Claudia spent 500+ hours across Latin America and found that the crypto payments narrative is fundamentally wrong. Crypto cards peaked—QR-based payments like Brazil's Pix (6B+ monthly transactions) and India's UPI are the structural winners, not card networks. The real opportunity isn't single-corridor dominance but cross-border scaling; stablecoin on/off-ramp margins are collapsing from 1.5-2% in 2023 to 0.3-0.8% in 2025, so winners will compete on wallets, cards, yield, and brand layered on top, not the ramps themselves.

Artemis argues agentic commerce is Coinbase's most underwritten thesis. As AI agents become primary internet users—Cloudflare expects bot traffic to surpass human traffic by 2027—they'll need programmable money for high-frequency, micropayment transactions that card rails can't serve economically. Coinbase uniquely owns the full stack: USDC stablecoin, Base settlement layer, and x402 payment protocol. x402 has processed 178.7 million agentic payments worth $42.4 million since October 2025, with 82.1% settling on Base and 99.8% in USDC. If agentic commerce reaches $7.5 trillion annually by 2031 with 20% on stablecoin rails, Coinbase could capture $4.25 billion in annual revenue—positioning it as financial infrastructure for AI-native finance, not just a crypto exchange.

6-month ground-truth piece across Brazil, Mexico, Argentina, Colombia, Peru. Most fintech LATAM decks get the corridors, the user, and the product all wrong. Eight findings: **(1) Mexico is plateauing, Central America is exploding.** Total LATAM remittances hit $174B in 2025 — but Mexico fell 4.5% (first time in 11 years) while Guatemala +15%, Honduras +19%, El Salvador +18%. Driven by deportation-risk panic-sending. The unfought territory: non-US corridors (Venezuela→Colombia, Spain→Ecuador, Argentina→Bolivia) — barely served by US-licensed MTOs. **(2) Wrong customer.** Actual user is 40-60yo, sends $131-648/month (6-23% of income), 80% goes to groceries, half send to mom. Not a 25yo crypto trader. Trust > features. WhatsApp + mobile-first beats web every time. **(3) The stablecoin balance IS the product, not the transaction.** Argentina is full digital dollarization (USDT+USDC = >70% of crypto purchases). Brazil at ~90% of crypto volume is stablecoin-tied. Colombia at ~52% (driven by peso depreciation + Colombia's $5K minimum on USD bank accounts). Users want to *hold* dollars, not transit them. Three problems they're solving: inflation hedge, capital controls, cheap cross-border. The transaction is a side effect. **(4) Western Union collapsed, only Remitly is winning so far.** US-LAC share 2020→2024: WU 29%→17%, Remitly 14%→23%, MoneyGram flat. Bitso processes ~10% of US-Mexico flow on stablecoin rails. Felix Pago has done $1B+ via USDC-to-SPEI through WhatsApp. **(5) Cost wedge.** Banks lose 3-5% to FX spread. Crypto rails compress total cost <2%. For a $300/month sender, that's a month of groceries per year. Worst legacy economics = where stablecoin disruption hits first (Venezuela went P2P-stablecoin years before any regulation). **(6) Regulatory map.** Colombia + Argentina first (faster path), Brazil + Mexico in parallel via licensed local partners, Venezuela via P2P stablecoin already happening organically. **The biggest 2025 regulatory shift is the US 1% remittance tax** — passed summer 2025, hits roughly half of all senders, digital + crypto exempt. Single biggest stablecoin-rail tailwind in a decade, handed to the industry by US policy. **(7) Winning stack** = local rails (Pix/SPEI/PSE/CVU) + stablecoin liquidity + card layer + earn layer (USDC at 4-6% beats every regional savings account) + dead-simple UX. Closed loop: on-ramp → remit → recipient holds USDC or off-ramps → spends via card or earns yield. Banks can't do this. MTOs can't. Pure crypto exchanges can't. Pure neobanks can't. **(8) Three things every team gets wrong:** treating LATAM as one market (each country needs different licenses/rails/stablecoins), debating whether stablecoin adoption will happen (it already did), under-marketing on trust (a marketing problem, not engineering).