Jul 9, 2026 · 1h 15m · mad
The "Token Heist" Wiping Out AI Startups | Emily Sands (Stripe)
gold bands on the timeline = statements, start to end. Hover to read, click to jump. CC turns on captions
In this episode of The MAD Podcast, Stripe's Head of Data and AI Emily Sands details the financial architecture powering agentic commerce, token monetization models, security defenses against token theft, and the future of autonomous economic agents.
How this conversation actually went
Every chapter scored 0–10 on four independent dynamics. Hover any point for the reasoning behind the score. Matt holds 13.1% of the talking time here. How this is scored →
speaking balance: gold is Matt, purple is the guest (3 minute bins)
Emily directly counters Matt's thesis that runaway token costs could cause an industry backlash, explaining that token costs are only a small single-digit fraction of headcount budget.
Hardest push from Matt ▶ 35:03 Matt highlights legal liability gapsMatt refuses to accept that technical protocols fully address transaction risks, pushing Emily on the inevitable court battles that will arise when autonomous agents make severe purchasing errors.
Biggest teaching moment ▶ 50:45 Emily reveals token theft statsEmily educates the host on how AI fraud has fundamentally shifted from stealing money to stealing tokens, dropping the startling metric that over 1 in 6 signups at AI startups are multi-account token abuse.
Matt holds his own ▶ 5:50 Matt introduces levels of autonomyMatt demonstrates high technical fluency by suggesting an autonomy framework (Level 1 to Level 5) modeled after self-driving cars, which Emily validates as Stripe's exact internal framing.
the scores for every segment, with the reasoning behind each
| Chapter | Topic | Matt as informed peer | Guest teaching | Guest disagreement | Matt pushing back | Why |
|---|---|---|---|---|---|---|
| The Evolution and Spectrum of Agentic Commerce | 2 | 6 | 0 | 0 | Matt opens by asking what has transitioned from theoretical to reality over the past year in agentic commerce. Emily educates the host on the full spectrum of agentic transactions, ranging from fully autonomous agent-to-agent protocols down to human-guided in-app buy buttons. | |
| Framing the Levels of Agent Autonomy in Commerce | 4 | 5 | 0 | 1 | Matt demonstrates industry knowledge by proposing an autonomy level framework akin to self-driving cars, which Emily confirms Stripe literally uses. Emily goes on to explain the adoption of the Agentic Commerce Protocol (ACP) co-developed with OpenAI. | |
| Economic Dynamics of Autonomous Agent-to-Agent Transactions | 3 | 7 | 0 | 0 | Matt asks about the ultimate vision for agent-to-agent negotiations and economic impact. Emily cites economic theory on transaction friction (Coase theorem) and Census Bureau data showing a surge in high-earning solopreneurs running AI-assisted businesses. | |
| Addressing Trust and Adoption Roadblocks in Financial AI | 4 | 5 | 0 | 1 | Matt inquires about adoption roadblocks, specifically trust versus technical limitations, contributing historical context about early consumer hesitation to input credit cards online. Emily explains how trust is built incrementally through repetitions and wallet guardrails. | |
| Evaluating User Experience Shortfalls in AI Shopping Apps | 2 | 5 | 0 | 0 | Matt asks for specific examples of where current AI shopping user experiences fall short. Emily explains that LLMs lack complete deterministic metadata regarding inventory levels and price parameters. | |
| The Link Wallet for Autonomous Agents and Consumer Controls | 3 | 6 | 0 | 1 | Matt asks how Link Wallet functions for agents and seeks clarification on whether it relies on virtual single-use credit cards. Emily explains the transition from scoped one-time virtual cards to consolidated multi-credential agent wallets. | |
| Shared Payment Tokens and Programmable Spending Rules | 3 | 5 | 0 | 1 | Matt prompts Emily on the mechanics of Shared Payment Tokens, double-clicking on programmable spending restrictions. Emily outlines how payment rules can restrict geography, spend caps, and merchant categories. | |
| Risk Management, Merchant of Record, and Transaction Safety | 3 | 6 | 0 | 2 | Matt presses on liability and legal recourse when autonomous purchases go wrong, questioning how courts will handle faulty agent actions. Emily explains design principles that keep businesses as the merchant of record and integrate real-time fraud scoring via Stripe Radar. | |
| Vibe Deployment and the Launch of Stripe Projects CLI | 3 | 7 | 0 | 0 | Matt asks Emily to define 'vibe deployment.' Emily presents striking internal data revealing that 40% of Stripe documentation traffic and 70% of command-line interface requests are generated directly by AI agents. | |
| Token Monetization and Usage-Based Billing Models | 3 | 6 | 0 | 1 | Matt asks whether AI startups have completely abandoned seat-based pricing for usage-based billing. Emily explains how non-zero marginal costs force AI companies into usage meters or hybrid subscription-plus-credit models. | |
| Streaming Real-Time Payments and Metronome Accounting | 2 | 6 | 0 | 0 | Matt asks how billing infrastructure changes when servicing autonomous agents. Emily details real-time metering via Metronome and streaming payments via Tempo, noting the emergence of hybrid software-engineer accountants. | |
| Combating Token Theft and Fraud Across AI Lifecycles | 3 | 8 | 0 | 1 | Matt asks Emily to elaborate on 'token theft' as a novel fraud vector. Emily educates the host on how fraudsters target compute tokens instead of cash, revealing that over 1 in 6 AI company signups involve multi-account abuse. | |
| Machine Payments Protocol and Streaming Stablecoin Payments | 2 | 5 | 0 | 0 | Matt invites Emily to explain the Machine Payments Protocol (MPP). Emily outlines the protocol's machine-readable handshake between autonomous agents and API providers using stablecoins. | |
| Macro Trends and Growth in AI Startups | 2 | 7 | 0 | 0 | Matt asks for broader growth trends in AI startups. Emily shares that startups from the 2026 Stripe Atlas cohort are tracking to five times the revenue of the previous year's cohort at the same age due to day-one international expansion. | |
| Managing Token Costs and AI Spend Efficiency | 4 | 5 | 1 | 2 | Matt challenges whether unexpected token costs could trigger an enterprise backlash against AI spend. Emily gently reframes the issue, explaining that token costs represent a manageable single-digit percentage of headcount expense that well-managed firms recalibrate via model routing. | |
| Predictions for AI Commerce and Autonomous Agents | 2 | 6 | 0 | 0 | Matt asks for predictions on where agentic commerce will land in 12 months. Emily outlines a vision where agents evolve into complete micro-firms that autonomously provision infrastructure, purchase APIs, and sell services. |