A Two-Person Startup Just Became the Default Payment App on Every iPhone

The Pitch: An Agent That Decides How You Pay

Uno Wallet, a two-person startup from Y Combinator's Spring 2026 batch, raised a $500,000 pre-seed in June and secured Apple's approval as the first third-party wallet eligible to serve as the default contactless payment option on iPhone. The funding settled on-chain in minutes via USDC stablecoins, part of YC's new policy disbursing half a million to every S26 founder, a signal that the accelerator now treats stablecoin treasuries as the default for its newest cohort. The product connects your credit cards and routes each transaction to the one offering the highest rewards for that specific merchant, category, and spending pattern. No manual switching. No mental math. The app reads the merchant, evaluates your card portfolio, and selects the optimal card in real time.

That approval changes the competitive geometry. Apple Pay and Google Pay have always been passive containers; they hold credentials and authenticate the tap. Uno inserts an intelligence layer between the credential and the transaction. The company describes it as "the system that decides how people pay." The distinction is structural: today's wallets are storage. Uno wants to be the router.

The market is large enough to sustain the bet. Research firms converge on a trajectory: the smart wallet market will more than double by the early 2030s. North America drives roughly a third of revenue. The numbers differ by methodology, but the direction is clear — the wallet is becoming a software surface, not a leather good.

Shadpour and Joshi met at a Harvard startup mixer in 2023 and stayed in touch as friends and idea sparring partners. They concluded that advances in large-language-model reasoning and new iOS entitlements for contactless payment created a window that didn't exist two years ago. Joshi brought large-scale machine learning experience from xAI, where he built recommendation systems at scale before leaving to co-found Uno. That pedigree, and the specific technical challenge of real-time card selection at the point of sale, is why the founding engineer role they're hiring for sits at the intersection of consumer product, fintech rails, and model deployment.

The pitch is not rewards optimization alone. It is the precedent: a consumer-facing AI agent that acts autonomously on a high-stakes financial decision, thousands of times a year, with no human in the loop. If that works for credit cards, the same architecture can route across debit, buy-now-pay-later, stablecoin, and eventually cross-border rails. The wallet becomes the control plane for how money moves.

What the Founding Engineer Must Build

The job description for Uno Wallet's founding engineer reads less like a typical mobile role and more like a systems challenge: build the iPhone app, the merchant prediction engines, the category prediction layer, and the backend data infrastructure, all from prototype through production. The salary band signals the scope. This is not a narrow mobile, backend, or ML role. The engineer works directly with co-founders Saket Joshi, formerly a Member of Technical Staff at xAI building recommendation systems at scale, and Mordi Shadpour, a former Bloomberg Terminal expert, and owns whether the product works.

The core problem is deceptively simple: at the moment a user taps to pay, the system must understand the merchant, the purchase context, the rewards structures across every enrolled card, any active offers, and the user's own goals, then choose the optimal card in milliseconds. Wallets today store cards. Uno decides which one to use. That decision pipeline starts with merchant categorization, a problem that has defeated rule-based approaches for years. Payment processing research shows a merchant's self-reported category is often unreliable; high-risk merchants routinely misreport to avoid higher fees. The multi-modal approach, combining merchant time-series data with merchant-merchant affinity signals, has become the standard for verifying business type. Uno's founding engineer inherits this problem: build a prediction engine that identifies the true category of a coffee shop, a gas station, or a foreign marketplace before the NFC handshake completes.

Reward optimization compounds the difficulty. Each card carries rotating categories, spending caps, quarterly enrollment requirements, and issuer-specific offers that change without notice. The engine must model these constraints alongside the user's preferences, including cash back versus miles, annual fee justification, and credit utilization targets, and output a single decision in milliseconds. Messy real-world data and ambiguous problems must turn into reliable product behavior. The founding engineer builds the data infrastructure that ingests issuer feeds, parses unstructured terms, and maintains a live rewards graph for every card in the user's wallet.

Mobile integration adds a hard constraint: the intelligence layer sits inside an iPhone app that must interact with Apple's Secure Element and StoreKit framework. The Core OS layer governs hardware access; the Media layer handles the tap animation. Any latency in the prediction engine shows up as a stalled checkout. The job description explicitly calls for "real-time payment decisions that must be fast, accurate, and reliable," requiring high uptime and resilient routing in payments engineering. Industry data shows a one percent conversion lift came from replacing "transaction failed" with "this card is expired, tap to update." Error handling becomes product design.

Smart routing, the industry term for directing a transaction to the processor with the highest success probability at the lowest cost, is where the intelligence layer meets the rails. Open-source projects like payment_routing_optimizer frame it as a cost-optimization problem with predictive modeling. The founding engineer builds the routing engine that weighs rewards optimization against routing costs.

Reconciliation closes the loop. When a payment orchestrator goes down mid-transaction (one part succeeded, the other failed), the system must salvage the state and ensure what the app shows matches what settles in the bank. That requires idempotency keys, distributed tracing, and an incident response playbook that distinguishes internal failures from upstream outages. The founding engineer owns the monitoring that catches the issue before users do.

The stack is iPhone-first, but the architecture is a payment intelligence layer designed to outlive the form factor. The founding engineer decides what matters, ships quickly, and helps shape the engineering culture for the team that follows. The first hire builds the foundation; the next ten scale it. That technical scope is why the role sits at that intersection — and why it's drawing talent from frontier AI labs.

Why Elite AI Talent Is Leaving Labs for Startups

AI-engineer headcount in the United States rose 13 percent year-over-year in the first half of 2026, while the same metric in Southeast Asia fell 4 percent — the first major cross-regional migration of AI talent since 2023. The largest currents run between the major labs (OpenAI, Anthropic, Google DeepMind, Meta AI, xAI) and from big-tech research divisions into smaller specialized labs and startups. The smallest current runs the other way: from labs to traditional enterprise tech companies.

Compensation and equity lead the decision matrix, followed by mission alignment and research freedom, in roughly that order. At xAI, Joshi built large-scale machine learning systems on GPU clusters. At Uno, he applies that scale mindset to a different problem: real-time merchant categorization, rewards optimization, and on-device inference that must complete in the few hundred milliseconds between a tap and a settlement signal.

The Y Combinator Spring 2026 batch, roughly 196 companies, showcased the trend. Some of the hottest startups commanded valuations before Demo Day even ended. TechCrunch's data shows that some of the hottest startups in the batch commanded valuations north of $175 million, according to VCs who spoke to the publication. Uno Wallet raised its pre-seed under the new USDC policy, which settled treasuries in minutes instead of days, removing cross-border banking friction for international founders and giving every team an on-chain capital record from day one.

The founding-engineer role Uno posted on Work at a Startup asks for someone who has "built technically difficult products that real users depended on" and is "exceptional in at least one of mobile, backend, or ML." Ownership at the earliest stage is framed as the primary incentive. That pitch aligns with migration drivers: compensation, equity, mission alignment, and research freedom.

Joshi's LinkedIn notes emphasize "speed of iteration, strong evaluation frameworks, and attention to detail" as critical when building modern AI systems. The same discipline that trains a 100-billion-parameter model now trains a routing engine that decides whether a $4 coffee earns 2x on a dining card or 3x on a grocery card — and does it before the NFC field collapses. The pipeline is self-reinforcing. Every ex-lab founder who ships a consumer product becomes a proof point for the next wave. The next wave is already interviewing. The talent flow matters because the incumbents control the rails these agents must run on.

Can Apple and Google Counter the AI Wallet?

Apple Pay and Google Pay sit atop a massive annual flow, 56 percent of global e-commerce spending and a third of point-of-sale transactions, the Worldpay Global Payments Report 2026 shows. Apple Pay alone processed trillions in 2025 and is on track to exceed that in 2026, with over 700 million active users across 80-plus markets. Google Pay claims more than 500 million active users and significant volume. In the United States, Apple Pay holds about half of mobile-wallet users and trillions in 2025 transaction volume. Analysts project combined wallet volumes growing substantially by 2028.

Yet both platforms remain, at their core, credential relays. They store tokenized card numbers and pass transaction requests to the same Visa and Mastercard rails that have intermediated payments for decades. Apple extracts 15 basis points (0.15 percent) of interchange on every U.S. credit-card tap and a flat half-cent on debit — a rent the U.S. Department of Justice cited in its March 2024 antitrust suit and a $2 billion class action followed in January 2026. Google takes no such cut but ties its wallet to the Android ecosystem, which carries 70 percent global smartphone share.

The threat from AI-driven wallets such as Uno Wallet is not volume — it is intelligence. Today's incumbents let users choose a card at checkout. An AI wallet chooses for them, evaluating merchant category, rewards structure, spending patterns, and real-time offers in the milliseconds before the NFC handshake completes. That shifts the decision layer from the consumer's thumb to a model running on the device or in the cloud. If the model consistently picks the card that maximizes cash back or miles, the wallet becomes the primary financial interface, not the card issuer's app.

Regulatory pressure is already prying open the distribution moat. The European Commission accepted Apple's DMA commitments in July 2024: NFC access in Host Card Emulation mode, free of charge for ten years, with full Face ID, Touch ID, and auto-launch parity. Vipps MobilePay went live in the Nordics in December 2024, processing over one million taps. PayPal launched tap-to-pay on iPhone in Germany in May 2025; Curve followed across the EEA the same month. Banks can now bypass Apple Pay entirely in the EU, saving the 15-basis-point fee, and Visa is actively supporting such projects, including a partnership with BBVA in Spain ahead of the June 1, 2026 full-compliance deadline.

The incumbents are not standing still. Apple posted a Head of Financial Product Strategy role on August 26, 2026 explicitly seeking stablecoin expertise, signaling a move toward holding value directly rather than relaying credentials. Samsung reportedly plans stablecoin support for Samsung Wallet. Google Pay has been redesigned as a comprehensive financial platform integrating loyalty cards, boarding passes, and digital IDs, a step toward the "super app" model Alipay and WeChat Pay perfected in China, where they control over 90 percent of mobile payments with 1.4 billion and 935 million monthly active users respectively.

Agentic commerce adds a second front. Stripe's holdback experiment across trillions in annual processing found merchants offering Apple Pay saw a 22.3 percent conversion lift and 22.5 percent revenue lift; surfacing it via Express Checkout doubled that lift. But the same data shows tokenized transactions already cut fraud 30 percent and lift authorization rates 4.6 percent. The next leap is AI agents initiating payments on users' behalf, selecting the optimal rail (card, A2A, stablecoin), negotiating discounts, managing subscriptions. The Worldpay report's figures put digital wallets at 63 percent of e-commerce and 46–48 percent of POS by 2030, representing trillions at the point of sale alone. Whoever owns the agent that decides how to pay captures the economics of that decision.

Apple and Google have the distribution, the biometric trust anchor, and the tokenization infrastructure. What they lack is a business model that rewards optimizing the user's spend rather than taxing it. Until that incentive aligns, the AI wallet startups, lean, unencumbered by legacy interchange revenue, and staffed by engineers who built large-scale ML systems at xAI and DeepMind, will keep chipping away at the decision layer. The giants' counter-move will likely be acquisition or rapid internal replication of the routing engine, but the DMA-mandated NFC openness means they can no longer block the alternative from reaching the home screen. But distribution alone doesn't guarantee trust.

Trust, Security, and the Regulatory Gauntlet

The Houston Police Officers' Union bought four $100 Visa gift cards at a Kroger in Texas in August 2023. When members tried to redeem them, each card held $2. Thieves had recorded the numbers off the rack, waited for activation, and drained the balances. The union sued Kroger, arguing the retailer knew gift-card fraud was rampant and failed to secure the display. The case illustrates a baseline problem for any wallet startup: consumers already distrust plastic. Digital wallets now handle over half of global e-commerce transactions, Digipay.guru's December 2025 analysis shows, but the same fraud vectors (compromised credentials, SIM swaps, one-time-password interception) have migrated to mobile. Statista projects 2.5 billion mobile payment users worldwide as of 2026. Each new user is a target.

AI has lowered the cost of attack. Investigations cited by Convergence Now in May 2026 show fraudsters using voice cloning, synthetic identities, and automated tooling to hit users at scale. Personal data from leaks and unsecured databases feeds AI models that build detailed victim profiles. Traditional rule-based filters cannot keep pace with adaptive patterns; the same research notes that AI-powered edge detection is becoming essential for fraud management. For any wallet routing transactions through a real-time decision engine, the model itself becomes an attack surface: adversarial poisoning of merchant categorization or reward logic could steer spend to compromised cards or malicious merchants. The research emphasizes that fraud networks now combine stolen data with AI-generated inputs to bypass verification checks, a layered threat that demands continuous model monitoring, not just static rules.

Data privacy compounds the challenge. Uno Wallet must ingest transaction histories, merchant categories, and spending patterns to optimize card selection. That dataset is a magnet for regulators. Apple and Google updated their app-store policies throughout 2025, adding requirements for privacy manifests, data-deletion timelines, and alternative payment pathways. Appscalelab's 2026 readiness guide flags 48-hour data-deletion mandates and new Google Play Data Safety disclosures. A wallet that stores raw card numbers on device or in the cloud expands its PCI DSS scope dramatically. Appscalelab's August 2026 guidance recommends tokenization as the baseline: replace card numbers with encrypted tokens so a breach yields only useless tokens. Smaller teams can offload PCI burden to a fully compliant gateway that never touches raw PANs, but that adds vendor risk and latency — both fatal to a sub-second routing promise.

Regulatory frameworks are tightening in parallel. The North America payments market faces strict AML and PCI DSS mandates that disproportionately burden smaller providers. India's Cyber Crime Coordination Centre has scaled its 1930 helpline and reporting portals; authorities there are pushing zero-trust architectures — "never trust, always verify" — for critical infrastructure. Government agencies and financial institutions are evaluating zero-trust frameworks that continuously verify user identity, device posture, location, and behavior patterns, Convergence Now reports. Real-time payment systems like FedNow, PIX, and UPI are enabling account-to-account payments that bypass card networks, adding compliance considerations for cross-border wallets. Regulators are also signaling clearer liability frameworks for platforms handling digital payments and user data.

When a user taps to pay at a coffee shop in 2027, the decision of which rail carries that $4 — credit, debit, stablecoin, BNPL — may not be theirs. It will belong to the agent that learned their patterns, parsed the merchant, and routed the value before the NFC field collapsed. The wallet that wins that millisecond becomes the control plane. Uno's bet is that the agent starts with rewards. The market will decide where it ends.

Working in AI? Zero G Talent tracks the openings: see every open Databricks role, browse AI jobs, openings at Anthropic, and the people building the field.