July 30, 2026
Blockchain
AI agents are evolving from information tools into participants that can make payments, select products and execute transactions. This article explores how tokenized infrastructure could connect money, financial products, rules and ownership—making blockchain a largely invisible execution layer for the financial system of the future.
The most important users of blockchain may eventually be machines.
AI agents are moving beyond answering questions. They are beginning to research services, compare alternatives and complete tasks on behalf of users. The next step is financial: making payments, purchasing products, allocating capital and managing assets. But recommending a financial action and executing it are two different things.
An AI can already recommend an investment. To complete it, the agent must confirm that the client is eligible, establish that it is authorized to act, transfer the money and ensure that ownership of the asset is recorded correctly. Today, these steps often take place across several institutions and systems. APIs can help an AI communicate with them, but communication alone does not turn the process into one reliable transaction. This is where blockchain and tokenization may become important - not because AI cannot use online banking, but because tokenized infrastructure can connect money, assets, product rules and ownership in one programmable process.
Our thesis for 2035 is simple:
AI will not merely use blockchain. AI may become one of the main reasons blockchain becomes part of the financial system.
Through APIs, an AI agent can check a balance, initiate a payment or retrieve transaction information. This may be sufficient when the task begins and ends within one institution. A fund investment, however, may involve a bank, distributor, administrator and transfer agent, each maintaining its own records. An AI agent could communicate with all of them, but it would still depend on every institution completing and confirming its part of the transaction.
Blockchain offers a different approach. Where institutions use compatible tokenized infrastructure, payment, product rules and ownership records can be connected. Instead of sending separate instructions and waiting for several systems to be updated, selected conditions can be checked and the exchange of money and assets can be coordinated as one process.
This does not mean blockchain is better for every process. If one bank controls the entire workflow, its existing systems may be simpler and more efficient. Blockchain becomes more relevant when money or ownership must move between institutions that otherwise rely on separate records and repeated handovers.
APIs allow an AI agent to send instructions to individual institutions. Tokenized infrastructure could reduce the handovers between them by connecting payment, product rules and ownership records.
Tokenization is often described as representing an asset on a blockchain. That definition is correct, but incomplete. A PDF is also a digital representation of an asset. So is an entry in a transfer agent’s database. Moving an existing record onto a blockchain does not automatically make the product easier to access or operate. The greater opportunity is to make parts of the product programmable.
A token can carry or reference rules defining who may hold it, how it may be transferred, which events trigger a payment and which parties may interact with it. It can connect ownership, eligibility, settlement and transfer restrictions in a form that software can process. For a human investor, the relevant information may be spread across a prospectus, subscription agreement, term sheet and website. A professional can read those documents and determine the product’s fees, liquidity, investor requirements and transaction restrictions. An AI agent needs the same information in a reliable and structured form. It should not have to determine every important condition by interpreting several separate documents and comparing records held by different institutions.
Tokenization creates the possibility of connecting an asset to authoritative product information and applying selected rules directly. A tokenized fund could, for example, restrict transfers to eligible investors and exchange ownership against tokenized money when the agreed conditions are met. The difference is that important product conditions would no longer exist only as text for people to read. Some could also become rules that software can check before acting.
Digitization allows software to read information about a financial product. Tokenization can also allow software to apply selected rules and complete permitted transactions.
This vision is no longer entirely theoretical.
In May 2026, AWS introduced Amazon Bedrock AgentCore Payments in preview, developed with Coinbase and Stripe. It allows AI agents to discover and pay for APIs, data, content and other digital services while completing a task. This initiative shows that AWS expects the role of AI agents to evolve. Services may increasingly be used not only by people, but also by software that can identify, select and purchase the resources required to achieve an objective. The user still remains in control. They authorize the agent and set a spending limit that cannot be exceeded. The agent can then choose and pay for the service it needs, while every payment remains visible through AWS’s monitoring systems.
In this example, the payment uses stablecoins through Coinbase’s x402 protocol, which enables automatic payments within the same internet interaction used to request an online service. Once the payment has been confirmed, the service can be delivered and the agent can continue its task. The important point is not the payment instrument itself. It is the division of roles. A person defines the objective and the spending limit. The AI selects the service. The infrastructure checks that the payment is permitted, completes it and records what happened. Today, this applies to APIs and digital content rather than funds or securities. The AWS service is also still in preview. But it already demonstrates that AI can make a purchasing decision and complete the transaction without receiving unrestricted access to the user’s money.
AWS is preparing for a world in which software does more than recommend a purchase. It completes one within boundaries defined by a human.
The AWS example also shows how blockchain could reach broader adoption. The user does not need to choose a network, understand the payment protocol or manually transfer stablecoins. They simply see an AI agent completing a task within an approved budget, while the infrastructure remains in the background. The same operating model could eventually extend to investment products. Instead of selecting a fixed ETF, an investor might choose a broader investment strategy and add personal preferences. One investor might exclude oil companies. Another might remove their employer’s shares to avoid concentrating both their income and investments in the same company. An AI agent could translate those preferences into a personalized portfolio while preserving the overall investment strategy.
Blockchain would not create the investment strategy itself, but tokenized assets could make it easier to assemble, adjust and settle individual holdings at scale. The investor would see the chosen strategy and personal preferences, while the underlying technology remains invisible. A direction already being explored within the Asset Management Association Switzerland. Blockchain could therefore become widely used without investors consciously choosing it - just as people use the internet without understanding its protocols or make card payments without seeing the institutions involved. Its adoption will be driven by services that use tokenized infrastructure in the background to interact more directly with money, assets and product rules.
Blockchain adoption will be complete when people stop describing products as blockchain products.
By 2035, we believe AI agents will perform an increasing share of the financial steps that people and institutions execute manually today. This does not mean that every asset will be tokenized, every bank account will disappear or every transaction will move onto a public blockchain. Existing systems will remain where they work well, and financial institutions will continue to identify clients, protect assets, define controls and resolve exceptions.
However, the financial system will increasingly need to support AI agents as active participants, not merely as tools that provide information. To act on behalf of people and institutions, they will need clearly defined authority, controlled access to money and financial products, and infrastructure that can verify the relevant conditions before executing a transaction across different systems. APIs will remain part of this architecture. But where transactions involve several institutions, tokenized infrastructure could offer something more: a way to connect payment, product conditions and ownership records instead of coordinating them through separate messages and databases.
The transition is unlikely to happen through one dramatic change. It will appear through tokenized payments, machine-readable investment products, automated settlement and AI agents operating within clearly defined mandates. Together, these developments describe a different financial architecture. People define the objectives, institutions establish the rules and protections, AI agents select the actions, and programmable infrastructure coordinates the execution. Humans will remain responsible for determining what the system should achieve. Machines will increasingly perform the steps between intention and outcome. The future of tokenization should therefore not be judged only by whether humans want to use blockchain. The more important question is whether AI agents will need infrastructure in which money, assets and transaction rules can be accessed and acted upon by software.
We believe they will.
AI may become the user that reveals what tokenized financial infrastructure is ultimately capable of.
