrastructure. \>\> \[laughter\] \>\> Okay. Okay. Amazing. So, maybe maybe you're Yolo vibe coding, but even if you're vibe coding, um there's still an important step around provisioning like your your modern software stack and that is still very manual, right? So, like you as a human are still creating accounts across multiple services, um you're managing credentials, you're clicking through to do a lot of setup, you're probably bouncing between dashboards. And so, like the coding is getting easier a lot faster than the setup is getting easier. And that's actually the idea of Stripe projects, which we launched I don't know, maybe 2 weeks ago. It's basically like \>\> amazing. Tell people what that is. \>\> Yeah. Okay, if you want in, let me know. We can we can do that. \>\> I want it. I absolutely want it. Yeah. \>\> Okay. You're in. Check. Um I won't slack right now, but I'll slack right \[laughter\] after this and get you in. But but basically the idea of Stripe projects for those who haven't explored it is just like you or your agents um can go create and manage parts of your staff software stack right from the command line. And so, um you know, resources are provisioned in accounts you own and credentials sync back to your environment and so on. Um and one of the things that stood out, besides your enthusiasm for it, which I appreciate, is just how sort of overwhelming the interest has been in general from the ecosystem. So, we launched with like Vercel and Supabase, um PostHog there, Neon, RunLoop, um there's a bunch of great companies involved. But then immediately after launch, over a hundred other great companies reached out wanting to join, which I just think reinforces that like the friction is real. And you talked earlier about like you know, some things get easier with AI, but there's like some counter effect. You know, I think coding gets easier, but like code reviews become more burdensome cuz who's reviewing all the AI code? This is another example of like building gets easier, but you still kind of have to like provision everything. And so that's just an example of how we're building for this world of like the developer is no longer just a human. Got it. And then tell me about agentic commerce. Okay, so agentic commerce is a bit of an overloaded term. And I think a mistake that people make with agentic commerce is they jump straight to kind of the most extreme versions. So they hear the phrase and they think like some system that knows everything about me and decides what I need and like goes off and buys it for me. And then they're underwhelmed with the world we're actually in. Like maybe we get to that extreme eventually in some form, but we're not we're not there yet. I prefer to think about it as a spectrum and you know, I think that the the economic infrastructure you need is actually pretty similar no matter where you are in the spectrum. But the spectrum also like brings some realism to it. So at the at the first level, which is like AI is just removing friction from the internet we already have, right? So it helps you research and compare options and fill out some forms and narrow down your choices. But you the human are still making the decision. We're just making, you know, the agent is just making that that experience easier. Then you move to like, okay, search is descriptive, right? No more like blunt keywords and filters and such. It's like I got little kids, like I need a summer camp for my kids in this budget on these dates with this driving radius. And that's already a better commerce experience than like search plus filter, which is like, you know, knowably blunt. Um, then you get to sort of real delegation. And I think this is what most people would consider like the minimum viable bar for saying agentic commerce. So, like, I give some constraints. Like, I give some for camps, like some budget, some dates, some category, maybe a few preferences, and then the system goes and makes the purchases on my behalf. Um, but then there's a further out version, which is like the ambient version, right? I don't prompt anything, and the system knows me, and it knows my seasonal needs, and it knows, you know, summer camp planning is happening. Uh, that would be music to my ears. And And sort of that's the That's the most, um, futuristic thing. I think the point is that no matter where you are on that, like, um, what the internet, uh, needs for economic infrastructure starts to change, right? Even the earlier stages force a redesign of payments infrastructure in particular, because uh, the today model, the old model, again, human sitting in front of browser, creating account, choosing plan, filling out the forms, clicking purchase, entering card details. Um, not all those steps are happening anymore. Um, and so, you know, I I think there's sort of two worlds that I reason about preparing for. One is agent assisted buying, so I'm ultimately in charge, but the discovery and checkout and payment happen inside AI interfaces instead of on a merchant website. So, I'm not going to Nordstrom, I'm buying within Gemini or ChatGPT or, you know, Meta, like Facebook ads, whatever. Um, and what's challenging here is two things. One, the AI agent needs to be able to understand the merchant's products and prices and checkout flow, so that they can act on behalf of the consumer. And two, uh, as you and I talked about a bit at the top, trust can break down, right? As a consumer, I don't want to hand off my credentials to an agent, as a merchant, I don't want to let just every like bot through. I want to know is it like a good bot acting on behalf of a legitimate customer. So, the agent commerce protocol, which we we co-created with OpenAI, is just the shared technical language between AI systems and businesses. And it shows up across a lot of surfaces. Again, we built it with OpenAI, but Microsoft co-pilot uses it, Meta's in-ad shopping experience uses it. And how it works is is basically the merchant only has to integrate once with Stripe for their product catalogs, their prices, their checkout flows, and then they can literally from the dashboard turn themselves on through a whole host of agents and be exposed through those through those shopping experiences. Importantly, the merchant remains the merchant of record, and that part really matters. Like, businesses want access to these new storefronts, these new channels, but they don't want to give up the customer relationship, they don't want to give up control over trust or fraud. So, kind of category one is the human is still sort of leading the buying, but the agent is like facilitating the transaction. You could call it agent to commerce, you could call it facilitated commerce. And how does that actually work? So, is the experience something like I'm in ChatGPT and it says, "Hey, here's like a thing you might want to buy." And I can click checkout from OpenAI, and that's using that protocol to then go send my information to the merchant, and then send me back, "Hey, like here here's your your your thing is on the way." That's kind of that's what you're talking about? Exactly. That. So, it's like one-click checkout. Yeah, same thing. Like, you're in Facebook, you get an ad in Meta, let's say you do like a one-click checkout, and you know, one of the primitives that we built for this is the shared payment token uh or SPT and it's basically just it just lets your payment credentials be passed securely from the AI agent to the merchant. So, the merchant can process the transaction. And the merchant processing the transaction is important because that allows the merchant to remain the merchant of record. But, you know, you don't want your credentials viewed by the agent, which is you know, why it's why it's a token and your actual payment credentials. And the merchant needs to know that you and the agent are good, which is why as part of the shared payment token, we pass over a whole host of of fraud scores. And can I integrate this? Like so, we have a CLI, we have a bunch of software. Can I offer agentic checkout easily or it does it have to go through the OpenAIs and the Facebooks of the world? So, yes, you can. And I think one of the one of the premises here is, you know, just like today we haven't seen one model provider to rule them all or one model to rule them all. We don't think there's going to be one agentic shopping experience to rule them all. And merchants will literally break if they have to integrate with every single potential new storefront, right? When they integrated with the internet, they built their own storefront. And yeah, they iterated on it, but basically they built it once. But, if you tell them, "Hey, you need to build your storefront for, you know, agent startup shopping startup X and Fiat and Open AI and Meta", like their eyes are going to get, you know, bigger than their heads and they're not going to be able to handle it. And so, we really want to abstract away that complexity for businesses. Like we spent the last decade plus helping businesses sell wherever their customers are. And first that was like on their websites, and then it was in apps, and then it was through platforms and marketplaces. And actually some in person, too, with our terminal product. But now, like, you know, where are the consumers? Where are they wanting to buy? Increasingly through sort of AI tools and agentic flows. And so, similarly, we just want to make it really easy um for merchants to agnostically participate in those in those different storefronts. And again, they can choose where they want to sell. They can turn it on, little toggle in the dashboard, um but it's not a different integration, which is which is the whole idea of the protocol. And then how often is this happening? Like, what what's the volume of agent com agent agentic commerce right now? The volume of consumer commerce is still relatively small. Relatively small. Relatively small as a percentage of all of the commerce we see. Um but it is growing quickly, particularly for you know, what I would think of as commodities, right? So, what is the first thing that people are comfortable buying through agents? It's like things that are reasonably known, reasonably observable, not super high-priced. You know, I think when when people started buying online, like you didn't imagine that they were going to go online and buy like a $2,000 couch, right? Oh, wasn't a couch a thing you really want to know the quality and you want to sit in it. A mattress. Oh my god, these mattress companies that have blown up. You know, and it took time for them to build comfort, you know, making higher price purchases, making more quality dependent purchases. Um and so, today it's it's it's predominantly um commodities. Um you know, in a similar vein, I've tried uh you know, to to book a whole summer trip uh using uh using an agent, and I wasn't sufficiently satisfied with, you know, the family-of-four choice of flights and hotels and transportation and itinerary uh to be willing to to one-click buy it. But, the models will get better, the interfaces will get better, the experiences will get better. We're pretty agnostic to those. Like, we trust that they will evolve in a bunch of different and interesting ways. And sort of the the primitives that are needed underneath, the uh ability to expose your catalog, the shared payment token, the fraud protections, are pretty agnostic to those experiences. And so, um that's where we're we're hyperfocused for merchants. Hm. Give me an example of one of these commodities, and also what the order of magnitude we're talking about when we say it's relatively small. Um an example of a commodity would be like a Halloween costume. Got it. Agents are buying Halloween costumes? For themselves? Agents are buying Halloween costumes. That's really \>\> believe how many lazy parents there are in the world. I mean, I think the consumer side is interesting, too, right? Because we talked about what do businesses need, right? They need a They need a fast, easy way to safely expose their products, their prices, their inventory, their checkouts, uh understand fraud, and be in control of the relationship. From the consumer angle, the question's a little different, right? Like, even if I'm a lazy parent, I'm not so lazy that I'm willing to give someone my payment credentials, and, you know, let it rip. So, like, the question for me is, how do I safely let an agent buy on my behalf? And have you heard of Link? Yeah, I've used Link. Okay, amazing. So, Link is our consumer wallet. What do you use it for? Do you remember the first thing you used it for? \>\> I mean, I use it all the time. So, it's like everywhere. So, uh Amazing. Yeah, it's everywhere, right? It speeds up You wouldn't believe where it is. I I was like a I was uh getting soccer lessons for one of my kids, like, you know, from a local guy, and I was on their website, and they they only accepted Visa and Mastercard, neither of which I had, you know, on me, or um you know, direct debit from my bank account, which I wasn't going to put in this very janky website, or Link. AND I WAS LIKE, OH, AMAZING. Link is here. Problem solved. Anyway, a lot of people know about Link as our consumer wallet for buying soccer classes. It speeds up checkout, but it's also um So, and and it's already used by about a quarter of a billion consumers. So it's not a small network. But what I think is most interesting about Link is it's a very dense network when it comes to AI. So lovable is an interesting example. 58% of their payment volume runs through Link. You are hyper AI pilled. It is not surprising that everywhere you are Link is. And so what's changing now is that we're evolving Link for the AI economy because so many of the Link consumers are already AI consumers. And acknowledging that like agents themselves are becoming economic actors. And so the model isn't you know, give a random agent your card and hope for the best. Instead, it's delegated authority with guardrails. So, you know, you as the consumer decide which agents are allowed to request credentials and under what conditions and with what limits and whether those purchases require approvals before they go through and you do all of that through Link. And it's just a much more sensible model for for delegated for delegated purchases. That makes sense. Emily, this was a fantastic conversation. I learned so much. Awesome. Thank you for having me. \>\> \[music\] \>\> Oh my gosh, folks. You absolutely, positively have to smash that like button and subscribe to AI and I. Why? Because this show is the epitome of awesomeness. It's like finding a treasure chest in your backyard. But instead of gold, it's filled with pure, unadulterated knowledge bombs about chat GPT. Every episode is a roller coaster of emotions, insights, and laughter that will leave you on the edge of your seat craving for more. It's not just a show. It's a journey into the future with Dan Shipper as the captain of the spaceship. \>\> \[laughter\] \>\> So, do yourself a favor, hit like, smash subscribe, and strap in for the ride of your life. And now, without any further ado, let me just say, Dan, I'm absolutely, hopelessly in love with you.