Leonard Botezatu
A digital wallet stores credentials. An AI agent can act on them. That shift changes what identity means online: from proving who you are to delegating what can be done on your behalf.
...but a digital entity representing you like an Ai agent. Imagine:
"Transaction declined," flashed the coffee shop terminal. Marcus smiled, knowing exactly what happened. His AI agent, Nova, had detected he was about to pay $10 for a basic coffee - way above his usual spend pattern. "Nova, it's fine. The prices went up," Marcus whispered to his wearable earpiece. Within seconds, the payment cleared, but not before Nova had compared prices across nearby cafes and made a note to suggest a more economical spot tomorrow.
This was just one way Nova had transformed from a simple digital wallet into becoming Marcus's digital twin. While his old wallet just stored payment cards and ID documents, Nova actively represented his interests in the digital world. Later that day, during a job interview, Nova seamlessly presented Marcus's credentials, academic records, and work history - all while ensuring his private data remained protected. The interviewer's AI agent and Nova negotiated the perfect information exchange: enough to prove Marcus's qualifications without oversharing.
"Remember when we had to manually show our IDs everywhere?" Marcus thought, as Nova automatically handled his gym entry, workspace access, and transit payments throughout the day. "Now my digital self just... exists."
That evening, as Marcus relaxed at home, Nova was busy declining suspicious authentication requests, updating his privacy settings across platforms, and even negotiating better terms for his subscription services - all based on understanding his preferences and patterns. This wasn't just technology anymore. Nova had become Marcus's trusted digital representation, making thousands of micro-decisions daily, exactly as Marcus would himself.
Why the wallet was the right answer to the wrong question
Wallets solved a storage problem. Cards, licences and passes used to live in a physical wallet; the digital wallet moved them into a phone. That is useful. It is also passive. A wallet only produces a credential when a human decides to open it and show it.
An agent is different because it acts without a human in the loop for every step. That is the whole point. It can compare prices, decline a suspicious request, renew a subscription, or prove a single fact to a merchant. But every one of those actions needs three things a wallet never had to provide: a scope, a record, and a way to undo it.
Delegation is an identity problem, not a UI problem
When Marcus asks Nova to pay for coffee, he is not really asking for a payment. He is granting a narrow permission: this merchant, this amount, this purpose, for the next ten minutes. The interface can make that permission easy to understand. It cannot make it safe on its own.
Safe delegation needs the same primitives as any access-control system:
- Scope. The agent can prove age without revealing a date of birth. It can pay without exposing an account number.
- Intent. The request is bound to what the human actually asked for, so a merchant cannot reuse it for something else.
- Provenance. The relying party can tell the difference between a human and an agent acting for one.
- Audit. Every action is logged in a form the human can read later.
- Revocation. A permission can be withdrawn immediately, and the withdrawal is respected everywhere.
The trust problem nobody wants to own
The hard question is not technical. It is liability. If an agent buys the wrong thing, who pays? If it proves the wrong attribute, who is responsible? If it is tricked by a malicious merchant, is that fraud, a product defect, or user error?
Today the answer is usually whoever has the best lawyer. That is not a foundation for a payment network. Standards such as verifiable credentials and mobile driving licences define how a claim travels. They do not define who carries the risk when an autonomous system gets it wrong. Until that is settled, agents will stay in low-stakes tasks: reminders, search, drafts. The moment money or identity is involved, a human will still be asked to confirm.
What has to be true before this works
Three conditions need to hold at the same time. First, credentials have to be verifiable without a phone-home to the issuer, or the system will not scale. Second, agents need a recognised identity of their own, so a merchant can apply a policy to a bot instead of guessing. Third, users need a consent experience that is honest about what is being delegated. A wall of checkboxes is not consent. A single, specific sentence is.
The wallet is not going away. It becomes the place where those permissions live. The interesting product work is not the card carousel. It is the moment a person decides how much of their digital life an agent is allowed to handle, and the moment they take it back.