The Money Stack Gets a New Customer: Machines

The wave right now

The last fintech cycle was about moving humans' money cheaper and faster. This one is about two brand-new counterparties: autonomous AI agents that need to spend, get insured, and be trusted — and prediction markets that suddenly have real venues, real volume, and zero institutional plumbing. Layer on a third, quieter shift — AI eating the analyst and the back-office underwriter — and you get the cleanest read of the 2025–2026 YC fintech batches: the money stack is being rebuilt for non-human users.

If you're a founder picking a lane this quarter, the question is not "which consumer fintech is left." It's "which piece of machine-money infrastructure will look obvious in 2028." If you're an angel, the asymmetry is in the rails, not the apps.

The landscape today

1. Payment credentials for agents — the new card-issuing land grab

The sharpest cluster is agent spending controls. Agentcard issues disposable debit cards AI agents can spend with, one approval per charge; Allowance does the same job as scoped, one-time card numbers with enforceable spending rules. These are nearly identical wedges, two batches apart — which tells you the program believes this is a category, not a feature. The honest critique: both are a thin layer on existing card networks today. The winner will be whoever turns "safe agent spend" into the policy and audit layer that enterprises mandate, because the card primitive itself will commoditize fast.

S26

Agentcard

debit cards for AI agents.

ActiveFintech

P26

Allowance

Scoped payment credentials for AI agents

ActiveAI Agent Payment Credentials1 ppl

2. Underwriting the machines — agent liability is a real insurance line now

Where agents spend, agents break things, and a parallel cluster is racing to underwrite that. Mount pairs vulnerability scanning with AI-specific coverage, positioning as security-plus-carrier; Klaimee sells liability cover and certification for autonomous agents outright. Both lean on first-mover loss data as the moat — and they're right that loss history is the only durable asset in insurance. The adjacent, less glamorous bet is AI-native distribution in boring commercial lines: Panta is an AI brokerage for construction and logistics, Hedge places hard-to-find E&S specialty risk. I'd rather back the boring ones near-term: premium exists today, and the AI-agent liability market still needs a few public disasters before buyers show up with budget.

P26

Mount

The AI Insurance Carrier

ActiveAI Agent Liability Insurance2 ppl

P26

Klaimee

Liability insurance for AI Agents. You deploy agents, we cover you.

ActiveAI Liability Insurance2 ppl

W26

Panta

AI Native Commercial Insurance Brokerage

ActiveAI Commercial Insurance Brokerage3 ppl

P26

Hedge

AI-Native specialty insurance company

ActiveSpecialty Insurance AI2 ppl

3. Prediction markets get their prime brokers

Kalshi and Polymarket created volume; this cohort is building everything around it. River Markets is a prime brokerage with cross-venue execution and unified risk; Valence aggregates venues with arbitrage detection; Sequence Markets routes low-latency execution across crypto, prediction, and tokenized assets; ValCtrl is building the search-and-taxonomy layer over fragmented markets. This is the most classically attractive pattern in the memo — fragmented venues plus professional demand always produces a brokerage winner — and also the most crowded inside a single program. Expect consolidation; the one that signs real trading desks first eats the rest.

P26

River Markets

Prime brokerage for prediction markets

ActivePrediction Market Brokerage2 ppl

W26

Valence

Unified prediction markets trading platform

ActivePrediction Market Trading Infrastructure3 ppl

W26

Sequence Markets

Low-latency execution across crypto, prediction, and tokenized assets

ActiveCross-Venue Trading Infrastructure5 ppl

4. The analyst seat and the loan file go autonomous

The fourth pattern is AI absorbing financial labor outright. On the buy side, Kimpton AI generates portfolio-personalized trade proposals for funds, while Standard Signal skips selling software entirely and runs as an AI-native hedge fund — the maximalist version of the same thesis. On the credit side, Zolvo automates servicing for factoring and ABL lenders, and Kita automates underwriting memos for emerging-market lenders. The credit-ops pair is underrated: it sells into cost pressure that exists regardless of AI sentiment, and the workflows (reconciliation, fraud checks, credit memos) are exactly the document-heavy drudgery LLMs are already good at.

P26

Kimpton AI

AI-Native Investment Research

ActiveAI Portfolio Research Platform5 ppl

P26

Standard Signal

Hedge fund where AI researches and executes every trade

ActiveAI-Native Hedge Fund1 ppl

P26

Zolvo

AI that automates servicing for commercial lenders

ActiveCommercial Loan Servicing AI4 ppl

W26

Kita

Automate credit review for lenders in emerging markets

ActiveEmerging Markets Credit Automation2 ppl

The cohort signal

The batch pattern is unambiguous. Winter 2026 seeded prediction-market infra (Valence, Sequence Markets) and credit ops (Kita, Zolvo following in Spring). Spring 2026 then went all-in: three prediction-market plays (River Markets, ValCtrl, Totalis), agent payments (Allowance), and two agent-insurance bets (Mount, Klaimee) in a single batch. Summer 2026 opens with Agentcard doubling the agent-payments bet. That's acceleration, not noise — same-thesis companies funded in consecutive batches means the program is deliberately portfolio-betting these categories. Also notable: an unusual share of Spring 2026 fintechs carry rename/pivot flags (Mount, Hedge, River Markets, Totalis) — founders are actively steering *into* these themes mid-batch, which is itself a demand signal.

Lessons from the last cycle

The veterans rhyme with today's clusters. Stripe won by being the developer-first rail when e-commerce was the new counterparty — the exact playbook agent-payments founders are rerunning, and the reason the policy layer matters more than the card. Coinbase proved that being the compliant on-ramp to a chaotic new asset class is worth a public listing — the prediction-market brokers should study it closely. Brex and Newfront both ended in acquisitions after pivoting their wedges (Brex literally renamed its way to corporate cards), a reminder that fintech wedges are disposable but the license, the loss data, and the distribution relationship are not. And Truebill's acquisition is the ceiling check for consumer money apps: a good outcome, rarely a category-defining one.

If you're building here

Openings I'd take this quarter:

• The agent-spend policy and audit layer. Agentcard and Allowance prove demand for the card primitive; nobody yet owns enterprise-grade authorization, attribution, and dispute tooling for agent-initiated transactions. That's the Stripe-shaped seat.
• Lending back-office, vertical by vertical. Zolvo (factoring/ABL) and Kita (emerging-market credit) each picked one lane and left a dozen open — equipment finance, SBA, CRE servicing. Real revenue, no hype dependency.
• Risk data for agent insurance. Before Mount or Klaimee can price anything well, someone has to standardize agent incident and loss telemetry. Selling that data to carriers is the safer side of the same trade.

Tarpits to avoid by name: another prediction-market terminal — with River Markets, Valence, and Sequence Markets already funded across two consecutive batches, the execution-aggregation seat is full. Likewise consumer AI money apps: Gravy and Uno Wallet are charming, but the last cycle's consumer outcomes ended at Truebill-sized exits, and distribution costs haven't gotten cheaper.

What you'd have to believe: that agents transacting autonomously becomes default behavior within ~24 months, and that prediction markets keep their regulatory green light. If either stalls, the infra cohort is early, not wrong — but your seed runway has to survive "early." Price accordingly.