Podcast, Interview, Fireside Chat
What Happens When AI Agents Can Spend Your Money?
- Shift in Agent Capabilities: AI agents have transitioned in the last three months from "spectators" to "actors" capable of executing tasks (e.g., booking a piano teacher) by utilizing payment credentials as functional "arms and legs."
- "Wallet Wars": A new competitive landscape has emerged where digital wallets (e.g., Stripe Link, ShopPay, PayPal) race to connect with AI agents, prioritizing two factors: the volume of stored consumer credentials and the breadth of merchant acceptance.
- Stripe Link connects ~300 million consumers to ~1 million merchants.
- ShopPay serves a similar consumer base but reaches ~1.7 million merchants.
- PayPal holds >400 million accounts across ~30 million merchants.
- Infrastructure Scale: VGS currently secures payment credentials for approximately one-third of all e-commerce Gross Merchandise Value (GMV), serving the largest online merchants, neobanks, and frontier AI labs.
- Vault Evolution: VGS is evolving from a "single-player" vault (securing credentials for subscription billing) to a "multiplayer" infrastructure that handles complex credential routing across multiple payment processors (Chase, Stripe, Adyen, etc.) for a single merchant.
- Network Tokenization: The industry has shifted from storing raw 16-digit Primary Account Numbers (PANs) to using domain-bound network tokens from Visa, MasterCard, Amex, and Discover, which reduce risk by rendering tokens unusable outside specific merchant contexts.
- Credential Maintenance: VGS automates the maintenance of ~1 billion stored credentials, addressing the annual 25% turnover rate caused by card expirations (approx. 10%) and network switches (e.g., Capital One moving from Visa to Discover) or fraud.
- Agentic Token Standards: New "agentic tokens" are being introduced by card networks to allow safe, verified handoffs of credentials from wallets to AI agents while preserving chargeback rights and merchant-consumer relationships.
- VGS is currently the largest provisioner of these tokens, processing millions weekly, up from zero three months ago.
- These standards enable agents to make purchases where the transaction appears as the consumer's original card, preserving eligibility for merchant loyalty programs and follow-on transactions.
- Adoption Friction Models:
- Walled Garden Approach: Stripe currently employs a "Stripe acts as merchant" model (using virtual cards) when agents interact with non-integrated merchants (e.g., buying chandeliers at Walmart via Stripe Link), which can sever direct merchant-consumer data links and complicate chargebacks.
- Open Standards Approach: The industry is moving toward open standards where the agent uses a tokenized version of the consumer's actual card to ensure the merchant sees the true customer for relationship building and dispute resolution.
- Merchant Stances on Agentic Commerce:
- Antagonistic: Amazon recently blocked the Muse agent, largely due to a business model reliant on selling user "eyeballs" (advertising) rather than direct transaction facilitation.
- Receptive: Walmart and Shopify merchants are actively embracing agents as vital sales channels; Walmart has not blocked agents, and small businesses view them as a necessary expansion.
- Liability and Fraud Risks: New challenges in "friendly fraud" and "hallucination liability" are emerging as consumers delegate authority to AI agents to spend money, complicating the traditional three-step transaction validation (card validity, ownership verification, and user soundness of mind).
- Economic Impact Predictions:
- Underhyped Demand: Agents will likely unlock significant GDP by automating repetitive purchases consumers forget (e.g., cycling gear, subscription renewals) or canceling forgotten trials.
- Overhyped Volatility: Significant volatility in financial services (e.g., automated bank runs or massive yield arbitrage between accounts) is unlikely due to friction in bill-pay integrations and user habits.
- Strategic Evolution for VGS: The company is advancing through three overlapping eras:
- The Vault Era: Establishing itself as the most trusted, PCI-compliant storage for credentials.
- The Maintenance Era: Managing complex network token updates and multi-processor routing.
- The Network Era: Brokering credentials between wallets, agents, and merchants to create network effects and ecosystem value.
- Reward System Disruption: The rise of AI intermediation threatens the integrity of current credit card rewards models (e.g., category-specific points) as agents optimize transactions for price rather than maximizing points, potentially necessitating new terms of service and reward structures.
- Asymmetry Warning: The current advantage AI agents hold over merchants (e.g., securing hard-to-get reservations or bypassing complex terms) relies on asymmetry; this advantage will diminish as merchants deploy their own counter-agents to enforce terms and negotiate.