The AI Species – Weekly Convergence Report

KW22/2026: When Banks Mint Coins, Agents Pay With Them, and Cars Drive Themselves

Dear readers,

the calendar week 22 of 2026 will be remembered as one of those quiet inflection points whose significance only becomes fully apparent in retrospect. While headlines on the surface seemed dispersed across disparate domains — a bank-issued stablecoin here, a self-driving expansion there, a regulatory rebuff in Frankfurt, a tokenization deal in Sofia — the underlying signal was remarkably coherent. We are witnessing, in real time, the operational stitching together of the machine economy: a system in which autonomous software agents transact via programmable money, regulated tokenized assets flow through decentralized protocols, and physical robots gradually assume tasks previously performed by human labor. Jefferies’ projection that crypto IPOs could constitute a one-trillion-dollar market within two years is not a speculative headline — it is a quantitative shadow cast forward by the convergence trends materializing this very week.

The thesis I would like to develop in this edition is the following: the machine economy is no longer an abstraction debated in white papers and conference panels. It is being built brick by brick, rail by rail, contract by contract — and KW22 delivered an unusually dense cluster of structural milestones. SoFi became the first U.S. national bank to issue a retail stablecoin to 15 million users. Keyrock documented that stablecoin rails are becoming the default payment infrastructure for AI agents. VanEck’s tokenized fund touched down on a permissionless DeFi protocol. Waymo expanded into the Washington, DC metropolitan area. Mercedes committed to a year-end urban autonomous rollout in Germany. And, sobering counterweight to all this enthusiasm, Cisco published evidence that not a single closed frontier model survives a determined multi-turn attack. The picture that emerges is one of accelerated convergence — but also of widening risk surface area. For investors, operators, and citizens, the strategic question is no longer whether the machine economy is arriving; it is which infrastructure layer one chooses to own before the network effects close.

I. AI & Agents: The Security Reckoning Meets the Regulatory Wave

The week’s most consequential AI story was not a model release but a security report. Cisco’s research concluded that none of the tested closed frontier large language models can be considered safe once a multi-turn attack is initiated. For anyone building autonomous agents — which is to say, for anyone building the operational backbone of the machine economy — this is a structural problem, not a cosmetic one. Single-turn benchmarks have long been a poor proxy for real-world robustness, but Cisco’s findings put numbers on what red-teamers have whispered for months: the more you let an LLM converse, plan, and act, the more its alignment guarantees decay. In a world where agents will autonomously dispatch payments, sign contracts, and execute trades, an exploit window of even minutes is sufficient to drain a treasury or corrupt a supply chain.

It is therefore no coincidence that AI security startup Gray Swan raised $40 million this week to scale its army of 15,000 hackers pressure-testing Claude, GPT-5, and Gemini. The economic logic is clean: as the value handled by AI agents scales from billions toward trillions, the marginal dollar spent on adversarial testing yields outsized risk-adjusted returns. Gray Swan’s positioning — embedded with every major frontier lab — makes it a critical piece of infrastructure, comparable to what credit bureaus or auditors became for the financial system of the twentieth century. Investors should note: the AI safety and red-teaming layer is shaping up to be a categorical winner-take-most market, and the valuations of pure-play security firms will likely diverge sharply from those of generalist AI startups in the coming quarters.

Regulators are responding in parallel. Illinois passed what is being described as America’s strongest AI safety bill, mandating third-party verification of safety standards for companies including OpenAI, Anthropic, and Google. This is a meaningful pivot in U.S. AI governance: from voluntary commitments toward externally auditable, jurisdiction-specific obligations. For frontier labs, the compliance overhead is non-trivial, but the second-order effect is more interesting — it accelerates the institutionalization of AI assurance as a tradable service, exactly the kind of market Gray Swan and its peers are positioned to dominate. As I argue throughout The AI Species, every technological transition produces both the productive machines and the watchdog machines; the latter is often the better long-term investment because it scales with regulatory complexity rather than with raw model capability.

II. Robotics: From Boulevard to Boardroom — The Physical Layer Hardens

If the digital agent economy depends on stablecoins and red-teamed LLMs, the physical machine economy depends on rolling steel that can actually navigate the world. This week brought three data points worth integrating. Waymo announced its expansion into Virginia, mapping Alexandria and Arlington — cities sitting directly across the Potomac from federal Washington. The geographic choice is not accidental. By placing autonomous vehicles within sight of the Capitol, Waymo is conducting an exercise in regulatory psychology as much as engineering: making robotaxis visible, familiar, and unremarkable to the lawmakers and staffers who will ultimately write the federal framework.

The human dimension of this expansion was captured movingly by the New York Times, which reported on visually impaired Waymo users experiencing the rare joy of riding alone. This is the often-overlooked welfare argument for autonomy: for millions of people, the marginal autonomous mile is not a luxury but an unlock of independence previously gated by another human’s availability. Such social externalities are, in my analysis, the most underpriced variable in long-term robotaxi adoption curves. Public sentiment, once it tilts decisively toward the lived experience of beneficiaries, tends to outpace regulatory caution.

Mercedes-Benz, meanwhile, committed to a year-end German rollout of urban point-to-point assisted driving. This places European premium OEMs in direct competition with Chinese and American autonomy stacks on home turf — a development that should accelerate, not slow, European industrial policy support for indigenous AI compute and sensor supply chains.

And then, the counter-narrative. Reuters’ investigation into Tesla’s self-driving program revealed that the company’s own AI trainers do not trust its full self-driving technology or the safety statistics built around it, striking at the heart of a $1.6 trillion valuation thesis. For investors, the lesson is uncomfortable but essential: autonomy is not a single market, and not all autonomy stacks are equivalent. The chasm between geofenced, sensor-redundant, mapped systems (Waymo) and vision-only, generalized claims (Tesla FSD) is widening, not narrowing. Capital allocation in the robotics layer of the machine economy must therefore distinguish ruthlessly between demonstrated operational safety and rhetorical promise.

III. Crypto, DeFi & RWAs: The Plumbing Becomes Visible

Three converging stories defined the financial infrastructure narrative this week. First, SoFi’s launch of SoFiUSD — a dollar-backed stablecoin issued on Ethereum and Solana, available to 15 million retail customers of a U.S. national bank. The symbolic weight of this is hard to overstate. For years, the stablecoin market was dominated by non-bank issuers operating in regulatory grey zones. With a federally chartered bank now minting digital dollars directly onto public blockchains, the firewall between traditional banking and crypto infrastructure has been formally breached — not by a fintech upstart, but from inside the regulated perimeter.

Second, VanEck’s tokenized fund arrived on the Euler DeFi protocol, an explicit signal that decentralized finance is actively courting Wall Street institutional capital. Tokenized real-world assets are no longer a thought experiment; they are collateral, yield instruments, and trading inventory in live protocols. Parallel to this, REAL Finance announced its first securities tokenization agreement with the licensed investment firm Factori AD, expanding the European tokenization corridor through regulated channels.

Third, the policy backdrop. The proposed Digital Asset Market Clarity Act of 2025 is, according to industry analysis, poised to trigger a boom in compliant “yield-as-a-service” offerings, forcing a migration from passive, opaque yield mechanisms toward active, transparent, jurisdictionally clean capital strategies. This is the regulatory unlock the U.S. crypto industry has lobbied for since 2018. If passed in its current form, it will rewire the economics of crypto-native asset management.

But not every regulator is moving in the same direction. The European Central Bank rebuffed proposals to boost euro-denominated stablecoins, calling them too risky for financial stability and warning of disintermediation pressure on bank lending. This stance, if maintained, risks ceding the euro’s role as a unit of account in the emerging on-chain economy to dollar-denominated alternatives — exactly the outcome European policymakers have publicly feared. The transatlantic divergence on stablecoin policy is becoming one of the most consequential macro-financial vectors of 2026, and investors should monitor it with the same intensity they once reserved for interest rate differentials.

Tying these threads together is the Jefferies forecast cited above: a one-trillion-dollar wave of crypto IPOs over the next two years. When traditional investment banks issue projections of this magnitude, the relevant question is not whether the number is precise — it is what behavioral signals it sends to allocators, pension funds, and sovereign wealth managers who anchor their positioning to such forecasts.

IV. Convergence Infrastructure: Where AI and Capital Markets Fuse

The single cleanest example of full-stack convergence this week was the Keyrock report documenting that stablecoins on blockchain rails are becoming the default payment layer for AI agents, with Coinbase, Stripe, and Visa actively positioning themselves to capture flow. This is the thesis of The AI Species in compressed form: software agents need to transact, traditional card networks are too slow, too geographically fragmented, and too expensive for sub-cent micropayments — therefore programmable money on programmable rails wins. The fact that Visa, the world’s most successful payment network, is publicly acknowledging the need to participate in this alternative architecture rather than compete against it tells us the question has shifted from if to which protocol.

Adjacent to this, Luffa AI secured strategic investment from GoFintech Quantum at a $220 million valuation, positioning itself at the intersection of Web3, AI, and fintech. The valuation itself is less interesting than the category. Investors are now actively pricing companies whose entire raison d’être is to operate at the convergence layer — not “AI” or “crypto” but the membrane between them. As I have argued repeatedly, the membrane is where the highest economic rents will accrue over the next decade, because that is where translation, settlement, and trust services must be performed at machine speed and machine scale.

V. Strategic Synthesis: Assembling the Full Stack

When we lay the week’s events side by side, a coherent architecture emerges. At the bottom we have the physical layer — Waymo, Mercedes, and the autonomous vehicles that move atoms through space. Above that sits the AI cognitive layer — frontier models from OpenAI, Anthropic, and Google, now subject to mandatory third-party verification in Illinois and continuous adversarial testing by Gray Swan. Above that, the agentic execution layer — software entities that plan, decide, and act, increasingly transacting via stablecoin rails as Keyrock documented. Then the financial settlement layer — SoFiUSD bringing bank-grade stablecoins to retail, VanEck and REAL Finance bringing regulated assets on-chain, the Clarity Act poised to legitimize compliant yield products. And surrounding all of it, the regulatory and assurance layer — Illinois on AI safety, the ECB on stablecoin risk, Cisco and Gray Swan on model robustness.

The strategic insight from The AI Species applies with unusual force this week: value in the machine economy does not accrue uniformly across layers. It concentrates at the chokepoints — the places where one layer must interface with another and where trust, settlement, or verification must occur. SoFi at the bank–crypto interface. Gray Swan at the model–deployment interface. Euler at the TradFi–DeFi interface. Waymo at the algorithm–street interface. Each of these companies is in effect minting a license fee on every transaction that crosses its boundary. For long-term investors, the portfolio question is not “do I own AI?” or “do I own crypto?” but “do I own the interfaces?”

The contrarian risk vector remains the security surface. If Cisco’s findings on multi-turn vulnerabilities generalize — and there is no reason to think they will not — then the agentic economy is being built on cognitive substrates that can be socially engineered. The race between capability deployment and adversarial hardening is the single most important technical race of 2026, and its outcome will determine whether the machine economy compounds smoothly or suffers a catastrophic trust event in the next 24 months.

VI. Outlook: What to Watch in KW23

Several threads from this week will demand continued attention. First, the legislative trajectory of the Clarity Act in the U.S. Congress — every committee vote, every amendment, will reprice large swaths of the crypto sector. Second, the ECB’s posture: will the rebuff of euro stablecoins harden into formal restriction, or soften under pressure from European banks watching dollar stablecoins absorb their deposits? Third, follow-on tokenization announcements from major asset managers — VanEck’s Euler launch will not remain isolated, and the next BlackRock or Fidelity move could compress timelines dramatically. Fourth, autonomy: Mercedes’ year-end commitment now sets a public deadline against which execution will be measured, and Waymo’s Virginia mapping will yield its first DC-area passenger rides faster than most analysts expect. Fifth, AI security: expect at least one additional state to introduce Illinois-style legislation, and expect adversarial research firms to publish further evidence of agent-level vulnerabilities.

The machine economy is not a forecast. It is, as of this week, a deployed system with named operators, regulated balance sheets, and human passengers. Our task as analysts, investors, and citizens is to understand its architecture clearly enough to participate in its construction rather than merely be processed by it. That is the work of The AI Species, and it is the work of this newsletter week after week.

Yours sincerely, Thomas Huhn